A closer look at 22 sectors

Research date: September 15, 2026. All funding figures, market sizes and investor-sentiment readings in this chapter were compiled on or before this date. Venture data is inherently revised upward for several quarters after a period closes (deals get reported late), so H1 2026 numbers cited here will likely be restated slightly higher in later reports.

How to Read This Chapter

What this chapter is. A working map of where startup capital, customer demand and competitive pressure actually sit in late 2026 — organized so a founder, operator or investor can find their sector, understand its economics, and calibrate expectations.

Three epistemic labels used throughout:

• [Verified] — A number reported by a primary source: a government statistical agency, a company filing, or a venture-data provider reporting on transactions it tracked (Crunchbase, PitchBook-NVCA, CB Insights, Rock Health, AgFunder). These are measurements of things that happened.
• [Estimate] — A figure derived by a third party using a methodology that is not fully transparent, or a number that different credible sources disagree about. Private-company revenue run rates, private valuations, and "market size" figures usually fall here.
• [Projection] — A forecast. Treat with the skepticism forecasts deserve.
• [Analysis] — My own interpretation, clearly separated from sourced fact.

On market-size numbers specifically. The "total addressable market" figures published by research vendors (Grand View, MarketsandMarkets, Precedence, Verified Market Reports, and dozens of near-identical outfits) are a genre of content marketing, not measurement. Their business model is selling $4,000 PDF reports, and a bigger number sells better. They are frequently produced by scaling a base year by an assumed CAGR with no bottom-up validation, and competing vendors routinely publish estimates that differ by 3–5x for the same market in the same year. The gaming sector below is a clean illustration: BCG says $263B, Grand View says $322.6B, and Statista says $577.9B for roughly the same thing (Shattered.io summary of BCG, 2026). Where a government source, an audited filing, or an industry body exists, I use it and say so. Where only vendor estimates exist, I say that too, and I do not treat the number as a fact.

On survivorship bias. Everything in the "notable companies" lines below is a survivor. The companies that raised at the same time on the same thesis and died do not have press coverage, and funding databases are systematically better at recording rounds than recording shutdowns. A sector that looks like it produces winners may simply be a sector whose losers are invisible. Wherever I could find shutdown or down-round data, I include it; usually I could not, and you should mentally discount the "notable companies" lists accordingly.

On the venture-data providers themselves. Crunchbase, PitchBook and CB Insights all sell data subscriptions and all benefit from a narrative of a large, dynamic, trackable venture market. They also use different sector taxonomies and different definitions of "AI company," which is why you will see AI's share of venture funding quoted as 70%, 80% and 86% in this chapter — all correct, all measuring different things. I flag these divergences rather than picking a favorite.

The Macro Picture: What 2026 Actually Looks Like

Any sector read in 2026 is a read on one sector's position relative to an unprecedented gravitational body. The numbers:

Global venture funding hit $510 billion in H1 2026 — a record half-year, against $440 billion for all of 2025 and a previous half-year record of $375 billion in H2 2021 (Crunchbase News, July 2026). [Verified, per Crunchbase's tracked transactions.]

Two companies took 43% of it. OpenAI and Anthropic raised a combined $217 billion in H1 2026. Anthropic alone raised $65 billion in Q2 (Crunchbase News, July 2026). PitchBook records that round as a step-up from a $350 billion pre-money to roughly $900 billion (Q2 2026 PitchBook-NVCA Venture Monitor); TechCrunch reported the post-money at $965 billion (TechCrunch, August 2026). [Estimate — private valuations are negotiated, not observed, and sources disagree by ~7%.]

AI's share, by three different measurements:

• Crunchbase: AI-focused companies took over 70% of global capital in Q2 2026, up from ~50% in Q2 2025; Q1 2026 was roughly 80% (Crunchbase News, July 2026).
• PitchBook-NVCA: AI companies captured 86% of all US H1 2026 venture dollars — $355.9 billion across 3,258 deals (Q2 2026 Venture Monitor).
• CB Insights: total funding exceeded $200B for a second consecutive quarter while deal count fell to its lowest point in over a decade, with 263 mega-rounds taking 81% of all capital and a single company representing nearly 40% (CB Insights, State of Venture Q2'26).

[Analysis] These three are not in conflict. The US skews more AI-heavy than the world; "AI company" is a broader label at PitchBook than at Crunchbase; and all three agree on direction and magnitude. The honest summary is: somewhere between seven and nine of every ten venture dollars in 2026 went to something an investor classified as AI.

The structural consequence, which matters more than the headline. Megadeals of $100M+ absorbed 87.5% of US capital deployed in H1 2026, up from 56.2% in 2024 (Venture Monitor). Sub-$100M rounds — which is to say, almost every round any normal startup will ever raise — compressed to 12.5% of dollars. Meanwhile first-time fund formation is on pace for its lowest year since 2016, and three firms (Andreessen Horowitz, Thrive Capital, Founders Fund) captured 48.1% of all US venture fundraising (Venture Monitor; NonPublic analysis, 2026).

[Analysis] This is the single most important fact in this chapter for a founder outside the AI-infrastructure core. The record headline number does not describe your fundraising environment. Capital available for a $3M seed or a $15M Series A in a non-consensus sector is not at a record; the supply of first-time managers who historically wrote those checks is shrinking. Read every sector profile below against that backdrop: "sector X raised $10B in 2026" is compatible with "it is harder than it was in 2021 to raise $5M in sector X."

One genuine bright spot: exits reopened. H1 2026 US exit value hit $2.19 trillion across 874 exits — but SpaceX's IPO alone accounted for the bulk of it, and PitchBook notes that without SpaceX, quarterly exit values reflect "a level more consistent with the constrained environment of recent years" (Venture Monitor). Globally, Q2 2026 saw 32 IPOs above $1B and 24 M&A deals at $1B+ totaling $113 billion — a record M&A quarter (Crunchbase News, July 2026). [Analysis] M&A at $1B+ is the more broadly relevant signal than the SpaceX-distorted IPO number: strategic acquirers are buying again, which raises the realistic exit floor across many sectors covered below.

Unicorn creation collapsed even as funding rose. New unicorn births fell more than two-thirds in a single quarter to a six-quarter low (CB Insights Q2'26), while the existing US unicorn herd grew to 945 companies worth a collective $5.3 trillion (Venture Monitor). [Analysis] Fewer new billion-dollar companies being minted while the old ones appreciate is what a late-cycle, concentration-driven market looks like — not what a broad boom looks like.

Part I: Sector Profiles

1. Artificial Intelligence (Foundation Models & AI Infrastructure)

Market size. There is no credible single "AI market size" number, and I decline to invent one. What can be measured: worldwide IT spending is forecast at $6.37 trillion in 2026, up 14.2%, with data center systems — the physical substrate of AI — growing 62.5% to $822 billion and IaaS growing 29.3% to $287 billion (Gartner, July 2026). [Projection, from a vendor that sells research but whose IT-spend series is the closest thing to an industry standard.] Gartner's analyst frames it plainly: "Building the compute capacity required for AI is the largest infrastructure project ever attempted by humanity."

Demand and growth. Real and extraordinary at the model layer. Anthropic's annualized revenue run rate went from $9B (end of 2025) to $47B (May 2026) to $65B (end of July 2026) — $18B of annualized revenue added in two months (CNBC, August 2026; TechCrunch, August 2026). OpenAI reportedly doubled from $20B to **$40B** over the same period. [Estimate — these are company-disclosed run rates shared with investors, not audited revenue, and "annualized run rate" flatters fast-growing businesses by extrapolating the best month.]

Capital intensity. Extreme, and the defining feature. The five largest US hyperscalers plan $660–690 billion in 2026 capex, roughly double 2025's ~$380 billion: Amazon ~$200B, Alphabet $175–185B, Microsoft $120B+, Meta $115–135B, Oracle ~$50B (Futurum Group, 2026). [Projection, built from company guidance.] Foundation-model training is now a capital expenditure category comparable to national infrastructure programs.

Regulatory. Lighter than expected in 2026. The EU agreed a "Digital Omnibus" deferring high-risk AI obligations under the AI Act from August 2026 to December 2, 2027 for stand-alone Annex III systems and to August 2028 for AI in regulated products. Article 50 transparency obligations did take effect August 2, 2026, with a grace period to December 2026 for watermarking existing systems, and new prohibitions on non-consensual intimate imagery and CSAM become enforceable December 2, 2026 (Gibson Dunn, 2026). [Verified.] Copyright remains the live legal risk: US courts have allowed Disney and other studios' infringement claims against AI video generators to proceed (Norton Rose Fulbright AI litigation update, 2026).

Competition. At the frontier: effectively a four-to-six player oligopoly with a capital moat that no new entrant can cross without a hyperscaler sponsor. At the application layer: brutally crowded, minimal defensibility, and exposed to the underlying model providers shipping your feature.

Business models. API/token metering; enterprise seats and committed-capacity contracts; consumer subscription; increasingly, revenue-share and outcome pricing at the application layer.

Investor interest. Maximal. Foundational AI startups raised $178 billion across just 24 deals in Q1 2026 alone — double all of 2025 ($88.9B across 66 deals) and 5.7x 2024 (Crunchbase News, 2026). Median Series D+ AI valuations run $4.25 billion versus $134 million for non-AI companies at the same stage (Venture Monitor) — a ~32x premium that is the cleanest single measure of the dislocation.

Risks. (1) The capex-to-revenue gap: at even the most generous accounting, model-layer revenue is a low-single-digit percentage of hyperscaler capex, and infrastructure precedes revenue by 18–36 months. (2) Depreciation schedules on GPUs that may not have the assumed useful life. (3) Circular financing — see the synthesis section. (4) Commoditization: capability gaps between frontier models have been narrowing, which is good for buyers and bad for pricing power. (5) A single disappointing capex guide from one hyperscaler could reprice the entire complex.

Notable companies. OpenAI, Anthropic, xAI (absorbed by SpaceX in a $250B transaction — the largest VC-backed M&A ever recorded), Mistral, Reflection AI ($2B Series B led by Nvidia), Safe Superintelligence, Advanced Machine Intelligence ($1.03B — Europe's largest seed round on record).

Underserved opportunities. [Analysis] Evaluation and observability for agent systems in regulated industries; inference cost optimization as a service (the arbitrage between what models cost and what applications charge is currently enormous and will compress); data rights, licensing and provenance infrastructure, which the copyright litigation is about to make mandatory rather than optional; and non-English, non-US-regulatory-context deployment tooling. Note that all of these are picks-and-shovels businesses serving the boom rather than bets on the boom — which is the appropriate risk posture for a startup that cannot raise $10B.

2. SaaS & Enterprise Software

Market size. Software spending is forecast at $1.468 trillion in 2026, up 15.5%; IT services at $1.570 trillion, up 5.3% (Gartner, July 2026). [Projection.] Notably, Gartner has revised software spend upward through 2026 — the widely predicted AI-driven SaaS collapse did not show up in aggregate spending.

Demand and growth. Solid in aggregate, bifurcated underneath. Budget is flowing to AI-adjacent software and to consolidated platform vendors; it is flowing away from single-purpose point solutions whose function an AI agent or an incumbent's bundled feature can replicate.

Capital intensity. Historically the lowest of any sector here — and rising. AI-native SaaS carries gross-margin structures materially worse than classic SaaS because inference is a variable cost of goods sold. [Analysis] A founder modeling 80% gross margins on an AI product in 2026 is probably modeling wrong; 55–70% is the realistic band until inference prices fall further, and that structurally changes how much a company can burn on go-to-market.

Regulatory. Light, except where vertical (healthcare, finance, education). SOC 2 / ISO 27001 are table stakes; EU AI Act Article 50 transparency now applies to any product where a user interacts with an AI system.

Competition. Extremely high and getting worse. The cost of building a credible v1 has collapsed, which means the number of competitors per category has risen while differentiation has fallen.

Business models. The live structural question of the sector is the migration away from per-seat pricing. If an agent does the work a seat used to do, per-seat pricing shrinks the vendor's revenue exactly when the vendor is delivering more value. Consumption, hybrid platform-fee-plus-usage, and outcome-based pricing are all being tried (Deloitte TMT Predictions 2026). [Analysis] Most of the popular commentary here overstates the speed of the change — procurement, budgeting and forecasting processes at large enterprises are built around predictable seat licenses, and buyers resist unpredictable bills. Expect hybrid models to win, not pure outcome pricing.

Revenue potential. Realistically: horizontal SaaS in a crowded category is now a $5–50M ARR business that gets acquired. Vertical SaaS owning a workflow plus payments in an underserved industry remains one of the most reliable paths to $100M+ ARR in all of startup-land, and has not been disrupted.

Investor interest. Present but unglamorous. Enterprise software appears regularly in Crunchbase's weekly top-10 rounds (Crunchbase News, June 2026), but the money goes to AI-labeled enterprise software. A non-AI SaaS company raising a Series B in 2026 faces a genuinely hostile market.

Risks. Feature absorption by incumbents and by model providers; gross-margin compression; pricing-model transition destroying net revenue retention mid-flight; and — underrated — the fact that AI lowering build costs means your customers can increasingly build a 70%-good internal version of your product.

Notable companies. Salesforce, Microsoft, ServiceNow (incumbents defending); Ramp, Mercury, Taktile, Hightouch, Netomi, Mews (breakouts, mostly vertical or infrastructure).

Underserved opportunities. [Analysis] Vertical software for industries with poor digitization and real regulatory moats — specialty healthcare practices, construction trades, marine/logistics, waste management, regulated industrial services. The pattern that works: own the workflow, then own the money movement. Also: the "AI cleanup" category — data quality, permissions, and audit tooling that enterprises need before agents can be trusted with production systems.

3. Developer Tools

Market size. No reliable standalone figure; devtools spending is embedded in the $1.468T software line. [Analysis] The meaningful measure is that AI coding assistants went from zero to among the fastest-revenue-scaling software products ever observed in roughly three years.

Demand and growth. Exceptional and real — this is the single most validated AI application category. Anysphere (Cursor) has been reported raising at valuations escalating from $30B to $50B and reportedly $60B within 2026 (Fortune, March 2026; The Next Web, 2026). Crunchbase records Anysphere as the subject of a ~$60 billion acquisition, the largest M&A of Q2 2026 (Crunchbase News, July 2026). [Estimate — reporting on this transaction is inconsistent across sources and I would not treat the $60B figure as settled.]

Capital intensity. Low to build, very high to compete. The category's defining economic problem: AI coding tools resell inference from model providers who are also their competitors, which caps gross margin and creates existential platform risk.

Regulatory. Minimal. Rising enterprise concern about code provenance and license contamination from generated code.

Competition. Among the most intense in software — model labs (Claude Code, OpenAI Codex), independent startups (Cursor, Cognition), and platform incumbents (GitHub) competing for the same developer.

Business models. Per-seat subscription with usage tiers; enterprise agreements; increasingly consumption-based as agentic coding shifts cost to tokens.

Revenue potential. Very high for the top three; poor for everyone else. [Analysis] This is a winner-take-most category with brutal gross-margin economics for non-model-owners. Fortune's framing of Cursor's position as a "crossroads" — enormous growth, very uncertain durability — is the right one.

Investor interest. High, concentrated at the top. Cognition appeared among Crunchbase's largest weekly rounds in 2026 (Crunchbase News, 2026).

Risks. Margin squeeze from model providers who can price their first-party products below a reseller's cost; rapid preference-switching by developers, who have essentially zero switching costs; and the possibility that coding agents become a bundled feature of the model subscription rather than a separate product.

Notable companies. Anysphere/Cursor, Cognition, GitHub (Microsoft), Replit, Baseten, Vercel.

Underserved opportunities. [Analysis] Everything downstream of code generation, where the bottleneck has moved: code review at agent-generated volume, test generation and verification, dependency and supply-chain security for AI-authored code, and infrastructure/observability for teams whose diff volume went up 5x. The scarce resource is no longer writing code; it is trusting code.

4. Fintech

Market size. Not meaningfully separable from financial services revenue globally; the relevant anchors are payment volumes and banking revenue pools, not a "fintech TAM." [Analysis] Treat any "$X trillion fintech market" claim as marketing.

Demand and growth. Recovered but selective. Global fintech funding reached $28.6 billion in H1 2026, up 22.7% year over year, while deal count fell 25.7% to 1,605 (Crunchbase News, July 2026). CB Insights, using a narrower definition, records Q2'26 fintech funding at $11.7B across 726 deals — down 20% and 25% QoQ, the lowest deal volume in over four years, with digital banking nearly doubling QoQ to $2.6B led by Ramp, Airwallex and Mercury (CB Insights, State of Fintech Q2'26). [Verified, both — the divergence is definitional.]

Capital intensity. Varies enormously by model. Software-only infrastructure: low. Lending and balance-sheet businesses: very high, and the capital is debt, not equity. Neobanks: high customer-acquisition cost with long payback.

Regulatory. Heavy and the primary moat. The GENIUS Act (stablecoins) became law in July 2025; a year later, implementing rules are still being written — the FDIC published 144 questions on custody, capital and liquidity, the OCC issued an interpretive proposal in February 2026, and KYC requirements comparable to traditional financial firms are proposed (CoinDesk, July 2026). [Verified.] [Analysis] Regulatory ambiguity of this kind favors incumbents and well-capitalized startups over new entrants: compliance cost is a fixed cost.

Competition. High in consumer; moderate in infrastructure; low in genuinely boring back-office niches.

Business models. Interchange; take-rate on payment volume; net interest margin; SaaS + payments ("payfac") hybrids; per-decision or per-API-call infrastructure pricing.

Revenue potential. Among the highest of any sector when it works, because fintech monetizes a percentage of money movement rather than a seat. Stripe remains privately valued at ~$159B, Ramp at ~$44B (Crunchbase News, July 2026). [Estimate.]

Investor interest. Recovering, concentrated on infrastructure and AI-for-financial-services rather than D2C. Geographic split of H1 2026: US $15B (52%), UK $2.7B, India $1.9B. Three fintech IPOs in H1 2026 (PicPay, AgiBank, PayPay), all foreign issuers listing in New York (Crunchbase News, July 2026); CB Insights counts 4 fintech IPOs in Q2, a four-year low against 25 in Q4'25 (CB Insights Q2'26).

Risks. Credit cycle exposure for any lending-adjacent model; interchange regulation; the extreme liquidity backlog — Stripe, Plaid, Revolut and Monzo have all been private for a very long time and late-stage investors are waiting; and AI-driven fraud, which is both a threat and a market.

Notable companies. Stripe, Ramp, Mercury, Airwallex, Taktile ($110M Series C led by Goldman Sachs Alternatives), Flutterwave (Series E at $3.2B).

Underserved opportunities. [Analysis] Stablecoin-rail cross-border payments now that the legal framework exists but incumbents haven't moved; compliance and AML automation (the cost base every financial institution most wants to cut); insurance and wealth infrastructure in non-US markets; and small-business financial operations outside the US and UK, which is dramatically under-built relative to the US market.

5. Healthtech / Digital Health

Market size. The reliable anchor is total spending, not a "digital health market." US national health expenditure reached an estimated $5.7 trillion in 2025, up 7.3% — a third consecutive year of 7%+ growth — and is projected to approach $9 trillion by 2034 (CMS via AHA, June 2026). [Verified for 2025 estimate; Projection thereafter.] Health price growth is expected to average only ~2.5% through 2026, meaning volume and intensity, not prices, drive the increase.

Demand and growth. Strong from a small base. US digital health startups raised $7.4 billion across 244 deals in H1 2026, up from $6.4B across 245 deals in H1 2025; median deal size rose to $14M from $12M (Rock Health, July 2026). [Verified.] Concentration is severe here too: 19 mega-deals took 45% of all capital, up from 22% in 2024 — 8% of deals absorbed nearly half the money.

Capital intensity. Moderate for software; high for anything touching clinical delivery, where you are buying clinician time and carrying medical cost.

Regulatory. Heavy: HIPAA, FDA for software-as-a-medical-device, state licensure for telehealth, and — the real barrier — reimbursement. [Analysis] Reimbursement pathway, not technology, is the primary determinant of outcome for most healthtech startups, and it is the thing founders most consistently underestimate.

Competition. High in the popular categories (mental health, weight management, scribing), low in unglamorous operational categories.

Business models. B2B2C through employers and payers; direct-to-consumer subscription; per-member-per-month; risk-bearing/value-based care (highest ceiling, highest blowup risk); and SaaS to providers. Notably, 64% of mental health and weight management startups pursue direct-to-consumer sales versus 29% industry-wide (Rock Health, July 2026).

Revenue potential. High ceiling, slow ramp. Enterprise health sales cycles of 12–24 months are normal and should be modeled as such.

Investor interest. Steady and rational, unlike 2021. Top-funded areas in H1 2026: mental health for a seventh consecutive year (Talkiatry $210M, Grow Therapy $150M) and weight management/obesity riding the GLP-1 ecosystem (eMed $200M, Nourish $100M, Midi $100M). M&A is busy: 115 acquisitions in H1 2026, with Q2's 71 deals the busiest quarter since Q3 2021; revenue cycle management is consolidating hard (Ensemble Health at $12B). Oura filed an S-1; Whoop raised $575M at $10.1B (Rock Health, July 2026).

Risks. Reimbursement changes; Medicaid enrollment and Marketplace policy shifts that CMS explicitly flags as decelerating spending growth and raising the uninsured share to a projected 9.5% by 2034; clinical liability for AI recommendations; and GLP-1 category dependence — a large share of recent funding rides one drug class.

Notable companies. Abridge, Talkiatry, Grow Therapy, Hinge Health (public, doubled from IPO price by Q2 close), Tempus (36% YoY revenue growth), Commure, Qualified Health.

Underserved opportunities. [Analysis] Back-office and administrative automation — the unsexiest, largest, most reliably fundable opportunity in healthcare, where roughly a quarter of US health spending is administrative. Also: post-GLP-1 maintenance and muscle-preservation care, which is a large cohort with no established care model; specialty care access outside metro areas; and long-term care / aging-in-place operations, which is demographically certain and technologically neglected.

6. Biotech & Life Sciences

Market size. Global pharmaceutical revenue is the relevant pool and is measured in the high hundreds of billions to low trillions depending on definition; I would not cite a specific vendor figure. The meaningful startup-side measure is R&D capital deployed.

Demand and growth. Structurally unlimited (disease burden), commercially gated by trials and payers. Global biotech venture funding is tracking within its historical $36–40 billion annual range in 2026 — remarkably stable given that everything around it changed (Crunchbase News, 2026). [Verified.] Over half of biotech investment is seed and early stage, and AI-focused biotechs raised over $6 billion in 2026.

Capital intensity. The highest of any sector except space and fusion. A single Phase III program can consume $100M–$1B+ with a binary outcome.

Regulatory. The most stringent environment covered here. FDA/EMA approval is the business model's central risk and its central moat.

Competition. High in hot modalities (obesity/metabolic, ADCs, cell therapy), low in neglected indications — which is precisely because neglected indications have poor commercial economics, not because nobody thought of them.

Business models. Out-licensing to pharma (most common realistic exit); platform-plus-pipeline; M&A by large pharma facing patent cliffs; occasionally, commercialization.

Revenue potential. Extreme and binary. Most biotech startups generate zero product revenue ever and exit via acquisition or fail.

Investor interest. Steady, and — critically — the exit market reopened. At least 12 funded biotechs sold for $1B+ in 2026, and the IPO window opened wide enough that companies are going public remarkably early: Kailera Therapeutics, founded in 2024, went public roughly six months after its Series B, raising $625M in one of biotech's largest-ever IPOs (BioPharma Dive, 2026; Crunchbase News, 2026). Largest 2026 rounds: Isomorphic Labs $2.1B Series B, Earendil Labs $787M, NewLimit $435M Series C, Chai Discovery $400M Series C.

[Analysis] The Kailera pattern — obesity/metabolic asset, in-licensed from Chinese pharma, IPO within months — is the defining biotech trade of 2026 and is being widely imitated. That is usually a sign the trade is late.

Risks. Clinical failure (the base rate is brutal and unchanged by AI); drug-pricing policy; the IPO window closing as fast as it opened; and — specific to 2026 — AI-drug-discovery valuations that price in a hit rate no one has yet demonstrated in the clinic.

Notable companies. Isomorphic Labs, Kailera Therapeutics, NewLimit, Chai Discovery, Parabilis Medicines, Recursion.

Underserved opportunities. [Analysis] Clinical-trial operations and patient recruitment (structurally broken, huge cost line, software-tractable); manufacturing and CDMO capacity, especially outside China amid supply-chain politics; antimicrobial resistance, which is a genuine public-health emergency with a broken commercial model that some kind of pull-incentive policy will eventually fix; and diagnostics for conditions where earlier detection changes cost curves.

7. Climate Tech & Energy

(Grouped: the two sectors have effectively merged, and the merger agent is AI power demand.)

Market size. The demand-side anchor is electricity. Over 90 GW of new US generating capacity, led by solar and storage, is expected online in 2026 (S&P Global Market Intelligence, July 2026). [Projection from a credible industry analyst.] Futurum estimates an ~$80 billion backlog of Azure orders constrained by power availability [Reported — single secondary source; I could not trace this to a Microsoft filing, earnings call or press release, so treat it as an analyst estimate rather than a first-party disclosure] (Futurum, 2026) — the single clearest statement of where the bottleneck now sits.

Demand and growth. Climate tech venture and growth funding was $40.5 billion in 2025, up 8% — the first increase since 2021–22 — but deal count fell 18% (Sightline Climate, 2026). [Verified.] The composition tells the story: clean energy grew 31% to $14.4 billion, a three-year high, driven by grid hardware, energy management software, batteries, nuclear fission and fusion at all-time highs, and next-gen geothermal. Data centers consumed 78% of all built-environment funding.

The stage split is the most important fact in this sector. Growth-stage (Series D+) funding rose 78% while Series C fell 32% to just 45 deals — an all-time low, and seed and Series A fell 20% and 7% (Sightline Climate, 2026; Heatmap News, 2026). [Analysis] Investors have, in Sightline's framing, "essentially declared winners." For a founder this means: if you are not already one of the declared winners, climate tech is one of the hardest places in the market to raise a first or second institutional round, despite the headline sector growth.

Nuclear specifically: US nuclear startups took $6.2 billion across 93 companies in 2025 and $4.5 billion across 81 companies by late July 2026, on pace to break the record (Yahoo Finance / Crunchbase data, 2026). [Verified.]

Capital intensity. Very high for hardware, generation and manufacturing; these are project-finance businesses wearing venture clothing. Low for grid software and energy management — which is where the risk-adjusted returns arguably are.

Regulatory. Decisive in both directions. Interconnection queues, FERC co-location rules, NRC licensing timelines, and — a 2026-specific factor — FEOC (foreign entity of concern) compliance rules that constrain battery supply chains (Davis Graham, 2026). Subsidy policy is politically unstable and should not be underwritten as permanent.

Competition. Low in deep-tech generation (few players can raise the capital); very high in climate software and carbon accounting.

Business models. Power purchase agreements; equipment sales; energy-as-a-service; project development and flip; SaaS for grid and industrial operators.

Revenue potential. Very high in absolute dollars for successful energy assets, but on utility timelines with utility margins. Grid software scales faster with less capital.

Investor interest. Strong but narrow and late-stage. Commonwealth Fusion appeared among Crunchbase's largest weekly rounds of 2026 (Crunchbase News, 2026); VoltaGrid and green-steel plays (Stegra $1.6B, Hydnum Steel $695M) also drew nine- and ten-figure rounds (Crunchbase News, 2026).

Risks. Dependence on AI capex continuing — if data center buildout decelerates, the entire "power for AI" thesis reprices at once. Also: policy reversal, interconnection delays that outlast runway, commodity price cycles, and a Series C funding chasm that will kill otherwise good companies.

Notable companies. Commonwealth Fusion Systems, X-energy, TerraPower, NuScale, Stegra, VoltaGrid, Base Power.

Underserved opportunities. [Analysis] Everything between the generator and the rack: interconnection process software, grid-edge flexibility and demand response, transformer and switchgear supply (a genuine physical shortage), and industrial heat decarbonization. Also, adaptation and resilience — insurance-adjacent, wildfire, water, and extreme-heat infrastructure — which is badly underfunded relative to how much damage is already being priced into insurance markets.

8. Robotics & Physical AI

Market size. No credible aggregate figure; the honest measure is capital deployed. [Analysis] Every "humanoid robot market will reach $X trillion by 2035" number in circulation is a vendor projection built on assumed adoption curves with no deployed base to calibrate against. Ignore them.

Demand and growth. Capital is far ahead of revenue. Physical AI startups raised $47.4 billion across 521 deals in H1 2026 — nearly 4x H2 2025 and more than the entire 2022–2024 period combined ($41.9B) (Crunchbase News, 2026). [Verified.] But Waymo's $16 billion Series D alone was roughly one-third of that total, and the category as defined includes autonomous vehicles, aerospace, drones and industrial automation — it is not principally humanoids.

Capital intensity. Very high. Hardware iteration, supply chains, field service, and safety validation all consume capital before a unit of revenue exists.

Regulatory. Moderate and fragmented: workplace safety (OSHA), machinery directives in the EU, and for autonomous vehicles a patchwork of state and municipal approvals.

Competition. Rapidly rising in humanoids; moderate in specialized industrial robotics where domain integration is the moat.

Business models. Robot-as-a-service (increasingly the default, because it solves the customer's capex objection and the vendor's proof problem); equipment sale plus service contract; per-task or per-outcome pricing in logistics and agriculture.

Revenue potential. [Analysis] Genuinely large in specific, structured, labor-scarce environments — warehouses, agriculture, construction earthmoving, inspection. Genuinely speculative for general-purpose humanoids, where no company has yet demonstrated a deployed fleet doing economically meaningful work at positive unit economics. The gap between those two statements is where most of the 2026 capital has gone, and it is the sector's central risk.

Investor interest. Extremely high. Notable: Physical Intelligence $600M Series B, Bedrock Robotics $270M Series B (autonomous construction), Neura Robotics (backed by Amazon and Nvidia) (CNBC, June 2026). Robotics and hardware account for roughly 15% of Series B funding in 2026 (Crunchbase News, 2026). Exits are starting: Mobileye acquired Mentee Robotics for ~$900M.

Risks. Demo-to-deployment gap; unit economics that only work at volumes no one has reached; hardware supply chains concentrated in China; safety incidents that trigger regulation; and capital requirements that mean a mid-tier player runs out of money before reaching scale.

Notable companies. Waymo, Physical Intelligence, Figure, Neura Robotics, Bedrock Robotics, Skild AI.

Underserved opportunities. [Analysis] The boring layer: teleoperation and human-in-the-loop infrastructure (every deployed robot fleet needs it, almost nobody sells it well), fleet management and OTA for heterogeneous robots, simulation and synthetic data for manipulation, robot-specific insurance and certification, and field service networks. Also: retrofitting existing industrial equipment rather than replacing it, which is what most customers can actually afford.

9. Defense Tech

Market size. The Pentagon's FY2027 request is approximately $1.5 trillion (Defense Security Monitor, April 2026), with the FY2026 request having included roughly $13.4 billion for AI and autonomy (CDO Magazine, 2025) and DoD planning its largest-ever investment in drones and counter-drone systems (DefenseScoop, April 2026). The Senate Armed Services Committee's policy bill creates a combatant command for drones (Breaking Defense, June 2026). [Verified as budget requests — note that requests are not appropriations, and appropriated is not obligated.]

Demand and growth. The fastest-growing venture sector by multiple. Defense tech startups raised $14.6 billion in the first five months of 2026, already exceeding the full-year 2025 record of $9.6 billion — against $1.6B in 2020 and $2.8–3.8B annually in 2022–2024 (Crunchbase News, 2026). [Verified.] That is roughly a 9x increase in six years.

Capital intensity. High. Hardware, testing, security clearances, classified facilities, and a sales cycle measured in years before revenue.

Regulatory. Severe and itself a moat: ITAR, security clearances, FedRAMP/IL levels, and a procurement system that structurally favors incumbents. The counter-trend is real acquisition reform intended to let software-speed companies sell to DoD.

Competition. Rising fast. Every seed fund now has a defense thesis; the primes are responding; and European sovereign-defense budgets have opened a second front of entrants.

Business models. Program-of-record contracts; OTAs and SBIR pathways in; increasingly, own-the-product-sell-the-capability (Anduril's model) rather than cost-plus services.

Revenue potential. Very high for the few that reach program-of-record status, near-zero for the many that do not. [Analysis] This is a sector with a brutal bimodal outcome distribution disguised by a rising tide of funding.

Investor interest. Maximal, and now anticipating exits. Anduril raised a ~$5 billion Series H in May 2026; reported valuations for that round vary by source — Crunchbase's defense snapshot cites $30.5B while its physical-AI snapshot cites $61B, and TechCrunch reported July 2026 talks at ~$100B, "more than 3x last year's mark" (Crunchbase News; Crunchbase News; TechCrunch, July 2026). [Estimate — I flag this discrepancy rather than resolving it; it is a good example of why private valuations should not be treated as data.] Other major rounds: Shield AI $2B Series G at $12.7B, Saronic $1.75B Series D at $9.25B, Mach Industries $300M Series C, Castelion. Nearly 50 defense companies are identified as likely IPO candidates, and drone company Swarmer's 2026 IPO rose over 500% on day one.

Risks. Political and budgetary reversal (a single administration change can reprice the sector); the gap between a budget line item and a signed contract; concentration of customer risk in a single buyer; export-control limits on international expansion; and — the risk few discuss — that valuations now assume program wins that have not happened.

Notable companies. Anduril, Shield AI, Saronic, Castelion, Mach Industries, Helsing (Europe).

Underserved opportunities. [Analysis] Counter-drone at affordable cost-per-intercept (the economics of shooting $100k missiles at $1k drones are unsustainable and everyone knows it); munitions and solid-rocket-motor manufacturing capacity; secure, low-bandwidth, contested-environment communications; European and allied sovereign capability, where the buyer base is expanding faster than the supplier base; and software for the acquisition process itself.

10. Space Tech

Market size. Startup-side capital is the measurable quantity: $20.3 billion in global seed-through-growth funding to space and satellite companies through August 2026 — already a record annual high with four months to run. US $12.7B (60%+), China ~20%, Europe ~10% (Crunchbase News, August 2026). [Verified.]

Demand and growth. Transformed by a single event. SpaceX went public in June 2026 at approximately a $1.7–1.8 trillion valuation, raising $75–80 billion — the largest VC-backed technology IPO ever, after absorbing xAI in a ~$250 billion transaction (Crunchbase News; Venture Monitor; Motley Fool, March 2026). [Verified as an event; the precise valuation and raise figures differ by a few percent across sources.]

Capital intensity. Among the highest in this chapter, with the longest payback periods. Launch and constellation businesses are infrastructure projects.

Regulatory. FAA launch licensing, FCC spectrum and orbital-debris rules, ITAR, and — increasingly binding — orbital slot and spectrum scarcity in LEO.

Competition. Launch is effectively a SpaceX-dominated market with a long tail of subscale competitors. Downstream applications (Earth observation, comms resale, in-space services) are more open and more crowded.

Business models. Launch services; satellite manufacturing and bus supply; data and imagery subscriptions; government and defense contracts (the majority of real revenue); ground-segment-as-a-service.

Revenue potential. [Analysis] Government and defense demand is the only proven large revenue base. Commercial Earth-observation data has repeatedly disappointed against projections — this is a sector where the 2015–2020 cohort's revenue forecasts were wrong by an order of magnitude, and there is no strong reason to think the current cohort's are better calibrated.

Investor interest. Very high, with a real exit track record now. IPOs in 2026: SpaceX, York Space Systems (January, $4B valuation — stock subsequently declined), HawkEye 360 ($416M, shares down from first-day close), Aevex ($320M). M&A: York acquired All.Space ($355M), Orbion and Solestial; Voyager Technologies acquired Astrobotic (~$300M). Large private rounds: Anduril $5B, Yuanxin/SpaceSail ~$1B, K2 Space $500M Series D (Crunchbase News, August 2026).

[Analysis] Note the pattern in those IPOs: the non-SpaceX space listings of 2026 are trading below their first-day closes. The public market is distinguishing sharply between SpaceX and everything else. That is a warning for any space company underwriting a 2027 IPO.

Risks. Launch-cost deflation destroying the economics of anything priced against historical launch costs; orbital congestion and debris liability; extreme government-customer concentration; and valuations set by proximity to the SpaceX narrative rather than by cash flow.

Notable companies. SpaceX (public), Rocket Lab (public), K2 Space, Varda, True Anomaly, Sierra Space.

Underserved opportunities. [Analysis] Ground segment and data pipelines — everyone builds satellites, almost nobody builds the unglamorous infrastructure to get bits from orbit to a usable product. Also: space domain awareness and traffic management (a regulatory requirement forming in real time), in-space servicing and deorbit, and radiation-tolerant compute. "Orbital data centers" are being marketed heavily in 2026; I would treat that category as pre-revenue narrative until someone demonstrates thermal management and downlink economics at scale.

11. Cybersecurity

Market size. Vendor estimates cluster around $200–250B for global security spending in 2026; these are estimates, not measurements, and I cite them only as an order of magnitude. The more reliable indicator is that security is a non-discretionary line item that survives budget cuts.

Demand and growth. Structurally permanent, and AI is expanding it on both sides — new attack surface and new defensive capability.

Funding. $10.6 billion in H1 2026 across all stages, with Q2 seed-through-growth at $4.4 billion, down roughly 30% from both Q1 2026 and Q2 2025, on a similar ~30% decline in round count — yet still producing eight $100M+ mega-rounds (Crunchbase News, 2026). [Verified.] At the seed stage, AI-security startups pulled $855M across 150+ seed rounds in 2026 (Crunchbase News, 2026).

[Analysis] Cybersecurity is one of very few sectors where seed activity is genuinely healthy in 2026. That is a meaningful signal: it means investors believe new categories are still being created, rather than that the incumbents have won.

Capital intensity. Low to moderate — classic software economics, with heavy go-to-market cost being the main capital sink.

Regulatory. A demand driver rather than a barrier: SEC disclosure rules, NIS2 in Europe, DORA for financial services, sector-specific mandates. Compliance deadlines create purchase events.

Competition. Very high, with aggressive platform consolidation by Palo Alto, CrowdStrike, Microsoft and Wiz/Google. The historical pattern — best-of-breed startup gets acquired into a platform — remains the base case.

Business models. Subscription per endpoint/user/workload; consumption-based cloud security; managed detection and response (services-heavy, lower multiple but sticky).

Revenue potential. High and reliable. Security buyers renew.

Investor interest. Solid, unglamorous, and with an unusually good exit market. Largest Q2 2026 rounds: Cyera $600M at $12B, NinjaOne $400M+ at $12.3B, Dream $260M at $3B. M&A: Motorola Solutions acquired counter-drone firm D-Fend Solutions for $1.5B; multiple hundred-million-dollar acquisitions (Crunchbase News, 2026).

Risks. Consolidation squeezing point solutions; buyer fatigue with tool sprawl (the average enterprise runs dozens of security tools and is actively trying to reduce that number); AI reducing the defensibility of detection-rule-based products; and the fact that "AI security" as a category is currently over-seeded relative to demonstrated budget.

Notable companies. Wiz (Google), CrowdStrike, Palo Alto Networks, Cyera, NinjaOne, Dream, Exaforce.

Underserved opportunities. [Analysis] Identity and permission management for non-human actors — AI agents with credentials are a rapidly growing, badly governed attack surface and the existing IAM stack was not designed for it. Also: security for OT/ICS and industrial systems (chronically neglected, increasingly targeted); supply-chain and SBOM verification for AI-generated code; and security tooling priced for mid-market companies, who have the same threat model as enterprises and none of the budget.

12. Edtech

Market size. Institutional education spending globally is measured in trillions, but the addressable software slice is small and famously hard to monetize. [Analysis] Edtech is the clearest case in this chapter where a huge "market size" number is misleading: schools have enormous budgets and almost no discretionary software procurement capacity.

Demand and growth. Weak funding, genuine product demand. Edtech venture funding: $525M across 30 deals in 2024, $1.17B across 63 deals in 2025, and only $435M across 24 deals in the first seven months of 2026 — against $774M across 37 deals in the comparable 2025 period (New Market Pitch edtech funding analysis, 2026). [Estimate — this is an aggregator's dataset with a narrower sector definition than Crunchbase's; directionally consistent with all other reporting, but the absolute levels depend heavily on what counts as edtech.] Concentration is extreme even at this small scale: the top 3 rounds took 63.2% of 2026 YTD capital and the top 10 took 90.4%.

Capital intensity. Low to build. High to sell, because of long institutional sales cycles tied to budget calendars.

Regulatory. FERPA and COPPA in the US; state procurement rules; accreditation requirements for anything credential-bearing; EU data protection for minors.

Competition. High in consumer language/tutoring; moderate in workforce learning; low in genuinely unglamorous institutional operations software.

Business models. Institutional licensing (slow, sticky, low ACV); D2C subscription (fast, churny); B2B2C via employers; outcome-based workforce training (income-share and job-placement models, which have a troubled regulatory history).

Revenue potential. Modest. [Analysis] Edtech has produced remarkably few large outcomes relative to attention, and the 2020–21 cohort largely underperformed. AI tutoring is a real capability improvement but has not yet changed the willingness-to-pay problem: the people who most need tutoring are the least able to pay for it, and the institutions that could pay are the slowest buyers.

Investor interest. Low and falling, with a notable geographic shift: Europe took 62.3% of 2026 YTD edtech capital, up from 34.5% in 2025, driven by Preply, Multiverse and Gizmo. Strongest categories: workforce learning software ($107M YTD, 24.7% of capital) and digital tutoring ($176M YTD) (New Market Pitch, 2026).

Risks. General-purpose AI assistants offering free tutoring that is good enough to destroy paid consumer edtech; public-funding cycles; the persistent gap between engagement metrics and learning outcomes; and reputational risk in the credential-and-placement business model.

Notable companies. Duolingo (public), Coursera (public), Preply, Multiverse, Khan Academy (nonprofit, but the relevant product benchmark), Gizmo.

Underserved opportunities. [Analysis] Workforce and trade upskilling tied to a specific hiring pipeline, where the employer pays and the outcome is measurable — this is the only edtech model with reliable unit economics. Also: teacher-facing administrative automation (grading, IEP documentation, compliance reporting), which saves the one resource schools actually lack; assessment and verification in a world where take-home work is no longer evidence of anything; and vocational training for the skilled trades, which faces a genuine demographic labor shortage.

13. E-Commerce & Retail Tech

Market size. Here we have real government data. US e-commerce sales were $340.2 billion in Q2 2026, 17.1% of $1,986.5 billion in total retail sales, up 12.2% year over year — the first double-digit annual growth in about five years — against total retail growth of 6.7% (US Census Bureau, Quarterly Retail E-Commerce Sales, Q2 2026). [Verified — this is a statistical agency measurement, the highest-quality market-size figure in this chapter.]

Demand and growth. Re-accelerating after several flat years. [Analysis] The reacceleration is the most underdiscussed sector fact of 2026: e-commerce penetration had plateaued post-pandemic, and it has resumed climbing.

Capital intensity. Very high for inventory-carrying D2C brands (working capital is the killer); low for software and infrastructure serving them.

Regulatory. Consumer protection, tariffs and de minimis import rules (a material 2025–26 variable for cross-border sellers), sales-tax nexus, and emerging questions about liability when an autonomous agent makes a purchase.

Competition. Extreme in D2C brands; extreme in commerce software; the "agentic commerce" layer is the one genuinely new competitive front.

Business models. Marketplace take rates; D2C margin; SaaS for merchants; payments attach; retail media (now the highest-margin revenue line for large retailers and the reason retail media networks proliferated).

Revenue potential. [Analysis] The lesson of the 2015–2022 D2C cohort should be treated as settled: venture-funded D2C brands with paid-acquisition-dependent growth are a bad risk-adjusted bet, and the small number of successes obscures a very large graveyard. This is the sector where survivorship bias in "notable companies" lists is most severe.

Investor interest. Modest and selective. Quince appeared among Crunchbase's largest weekly rounds in 2026 (Crunchbase News, 2026), but commerce is not a priority allocation for most funds.

Risks. Customer-acquisition cost inflation; platform dependency (Amazon, Meta, Shopify all extract the surplus); tariff volatility; and the open question of what happens to brand discovery and merchandising if a meaningful share of purchases are mediated by AI agents that optimize for price and specification rather than brand.

Notable companies. Amazon, Shopify, Temu/Shein (public-market and private disruptors), Quince, Faire, Whatnot.

Underserved opportunities. [Analysis] Agentic-commerce infrastructure — the protocols, product-data feeds, authentication and payment rails that let an agent transact on a consumer's behalf — is genuinely new, genuinely early, and being defined right now by open protocol efforts. Also: returns and reverse logistics (a large, growing, structurally unprofitable cost line); B2B wholesale commerce, which remains far less digitized than consumer; and retail media infrastructure for mid-sized retailers who cannot build what Amazon built.

14. Creator Tools & Media

Market size. Creator economy estimates cluster around $200–260 billion for 2026 at ~20%+ CAGR (New Market Pitch; Archive.com compilation). [Estimate — these figures come from marketing-analytics vendors with a direct commercial interest in a large creator economy, definitions vary wildly (some include all influencer marketing spend, some include creator earnings, some double-count both), and I would not build a plan on them.] A harder number from an adjacent market: major gaming platforms' creator payouts exceeded $1.5 billion in 2025 (Roblox $923M in 2024; Fortnite $352M) (BCG via Shattered.io, 2026).

Demand and growth. Growing, but value accrues to platforms. [Analysis] The durable structural fact of the creator economy is that creators capture a small and platform-determined share of the value they generate, and tools sold to creators inherit that weakness — you are selling to a customer base with volatile, platform-dependent income.

Capital intensity. Low for software; high for anything involving licensed content or generative video compute.

Regulatory. Copyright is the binding constraint. US courts have permitted Disney and other studios' infringement claims against AI video generation companies to proceed (Mealey's, 2026; Norton Rose Fulbright, 2026). EU AI Act Article 50 requires disclosure of AI-generated content from August 2026 and watermarking for existing systems from December 2026 (Gibson Dunn, 2026). [Verified.]

Competition. Extremely high; near-zero switching costs; features get copied by platforms within months.

Business models. Creator subscriptions (low ARPU, high churn); take-rate on creator revenue (better aligned, harder to win); brand-marketplace fees; licensing to studios and brands (the highest-value but slowest path).

Revenue potential. Modest for tools; potentially large for licensing infrastructure and for generative-media companies that solve rights.

Investor interest. Low as a standalone category; high where it overlaps with generative AI. Odyssey (world models) raised $310M in a week Crunchbase described as otherwise slow (Crunchbase News, 2026).

Risks. Platform policy changes that destroy a business overnight; copyright liability, now a demonstrated and not merely theoretical risk; commoditization of generation as model quality converges; and the concentration of creator income in a tiny top percentile, which caps the paying customer base.

Notable companies. Patreon, Substack, ElevenLabs, Runway, Midjourney, Odyssey.

Underserved opportunities. [Analysis] Rights, licensing and provenance infrastructure is the clearest opportunity in this space — the litigation makes it a requirement, the studios want it, and almost nobody has built the clearing-house layer. Also: financial services for creators (income smoothing, advances, tax), where the customer has a real unmet need and a measurable income stream; and B2B generative media for the long tail of advertising and training content, which is a larger and less contested market than consumer creative tools.

15. Consumer Apps

Market size. No credible aggregate. The most useful hard data comes from subscription infrastructure: RevenueCat's 2026 report analyzes over 1 billion transactions and $11 billion in annual developer revenue (via TechCrunch, March 2026). [Verified within that dataset, which skews toward small and mid-sized independent developers.]

Demand and growth. The AI consumer app data is the most instructive dataset in this entire chapter, because it quantifies exactly what "AI hype" means at the product level:

(RevenueCat State of Subscription Apps 2026, via TechCrunch) [Verified.]

[Analysis] Read this table carefully, because it is the cleanest available empirical statement of the AI consumer thesis. AI apps convert 52% better and monetize ~40% higher per user — and churn 30% faster with 20% higher refunds. This is the profile of a product that is easy to sell and hard to keep: novelty-driven trial, insufficient durable habit. For a founder it means AI features will flatter your top-of-funnel metrics and punish you 6–12 months later. Underwrite to retention, not conversion.

Capital intensity. Low to build, very high to distribute. Paid user acquisition against incumbents with better data is usually unwinnable.

Regulatory. App store rules (the real regulator), children's privacy, and increasingly AI disclosure obligations.

Competition. Maximal. The single most crowded category in this chapter.

Business models. Subscription; freemium; ads; in-app purchase.

Revenue potential. Power-law distributed in the extreme. Most consumer apps never reach meaningful revenue; a handful reach billions.

Investor interest. Low outside AI-native consumer. a16z's Top 100 Gen AI Consumer Apps series remains the best periodic map of what is actually getting used (a16z, 2026).

Risks. Retention (see table); platform tax and policy; acquisition cost inflation; and the specific 2026 risk that general-purpose assistants absorb single-purpose AI app use cases at zero marginal price.

Notable companies. ChatGPT and Claude consumer apps, Duolingo, Strava, Whoop ($575M raise at $10.1B valuation), Oura (S-1 filed).

Underserved opportunities. [Analysis] Consumer products where AI produces a durable, compounding personal asset — health records, financial state, a knowledge base — rather than a one-shot output, because that is what creates switching costs. Also: aging and caregiving, an enormous, wealthy, underserved demographic; and non-US consumer markets where the local-language product gap is real.

16. Proptech (Real Estate Technology)

Market size. Real estate is the largest asset class in the world; the software slice is tiny and historically hard to monetize. [Analysis] Another sector where the TAM number is worse than useless.

Demand and growth. Stabilized at a lower level. Global proptech startups raised ~$8.7 billion through 2026 to date across 794 deals, on pace to match or slightly exceed 2025's $12.3 billion — but deal count has collapsed from 2,400+ in 2019 and 1,446 in 2025 (Crunchbase News, 2026). [Verified.]

Capital intensity. Extreme for asset-heavy models (iBuying, which failed publicly and expensively); low for software.

Regulatory. Fragmented by municipality: zoning, building codes, licensing, landlord-tenant law, fair housing. Also: the aftermath of US real-estate commission litigation continues to reshape brokerage economics.

Competition. High in brokerage and consumer search; low in construction and property operations.

Business models. SaaS to owners/operators; transaction fees; mortgage and insurance attach; construction-tech equipment and services.

Revenue potential. Good in vertical SaaS with payments attach; poor in consumer search, which is a two-horse race in most markets.

Investor interest. Selective and notably non-US. Four of the five largest 2026 proptech deals were outside the US: Stegra green steel $1.6B (Stockholm), Hydnum Steel $695M (Madrid), Mews $300M Series D (Amsterdam), Nesto $216M Series E (Montreal), with Bedrock Robotics $270M Series B (San Francisco, autonomous construction) the US entry (Crunchbase News, 2026). Investors are backing AI applied to construction, property operations and transaction cost — "generic real estate software faces significant funding challenges."

Exits are the good news. EquipmentShare IPO'd in January raising $747M; Autodesk acquired MaintainX for $3.6B; Compass acquired Anywhere for $1.6B; Procore acquired DroneDeploy for $845M; CoStar bought Zonda for $800M (Crunchbase News, 2026). [Analysis] That is a healthier strategic-acquirer market than proptech has had in years, and it materially raises the realistic exit floor for a decent construction- or operations-software company.

Risks. Interest-rate sensitivity of the underlying asset class; transaction-volume dependence; slow, fragmented, technology-averse buyers; and the memory of iBuying, which has made LPs allergic to balance-sheet real-estate models.

Notable companies. Procore, CoStar, Compass, Mews, EquipmentShare, Bedrock Robotics.

Underserved opportunities. [Analysis] Construction productivity is the standout: it is one of the only large sectors with essentially flat multi-decade labor productivity, it has a severe skilled-labor shortage, and autonomous earthmoving and robotic site work now have credible technology. Also: building retrofit and electrification (regulatory deadlines are arriving in major cities), insurance-driven resilience assessment, and property operations for the vast mid-market of owners too small for enterprise software.

17. Food & Agriculture (Agtech / Foodtech)

Market size. Global agrifood output is measured in trillions; the technology slice is small. The credible startup-side figure: global agrifoodtech funding was $16.2 billion in 2025, flat year over year, with deal count down 12% (AgFunder Global AgriFoodTech Investment Report 2026, via AgFunderNews). [Verified — note that AgFunder is a venture firm publishing sector research it also invests against, a mild conflict of interest; its methodology is nonetheless the most transparent available for this sector.] Crunchbase, using a narrower "agtech" definition, records $1.4 billion across 187 deals through early May 2026, pacing at or slightly below 2025's $4.4B and 2024's $4.6B — and far below the 2021 peak of $10.5B (Crunchbase News, May 2026). [Verified.]

Demand and growth. Weak funding, real operational demand. Composition shifts in 2025 are more informative than the flat headline: upstream (farm and food production) rose 7% to $9 billion while downstream grocery delivery kept contracting — the largest downstream mega-rounds shrank 35% versus 2021. Climate-focused agrifood recovered to $3.9B from $2.8B; China grew 43% and South Korea 171%; debt reached 18.2% of total funding, the highest in a decade; deeptech rose to 32% of agrifood deals from 22% (AgFunder, 2026).

Capital intensity. High for hardware, bioprocessing and anything needing physical facilities. Alternative protein and vertical farming in particular are capital-intensive with a demonstrated record of capital destruction.

Regulatory. USDA/FDA/EPA in the US; EFSA and notably slow novel-food approvals in the EU; gene-editing rules that differ sharply by jurisdiction and are a genuine strategic variable in where you domicile and launch.

Competition. Low in most niches — which reflects low investor appetite rather than an open field.

Business models. Equipment sales and robot-as-a-service; input sales (seed, biologicals); per-acre SaaS; marketplace take rates; B2B ingredient supply.

Revenue potential. Modest, with slow-adopting, price-sensitive, seasonally cash-constrained customers. [Analysis] Farmers are among the most rational and most skeptical buyers in any sector covered here; a product must pay for itself within one growing season or it does not sell.

Investor interest. Low and cautious. The most telling AgFunder finding: deeptech agrifood companies commanded a 78% seed-stage valuation premium but received zero mega-rounds ($200M+) in 2025, versus seven for non-deeptech companies. Investors will fund the science early and refuse to fund the scale-up — a direct legacy of vertical-farming and alt-protein failures. [Analysis] This is structural capital starvation, and founders should plan for it explicitly: assume there is no Series C, and design a business that reaches cash-flow breakeven on Series A/B money.

Notable 2026 rounds. Halter $220M Series E (smart cattle collars, New Zealand), Tomorrow.io $175M Series F (weather), Hynaero $135.2M Series A (amphibious wildfire aircraft), Tropic Biosciences $105M Series C (gene-edited crops). Indian startups took three of the eleven largest deals. Exits are strategic, not IPO: John Deere acquired Guss Automation; BASF acquired AgBiTech (Crunchbase News, May 2026).

Risks. Commodity price cycles; weather; the Series C chasm; sales cycles tied to annual planting decisions; consumer rejection of novel foods; and the reputational drag that the 2021 cohort's failures placed on the whole category.

Notable companies. John Deere (incumbent), Halter, Tomorrow.io, Tropic Biosciences, Carbon Robotics, Upside Foods (cautionary case).

Underserved opportunities. [Analysis] Farm labor automation for specialty crops — the labor shortage is acute, the work is skilled and manual, and almost nothing viable exists. Also: on-farm financial and risk tooling (insurance, hedging, carbon-credit verification) where the customer already buys an analog version; food-safety traceability, which regulation is mandating; and post-harvest loss reduction, the largest and least glamorous efficiency gain available in the food system.

18. Manufacturing & Industrial Tech

(Grouped: the buyer, the sales motion and the regulatory environment are the same.)

Market size. Manufacturing output is measured in trillions globally; the startup-addressable slice is industrial software, automation and supply chain, and no reliable aggregate exists. [Analysis] Size this bottom-up — number of plants in your segment times realistic spend per plant — not from a vendor TAM. Any "smart manufacturing market will reach $X hundred billion" figure is a CAGR extrapolation with no bottom-up validation.

Demand and growth. Driven by three durable forces: (1) reshoring and supply-chain reconfiguration, with CHIPS Act semiconductor fabs and supplier ecosystems in active construction and mid-2026 progress reports showing real capacity coming online (SupplyICs, 2026; Accuris); (2) a skilled-labor shortage as the manufacturing workforce ages out; (3) AI and robotics finally cheap enough for mid-sized plants rather than only for automotive OEMs.

Capital intensity. Very high for anything that builds physical things; moderate for industrial software. [Analysis] "Factory-as-a-startup" models have consistently required several times the capital founders projected, and the venture structure fits them poorly.

Regulatory. OSHA, EPA, product-specific certifications, export controls on advanced manufacturing equipment, and — a live 2026 factor — FEOC and content-origin rules that determine subsidy eligibility for batteries and clean-energy components.

Competition. Low in industrial software (incumbents Siemens, Rockwell and SAP are entrenched but their products are widely disliked, which is the classic setup for displacement); high in contract manufacturing, where the competition is global and price-based.

Business models. Equipment sales plus service contracts; industrial SaaS priced per site or per machine; manufacturing-as-a-service; parts and materials marketplaces.

Revenue potential. Solid and durable rather than spectacular. Industrial contracts are long, renewal rates are high, and customers do not churn casually. The ceiling is lower than pure software; the floor is much higher.

Investor interest. Rising, and one of the clearest beneficiaries of the "physical AI" reframing that made hardware fundable again. Crunchbase's weekly round-ups in 2026 repeatedly feature manufacturing and energy among billion-dollar raises (Crunchbase News, 2026), and semiconductor startup funding is active (SemiEngineering, Q1 2026). Robotics and hardware together account for roughly 15% of 2026 Series B funding (Crunchbase News, 2026).

Risks. Long sales cycles that outrun runway; capital intensity that outruns venture risk appetite; policy dependence for reshoring economics (subsidy regimes change with administrations); and the pilot-to-fleet gap — succeeding at one plant and failing to roll out across twelve is where most industrial startups die.

Notable companies. Siemens, Rockwell Automation (incumbents), Hadrian, Machina Labs, Nominal, EquipmentShare.

Underserved opportunities. [Analysis] The largest genuinely neglected opportunity is software for small and mid-sized manufacturers — tens of thousands of job shops and contract manufacturers running on spreadsheets and 1990s-era ERP, who cannot afford SAP, are facing a labor cliff, and are seeing reshoring demand they cannot currently serve. Also: industrial data infrastructure (getting data off legacy PLCs remains a real, unsolved, unglamorous problem), computer-vision quality inspection, and maintenance-workforce tooling for a retiring technician base.

19. Mobility & Transportation

Market size. Automotive is a multi-trillion-dollar industry; the venture-addressable slice is autonomy, fleet software and charging infrastructure.

Demand and growth. Autonomy crossed from pilot to commercial scale in 2026. Waymo raised $16 billion in February 2026 at a $126 billion valuation — co-led by Alphabet with Dragoneer, DST Global and Sequoia — explicitly to scale internationally to London and Tokyo (TechCrunch, February 2026; CNBC, February 2026). [Verified as an event; note eWEEK reported the valuation at ~$110B — another instance of private-valuation reporting variance (eWEEK, 2026).] By September 2026 Waymo had launched in Denver, San Diego and Tampa, on top of Houston, Dallas, San Antonio and Orlando earlier in the year (TechCrunch, September 2026; CNBC, February 2026).

Capital intensity. Among the highest in this chapter. Waymo's single round exceeded the entire annual venture funding of most sectors profiled here, and represented roughly one-third of all H1 2026 physical-AI funding (Crunchbase News, 2026). [Analysis] This is the defining fact about mobility as a startup sector: the autonomy race is now a contest between entities that can deploy $10B+, which means core autonomy is not a venture-startup opportunity anymore. It is a corporate capital-allocation contest with a venture wrapper.

Regulatory. City-by-city and state-by-state approval; NHTSA oversight; insurance and liability frameworks still being written; municipal labor politics around driver displacement.

Competition. Concentrated at the top (Waymo, Tesla, Zoox, Chinese operators), with a long tail of trucking and constrained-environment autonomy companies.

Business models. Per-ride marketplace; fleet operations; licensing autonomy stacks to OEMs; charging networks; fleet management SaaS.

Revenue potential. Enormous if autonomy works at city scale. Unit economics at current vehicle costs and remote-operator ratios remain unproven in any public disclosure, and should be treated as an open question rather than a solved one.

Risks. A single high-profile safety failure resetting regulatory posture nationally; vehicle capital costs; city-level political resistance; and, for anyone outside the top four, being outspent into irrelevance regardless of technical merit.

Notable companies. Waymo, Tesla, Zoox (Amazon), Aurora, Motive, Nuro.

Underserved opportunities. [Analysis] Everything adjacent to autonomy rather than competing with it: depot operations, cleaning and charging logistics for robotaxi fleets, remote assistance and teleoperation, AV-specific insurance and claims adjudication, and autonomy for constrained environments (ports, mines, yards, airports) where regulatory burden is low and labor economics are compelling. Also: commercial fleet electrification, where total-cost-of-ownership math already works and the real bottleneck is charging infrastructure and financing rather than vehicles.

20. Gaming

Market size. BCG estimates the industry at $263 billion in 2025, growing to $353 billion by 2030 (~6% CAGR), with mobile in-app purchases at roughly $130 billion — about half the market. Cloud gaming is projected to grow from $1.4B to $18.3B by 2030 with users rising from 5 million to 65 million; console hardware is expected to decline 7%; Steam recorded a record $11.1 billion in H1 2026 (BCG via Shattered.io, 2026). [Estimate — and note the vendor spread flagged in the introduction: Statista projects $577.9B and Grand View $322.6B for adjacent definitions of the same industry. Gaming is the textbook illustration of why market-size figures should not be treated as facts.]

Demand and growth. Large, roughly flat, and brutally competitive for attention. The sector's 2026 paradox is well documented: record industry profits alongside record layoffs, with forecasts of roughly 14,666 layoffs in 2026 (Outlook Respawn, 2026; Tech Insider, 2026; Forbes, March 2026).

[Analysis] The explanation is that player time and spend have consolidated into a small number of persistent live-service titles and UGC platforms. A growing market with consolidating attention is the worst possible environment for a new entrant, and it is why gaming is one of the few large sectors where the venture case has genuinely deteriorated rather than merely cooled.

Capital intensity. High and rising for AAA production; low for UGC-platform-native development; moderate for tools and infrastructure.

Regulatory. Loot-box and gambling-adjacent monetization rules in several jurisdictions; children's privacy and age-assurance requirements tightening in the UK and EU; app-store policy, which functions as de facto regulation.

Competition. Maximal, with the additional hazard that your competition includes an effectively infinite supply of free user-generated content.

Business models. Free-to-play with in-app purchase; premium; subscription; UGC platform revenue share; advertising.

Revenue potential. Extreme power law — the most severe distribution in this chapter. A small number of titles earn nearly everything, and the median funded studio returns nothing.

Investor interest. Low, and shifting toward tools, infrastructure and UGC platforms rather than content studios (Forbes Business Council, July 2026). [Analysis] That shift is correct. Funding content studios resembles funding film production more than funding software, and venture fund structures fit it badly — which the 2021–23 gaming-fund cohort demonstrated expensively.

Risks. Hit-driven revenue; rising production costs; platform fees; AI-generated content flooding discovery surfaces; and UGC platforms capturing the next generation of both creators and players before traditional studios reach them.

Notable companies. Roblox, Epic Games, Valve, Tencent, Krafton, Discord.

Underserved opportunities. [Analysis] Tools and infrastructure for UGC creators: BCG's data shows 40% of surveyed gamers consuming more UGC while only 10–15% create it, with major platform creator payouts exceeding $1.5 billion in 2025 (Roblox $923M in 2024, Fortnite $352M). The creation funnel is the bottleneck and therefore the opportunity. Also: player-safety and moderation infrastructure (regulatorily forced, technically hard, poorly served); and AI live-ops and content generation sold to studios as cost reduction rather than sold to players as a feature.

21. Web3 & Blockchain

Market size. Crypto market capitalization is observable but volatile and a poor proxy for a startup market. The meaningful development in 2026 is regulatory, not market-size.

Demand and growth. Bifurcated with unusual clarity: stablecoins and payment rails are attracting capital; consumer web3 applications are not. Q1 2026 saw roughly $2.8 billion of crypto VC flow disproportionately toward stablecoin infrastructure rather than web3 apps (bex.co, March 2026) — [Estimate, secondary source, treat the exact figure as indicative]. Crunchbase's fintech data corroborates the direction, listing stablecoins and blockchain-based asset tracking among the areas where fintech capital concentrated in H1 2026 (Crunchbase News, July 2026).

Capital intensity. Low for protocol software. The historical model of funding via token issuance has largely closed for institutionally credible projects, which is a net positive for capital discipline and a net negative for founders who relied on it.

Regulatory. The dominant variable, and it moved decisively. The GENIUS Act created a federal stablecoin framework in July 2025. A year on, implementing rules remain unfinished: the FDIC published 144 questions on custody, capital and liquidity; the OCC issued an interpretive proposal in February 2026; and rules requiring stablecoin issuers to run KYC comparable to traditional financial firms are proposed but not final (CoinDesk, July 2026). [Verified.] Banks and stablecoin advocates are actively fighting over rewards structures.

Competition. High in stablecoin issuance and payments, where banks are now entering directly; low in institutional-grade tokenization infrastructure.

Business models. Float and reserve yield on stablecoins (the actual business model, and a good one at current rates); transaction and bridging fees; infrastructure SaaS; custody.

Revenue potential. Strong for stablecoin issuers and rails; poor for nearly everything else. [Analysis] The seven-year experiment in consumer web3 applications — NFTs, play-to-earn, DAOs, decentralized social — produced negligible durable revenue. Treat it as a closed question, not an underexplored opportunity. A founder pitching consumer web3 in 2026 is competing against investors' memory of losses.

Investor interest. Focused, disciplined and much smaller than 2021. SVB's framing of 2026 as crypto's "integration year" — institutions adopting crypto rails rather than crypto replacing institutions — is both the consensus and, on the evidence, accurate (CoinDesk, February 2026). VCs surveyed for 2026 describe the shift as "less hype, more maturity" (DL News, 2026).

Risks. Rule-making outcomes that advantage incumbent banks; interest-rate compression destroying stablecoin float economics (the entire business model is rate-dependent); persistent fraud and reputational problems; and political entanglement — CoinDesk notes the sitting US president's roughly $800 million stake in World Liberty Financial's token, a conflict-of-interest overhang across the whole policy process.

Notable companies. Circle, Tether, Coinbase, Bridge (Stripe), Fireblocks, Chainalysis.

Underserved opportunities. [Analysis] Cross-border B2B payments on stablecoin rails, where correspondent banking is genuinely slow and expensive and a regulatory path now exists. Also: compliance and travel-rule infrastructure that the new rules will mandate; tokenized money-market and treasury products for corporate treasurers; and emerging-market dollar access, the one consumer crypto use case with demonstrated organic demand rather than speculative demand.

22. Accessibility Tech

Market size. Vendor estimates for "assistive technology" and "digital accessibility software" range across a very wide band, and I do not find any of them credible enough to quote as fact — segment definitions vary from durable medical equipment to web-compliance SaaS, and the publishers are report-selling outfits with an incentive toward larger numbers (representative vendor estimate: Grand View Research; Market.us compilation). [Estimate, low confidence.] The more useful framing: the WHO estimates over a billion people need assistive products, and aging demographics in high-income countries guarantee that number grows.

Demand and growth. Real, underserved, and increasingly regulatorily compelled. The European Accessibility Act obligations began applying in June 2025 to a broad range of consumer products and services sold in the EU, and the ADA Title II web-accessibility rule phases in for US public entities across 2026–2027. [Analysis] Compliance deadlines are the most reliable demand generator in B2B software, and accessibility now has them — which is the single biggest change in this sector's commercial prospects in a decade.

Capital intensity. Low for software; moderate-to-high for hardware devices, many of which also face medical-device regulatory paths.

Regulatory. EAA, ADA, Section 508 procurement rules, WCAG conformance standards, and FDA where a device makes clinical claims.

Competition. Low — genuinely one of the least crowded areas in this chapter. That is partly opportunity and partly warning: low competition frequently reflects difficult underlying economics rather than collective oversight.

Business models. B2B compliance SaaS (audit, remediation, monitoring); device sales, often reimbursed through insurance or government disability programs; per-seat enterprise licensing; public-sector procurement.

Revenue potential. [Analysis] Moderate, and constrained by a structural payer mismatch: the person who benefits is rarely the person who pays, and the person who pays — an enterprise, a government, an insurer — is buying compliance or cost avoidance rather than capability. Businesses that sell to the compliance buyer scale. Businesses selling directly to disabled consumers generally do not, absent reimbursement.

Investor interest. Low. Accessibility does not appear as a tracked category in any major quarterly venture report reviewed for this chapter — Crunchbase, PitchBook-NVCA and CB Insights all omit it. [Analysis] That absence is itself the finding: it is not tracked because it is not funded at venture scale. Founders here should plan around non-dilutive funding, government procurement, strategic partnerships and early revenue rather than a conventional venture path, and should be skeptical of any investor who claims this is a large near-term venture market.

Risks. Enforcement intensity determines demand, and enforcement is politically variable; reimbursement coverage decisions; and the risk that mainstream AI products — live captioning, real-time translation, screen understanding, voice interfaces — absorb accessibility use cases as free bundled features. That last one is excellent for users and existentially bad for standalone accessibility startups, and it is already happening.

Notable companies. Be My Eyes, Deque Systems, Level Access, Cognixion, Whisper.ai, Voiceitt.

Underserved opportunities. [Analysis] Accessibility conformance tooling built into the developer workflow rather than bolted on as a periodic audit — the compliance deadlines are creating a procurement event and the incumbent tools are widely disliked. Also: cognitive accessibility, which has far less tooling than visual or auditory; workplace accommodation management for enterprises; and, largest and least served, aging-related functional decline, where the customer has money, the need is near-universal, and the existing products are uniformly poor.

Part II: Synthesis — What's Promising, What's Crowded, What's Hype

Everything in this section is [Analysis] — my judgment, built on the sourced data above. Reasonable people disagree, and I try to show my reasoning so you can disagree productively.

The Four Questions That Sort Sectors

Before the lists, the method. I ranked sectors on four things, not on funding volume:

1. Is demand paid for by someone with a budget and a deadline? Regulatory mandates, labor shortages and cost lines a CFO already owns are the most reliable demand. "This would be valuable" is not demand.
2. Is the capital requirement matched to the capital available? A sector can have great fundamentals and still be a bad place to start a company if the funding required to reach scale does not exist at the stage you need it. Climate tech's Series C collapse and agtech's missing mega-rounds are both examples.
3. Does the winner keep anything? Defensibility in 2026 comes from regulatory position, proprietary data with a feedback loop, workflow entrenchment plus money movement, or physical assets. It does not come from a model, a UI, or a six-month head start.
4. What happens to this sector if AI capex decelerates? This is the systemic risk question, and most sector analyses skip it.

Genuinely Promising

1. Cybersecurity. The strongest risk-adjusted profile in this chapter. Demand is non-discretionary and compliance-forced; the exit market is functioning (Cyera at $12B, NinjaOne at $12.3B, Motorola/D-Fend at $1.5B); and crucially, seed activity is healthy — $855M across 150+ AI-security seed rounds in 2026 (Crunchbase News, 2026). Healthy seed activity in a mature sector means investors believe new categories are still being created. The specific opening: identity and permissions for non-human actors. Enterprises are deploying agents with credentials into systems whose IAM stack was designed for humans, and nobody has solved it.

2. Vertical SaaS with payments attach. Unglamorous, unfashionable, and still the most reliable path from zero to $100M ARR available to a founder without access to billions. The AI-disruption narrative has been aimed at horizontal SaaS; vertical software that owns a regulated workflow and the money moving through it has not been disrupted, and the proptech exit data (Autodesk/MaintainX $3.6B, Procore/DroneDeploy $845M) shows strategics paying real prices for it.

3. Healthcare administration and operations. Roughly a quarter of $5.7 trillion in US health spending is administrative (CMS via AHA, June 2026). The revenue-cycle-management consolidation wave — Ensemble Health at $12B, IKS/TruBridge, Med-Metrix — demonstrates acquirer appetite (Rock Health, July 2026). This is a cost line every CFO in healthcare is actively trying to cut, which is the cleanest form of demand that exists.

4. Grid, interconnection and energy infrastructure software. [Reported — single secondary source] Futurum estimates an $80 billion Azure order backlog Microsoft cannot fulfill because of power constraints; I could not trace this to a Microsoft filing or earnings call, and it should be treated as an analyst estimate rather than a first-party disclosure (Futurum, 2026) is the most specific statement of a bottleneck anywhere in this chapter. Clean energy funding grew 31% to $14.4B in 2025 (Sightline Climate, 2026). The software layer — interconnection process management, grid-edge flexibility, demand response — requires a fraction of the capital of generation and captures a real bottleneck. Caveat: this thesis is levered to AI capex continuing. See the risk note below.

5. Construction and industrial productivity. Flat multi-decade labor productivity, a severe skilled-labor shortage, credible autonomous-equipment technology (Bedrock Robotics $270M Series B), and — newly — a functioning strategic acquirer market. The mid-market manufacturer running 1990s ERP is the single most underserved software buyer I encountered in this research.

6. Defense tech — with a specific caveat. The growth is real: $14.6B in five months against a $9.6B full-year 2025 record (Crunchbase News, 2026). But the promising part is not competing with Anduril. It is the supply chain beneath the primes: affordable counter-drone intercept economics, solid rocket motor and munitions capacity, contested-environment comms, and European sovereign capability where the buyer base is expanding faster than the supplier base.

7. Biotech — selectively, and countercyclically. Funding held steady at $36–40B while everything else distorted, the IPO window reopened (Kailera: founded 2024, public within six months of its Series B), and at least 12 companies sold for $1B+ in 2026 (Crunchbase News, 2026). Biotech's relative stability during the AI distortion is evidence that its capital base is structurally separate. The opportunity is in clinical-trial operations and manufacturing, not in another obesity asset.

Overcrowded

1. The AI application layer, generally. Thin wrappers over models with no proprietary data, no workflow entrenchment and no regulatory position. The empirical case is the RevenueCat consumer data: AI apps convert 52% better and churn 30% faster with 20% higher refunds (via TechCrunch, March 2026). That is a novelty-purchase profile. Easy to sell, hard to keep.

2. AI coding tools. The demand is the most validated in AI, which is exactly why this is crowded: model labs, independent startups and platform incumbents are all competing for the same developer, who has near-zero switching costs, while the independents resell inference from their own competitors at capped gross margin. Fortune's "crossroads" framing of even the category leader is apt (Fortune, March 2026).

3. Humanoid robotics. $47.4B into physical AI in H1 2026 (Crunchbase News, 2026) — but no company has publicly demonstrated a deployed humanoid fleet doing economically meaningful work at positive unit economics. Capital is years ahead of evidence. Specialized industrial robotics in structured environments is a different and much better business.

4. Consumer apps and creator tools. Maximal competition, near-zero switching costs, platform dependency, and features copied within months. Creator tools additionally sell to a customer base with volatile, platform-determined income.

5. Gaming content studios. Record profits with ~14,666 forecast layoffs (Tech Insider, 2026) is the signature of a market where attention has consolidated into a few live-service titles and UGC platforms. Growing market, consolidating attention, hit-driven revenue: the worst possible entry conditions.

6. Horizontal AI infrastructure below the frontier. Vector databases, generic orchestration frameworks, prompt management. Heavily seeded in 2023–24, and the functionality is being absorbed into model provider platforms and cloud vendors.

Mainly Hype Right Now

I use "hype" narrowly: the narrative is running materially ahead of the demonstrated economics, not "this will never work."

1. General-purpose humanoid robots. See above. The technology is genuinely improving; the timeline claims and the valuations are not supported by any published deployment economics.

2. Orbital data centers. Heavily marketed in 2026 as a solution to terrestrial power constraints. Nobody has demonstrated thermal management or downlink economics at scale. Treat as pre-revenue narrative. Note also the warning embedded in the 2026 space IPOs: York Space and HawkEye 360 are both trading below their first-day closes (Crunchbase News, August 2026). Public markets are distinguishing sharply between SpaceX and everything else in space.

3. "Agentic" as a product category. Agentic branding is now applied to roughly everything. Where agents genuinely work — coding, structured research, narrow customer support — the value is real. Where "agentic" is a prefix on existing workflow software, it is a pricing strategy.

4. Consumer web3. Not emerging — concluded. Seven years, negligible durable revenue. Stablecoin rails are the real business; the rest is a closed question.

5. Pure outcome-based SaaS pricing. The commentary vastly overstates the speed of this transition. Enterprise procurement, budgeting and forecasting are built around predictable seat licenses, and buyers actively resist unpredictable bills. Hybrid platform-fee-plus-usage will win; the "death of SaaS" thesis is a content genre, and Gartner's upward revision of 2026 software spend to $1.468 trillion (+15.5%) is the evidence against it (Gartner, July 2026).

6. AI drug discovery valuations. Not the science — the valuations. Isomorphic Labs raised $2.1B; the sector has yet to demonstrate an improved clinical success rate, which is the only thing that would justify the multiple. Preclinical speed is not the binding constraint in drug development; Phase II efficacy is.

Capital-Starved (a category most analyses omit)

Sectors where fundamentals are decent but the funding to scale does not exist — which matters enormously for founders and is invisible in funding totals:

• Climate tech Series C. Down 32% to 45 deals, an all-time low, while Series D+ rose 78% (Sightline Climate, 2026). Investors have declared winners. If you are not one, plan for a chasm.
• Agtech scale-up. Deeptech agrifood got a 78% seed valuation premium and zero $200M+ rounds in 2025, versus seven for non-deeptech (AgFunder, 2026). Fund the science, refuse the factory.
• Edtech, broadly. $435M across 24 deals in seven months of 2026, with the top three rounds taking 63.2% (New Market Pitch, 2026).
• Accessibility tech. Not tracked as a category by any major venture data provider.
• Non-AI seed generally. First-time fund formation is on pace for its lowest year since 2016 (Venture Monitor) — and first-time managers are historically the buyers of non-consensus early rounds.

The AI Funding Concentration Debate, Honestly

Somewhere between 70% and 86% of venture dollars went to AI in H1 2026, depending on whose definition you use. Two companies took 43% of global funding. Whether this is a bubble is the most consequential open question for anyone planning a company, and it deserves better than a one-word answer.

The bubble case

Circular financing. The most specific bear argument, most prominently made by Michael Burry: hyperscalers fund AI labs, which spend that money on chips and cloud from the same hyperscalers, so "the same dollar can show up as revenue at more than one stop on the chain." Roughly $879 billion in multi-year purchase commitments circulate among Microsoft, Google, Meta, Amazon, OpenAI and Nvidia — Oracle committed $300B in OpenAI purchases, Microsoft roughly $250B, OpenAI $90B to AMD while taking equity in Nvidia (24/7 Wall St., August 2026; Bloomberg's circular deals graphic, 2026). [Estimate — these are aggregated commitment figures, and commitments are not the same as recognized revenue.]

Hidden leverage. The same reporting alleges the five largest hyperscalers carry roughly $1.65 trillion in off-balance-sheet obligations via special purpose vehicles, exceeding their reported $1.35 trillion in debt. [Estimate, contested, single-source — I flag it as an allegation that has not been independently verified in the reporting I reviewed, not as an established fact.]

The capex-to-revenue gap. $660–690 billion of 2026 hyperscaler capex (Futurum, 2026) against model-layer revenue that, even taking the most recent figures — Anthropic's $65B run rate, OpenAI's ~$40B (CNBC, August 2026) — plus everyone else, remains a modest fraction of the spend. Infrastructure precedes revenue by 18–36 months, so the gap must close on a schedule nobody controls.

Credit market signals. Nvidia's five-year credit default swap spread roughly doubled over two months to mid-2026 (24/7 Wall St., August 2026). Credit markets and equity markets are pricing different scenarios, and credit markets are usually the better forecaster.

Depreciation. GPU useful-life assumptions drive reported earnings across the complex. If the real economic life is shorter than the accounting life, current profits are overstated in a way that compounds.

The fundamentals case

The revenue is real and it is accelerating. Anthropic went $9B → $47B → $65B annualized within seven months, adding $18B of annualized revenue in two months (TechCrunch, August 2026). Whatever else is true, this is not 1999, when the companies at the center had no revenue. These are among the fastest-scaling revenue lines in the history of software.

The capex is funded from cash flow, not primarily from debt. Nvidia recently generated close to $48 billion in free cash flow in a single period (24/7 Wall St., August 2026). The hyperscalers funding this are among the most cash-generative businesses ever built. The dot-com bust was catastrophic partly because it was financed by debt against companies with no cash flow.

Demand is constrained by supply, not the reverse. Microsoft's $80B unfulfillable Azure backlog and Amazon's statement that "AI capacity is being monetized as quickly as it is installed" (Futurum, 2026) describe a shortage. Bubbles are characterized by unsold inventory. This is the strongest single argument against the bubble case and it deserves more weight than it usually gets.

Enterprise adoption is showing up in third-party spending data. Gartner revised 2026 software spend upward to $1.468 trillion and data center systems to $822 billion (+62.5%) (Gartner, July 2026). Buyers are paying, not just piloting.

Jevons paradox. Falling inference cost has historically increased total inference consumption. Efficiency gains may expand rather than reduce infrastructure demand.

My reading

[Analysis] The most defensible position is that the frontier-model layer has real, accelerating, supply-constrained revenue, and that the financing structure around it has bubble characteristics anyway. Those are not contradictory. The 1990s telecom buildout involved real demand for bandwidth and a catastrophic capital misallocation; the fiber got used eventually, but the companies that laid it mostly went bankrupt first.

Three things I would flag as underweighted in most commentary:

One: the risk is not distributed where the narrative says. Nvidia, Microsoft, Amazon and Google have diversified revenue and enormous cash generation; they can absorb a bad AI outcome. The fragile positions are the neoclouds, the SPV-financed data center developers, the application-layer companies priced at frontier-lab multiples, and the sectors whose entire investment thesis is levered to AI capex continuing — which, per this chapter, explicitly includes clean energy, nuclear, and a meaningful share of industrial and space tech. A capex deceleration would not be contained within "AI."

Two: the concentration is a more immediate problem than the valuation. Even if AI fundamentals fully justify every dollar, a market where 87.5% of capital goes to $100M+ rounds and first-time fund formation is at a decade low is structurally hostile to new company formation across every sector in this chapter. That is happening now, regardless of how the bubble debate resolves. For most readers of this chapter, it is the more actionable fact.

Three: "is it a bubble" is the wrong question for a founder. The useful question is: does my business survive a 50% cut in AI capex? If your customer is a hyperscaler, a neocloud, or a company whose funding depends on AI enthusiasm, the answer is probably no, and you should be pricing that risk into your runway planning. If your customer is a hospital cutting administrative cost, a manufacturer facing a labor cliff, a bank facing a compliance deadline, or a defense ministry rearming, you are substantially insulated — and you are also, not coincidentally, in the sectors this chapter rates as most promising.

A closing note on the data itself

Every headline number in this chapter comes from an organization with a commercial interest in the story it tells. Crunchbase, PitchBook and CB Insights sell subscriptions to a narrative of a large, trackable, dynamic venture market. Market-research vendors sell reports that get bigger when the market does. Rock Health, AgFunder and Sightline Climate are all sector advocates as well as sector analysts. a16z publishes rankings of the category it invests in. None of this makes the data false — it is the best data available, and I have used it — but it does mean every number here tilts, gently, toward optimism. The companies that failed are missing from the databases. The rounds that did not happen are not counted. Adjust accordingly.

Sources

All URLs accessed on or before September 15, 2026. Organized by source type, with a reliability note where it matters.

Primary venture-data reports (highest reliability for funding figures; note all providers sell data subscriptions)

• Q2 2026 PitchBook-NVCA Venture Monitor (PDF) — July 2026
• Crunchbase News: Global Startup Investment Hit Record $510B In H1 2026 — July 2026
• CB Insights: State of Venture Q2'26 — July 2026
• CB Insights: State of Fintech Q2'26 — July 2026
• Rock Health: H1 2026 funding and market overview — July 2026
• AgFunderNews: Agrifoodtech funding is flat (AgFunder Global AgriFoodTech Investment Report 2026) — 2026 (publisher is also a venture investor in the sector)
• Sightline Climate: $40.5bn and 8% uptick as power demand drives 2025 investment — 2026 (sector advocate as well as analyst)
• NonPublic: The Q2 2026 Venture Monitor — Record Numbers, Narrow Recovery — 2026

Government and official statistical sources (highest reliability for market size)

• US Census Bureau: Quarterly Retail E-Commerce Sales, Q2 2026 (PDF) — August 2026
• AHA: CMS projects national health spending grew to $5.7 trillion in 2025 — June 2026
• CDO Magazine: Pentagon Seeks $13.4bn for AI and Autonomy, FY 2026 Budget Request — 2025
• DefenseScoop: DOD moves to make its largest-ever investment in drones and anti-drone weapons — April 2026
• Defense Security Monitor: Inside the Pentagon's Historic $1.5 Trillion FY27 Budget Request — April 2026
• Breaking Defense: SASC's $1.14T defense policy bill creates combatant command for drones — June 2026

Crunchbase News sector snapshots

• Physical AI funding, H1 2026 — 2026
• Defense startup funding hits an all-time record — June 2026
• Space tech startup funding orbits new highs — August 2026
• Cybersecurity startup funding, H1 2026 — 2026
• AI seed investors flock to cybersecurity — 2026
• Fintech funding surges 23% in H1 2026 — July 2026
• Biotech startup investment held steady even as AI funding surged — 2026
• Proptech funding holds up, but investors are placing different bets — 2026
• Agtech startups face a drier funding climate — May 2026
• Foundational AI startup funding in Q1 was double all of 2025 — 2026
• The Series B pipeline looks refreshingly diversified — 2026
• Biggest funding rounds: megarounds led by enterprise software, AI, and space tech — June 2026
• Biggest funding rounds: billion-dollar raises in manufacturing, energy and AI — 2026
• Biggest funding rounds: The Boring Co., Cognition and Motive — 2026
• Biggest funding rounds: Safe Superintelligence and Commonwealth Fusion — 2026
• Biggest funding rounds: Odyssey leads in a slower week — 2026
• Biggest funding rounds: AI, robotics and e-commerce top the ranks — 2026

Original reporting — company and sector news

• CNBC: Anthropic tells investors annualized revenue run rate climbed to $65 billion in July — August 2026
• TechCrunch: Anthropic's annualized revenue surges to $65B — August 2026
• TechCrunch: Anduril reportedly in talks to raise at $100B valuation — July 2026
• TechCrunch: Waymo raises $16 billion to scale robotaxi fleet — February 2026
• TechCrunch: Waymo accelerates robotaxi expansion — Denver, San Diego, Tampa — September 2026
• TechCrunch: AI-powered apps struggle with long-term retention (RevenueCat State of Subscription Apps 2026) — March 2026
• CNBC: Waymo announces $16 billion funding round — February 2026
• CNBC: Waymo opens robotaxi service to select riders in four more US cities — February 2026
• CNBC: Humanoid robotics company Neura Robotics backed by Amazon, Nvidia — June 2026
• eWEEK: Waymo eyes $16B raise — February 2026
• Fortune: Cursor's crossroads — March 2026
• The Next Web: Cursor raising $2 billion at $50 billion valuation — 2026
• BioPharma Dive: Kailera nets $625M in one of biotech's biggest-ever IPOs — 2026
• Motley Fool: SpaceX absorbed xAI at a combined $1.25 trillion valuation — March 2026
• Yahoo Finance / Crunchbase data: VC money floods into US nuclear startups — 2026
• Heatmap News: Funding for early-stage climate tech is drying up — 2026
• Forbes: The game business is bigger than ever — so why are so many having problems? — March 2026
• Forbes Business Council: Where gaming capital is moving in 2026 — July 2026
• Outlook Respawn: Record profits, record layoffs — inside gaming's 2026 paradox — 2026
• Tech Insider: Games industry reset 2026 — 14,666 layoffs forecast — 2026
• CoinDesk: The GENIUS Act turns 1 — State of Crypto — July 2026
• CoinDesk: 2026 is crypto's integration year, Silicon Valley Bank says — February 2026
• DL News: What VCs expect for crypto investments in 2026 — 2026
• SemiEngineering: Startup funding Q1 2026 — 2026

Legal and regulatory analysis

• Gibson Dunn: EU AI Act Omnibus Agreement — postponed high-risk deadlines — 2026
• Norton Rose Fulbright: An update on AI copyright cases in 2026 — 2026
• Mealey's: AI video generation companies must face copyright suit from Disney, others — 2026
• Davis Graham: Battery storage for data centers in 2026 — FEOC compliance, FERC co-location — 2026

Industry analysis and forecasts (vendor sources — treat projections with caution)

• Gartner: Worldwide IT spending to grow 14.2% in 2026, totaling $6.37 trillion — July 2026
• S&P Global Market Intelligence: US Grid Outlook 2026 — July 2026
• Futurum Group: AI Capex 2026 — The $690B Infrastructure Sprint — 2026
• Deloitte TMT Predictions 2026: SaaS and AI agents — 2026
• a16z: The Top 100 Gen AI Consumer Apps, 6th Edition — 2026 (publisher invests in the category it ranks)
• Shattered.io summary of BCG gaming industry data — 2026
• SupplyICs: CHIPS Act and global semiconductor reshoring — mid-2026 progress — 2026
• Accuris: The CHIPS Act and semiconductor reshoring — 2026

Lower-reliability sources (cited with explicit caveats in the text)

These are aggregators, secondary summaries, or market-research vendors whose figures I flagged as estimates rather than verified data. I include them for traceability, not endorsement.

• New Market Pitch: EdTech funding trends 2026 — aggregator dataset, narrower sector definition than Crunchbase
• New Market Pitch: Creator economy market size 2026 — vendor estimate
• Archive.com: Creator economy market size statistics — marketing-analytics vendor with commercial interest in a large creator economy
• Grand View Research: Digital accessibility software market report — report-selling vendor; cited only as an example of the genre
• Market.us: Assistive technology statistics — report-selling vendor
• bex.co: The great crypto VC pivot — $2.8B in Q1 2026 — secondary analysis
• 24/7 Wall St.: Michael Burry sounds the alarm again on circular AI financing — August 2026; reports contested allegations about off-balance-sheet leverage that I have flagged as unverified
• Bloomberg: AI circular deals — how Microsoft, OpenAI and Nvidia keep paying each other — 2026; paywalled graphic, cited for the mapping of commitment flows

End of Chapter 2. Research compiled September 15, 2026.