$258.7 billion of AI venture capital is chasing disruption — but it's concentrated in the wrong places. The biggest opportunities are in the industries founders aren't yet targeting.
AI captured 61% of all global venture capital in 2025 — $258.7 billion out of $427.1 billion total.[1] But this capital is heavily concentrated in a few sectors (legal, healthcare, fintech) while industries with equal or greater disruption potential remain founder-starved. This report maps 12 legacy industries by AI vulnerability, cross-references with startup activity, and identifies where the gap between disruption potential and founder attention creates the best seed-stage opportunities.
Bottom line: Seed capital should flow to vertical AI companies targeting high-vulnerability, low-founder-activity industries — particularly government services, construction, and agriculture. These sectors have the structural characteristics (fragmented, paper-heavy, regulated, labor-intensive) that make them most susceptible to AI disruption, but founders haven't shown up yet. The window is open.
Not all industries are equally vulnerable to AI. The factors that predict disruption potential are structural, not just technical:
A critical insight from Goldman Sachs: despite $258.7 billion in AI venture capital in 2025, AI's contribution to U.S. GDP growth was "basically zero."[3] Over 80% of companies report no productivity gains from AI so far. This isn't a failure of AI — it's a failure of application. Capital has flowed to horizontal tools and foundation models while the vertical application layer — where AI actually touches industry workflows — remains underdeveloped.
TrueBridge Capital's analysis (Forbes, Feb 2026) confirms the shift: "If 2024 and 2025 were about foundation models, 2026 is increasingly about application-layer execution."[4] Vertical AI companies like Harvey (legal) and Abridge (healthcare) are demonstrating growth curves dramatically compressed compared to prior software generations — reaching hundreds of millions in ARR within 2–3 years of launch.
| Industry | AI Displacement Risk | Market Size | Current AI Penetration | Founder Activity | Gap Score |
|---|---|---|---|---|---|
| Office & Admin | 90%[5] | $550B (US BPO) | High — horizontal tools (Copilot, etc.) | High | Low |
| Finance & Accounting | 84%[5] | $624B (global accounting) | Medium — Big Four investing $10B+[6] | High | Low |
| Retail & Customer Support | 82%[5] | $5.5T (US retail) | High — chatbots, personalization ubiquitous | Very High | Low |
| Legal Services | 75%[5] | $1.1T (global) | Medium-High — $4.3B legaltech funding in 2025[2] | Very High | Low |
| Insurance | 72% | $6.4T (global premiums) | Medium — insurtech → $239B by 2033[7] | High | Medium |
| Government / Public Sector | 68% | $8.9T (US gov spending) | Very Low — pilot projects only[8] | Very Low | ★ Very High |
| Construction | 46%[5] | $13T (global) | Very Low — <10% meaningful adoption[7] | Low | ★ Very High |
| Agriculture | 52% | $3.5T (global ag) | Low — $7.8B digital ag market[9] | Low-Medium | ★ High |
| Logistics & Supply Chain | 65% | $9.5T (global logistics) | Medium — Flexport, project44, etc. | Medium | Medium |
| Healthcare (clinical) | 55% | $4.5T (US) | Medium — Abridge, Ambience, Nabla | Very High | Low |
| Education | 60% | $7.3T (global) | Low-Medium | Medium | Medium |
| Energy & Utilities | 45% | $8T+ (global energy) | Low — Schneider/NVIDIA digital twins[7] | Low | ★ High |
Note on displacement scores: The FAIR Framework scores (marked with citations) measure job-level displacement risk — how much competitive pressure AI creates. A 46% displacement score for construction doesn't mean 46% of the industry will be automated. It means nearly half of construction work processes will be fundamentally reshaped by AI, creating massive demand for tools that help the industry adapt.[5]
The U.S. government alone spends $8.9 trillion annually. State and local governments process millions of permits, licenses, benefits applications, and compliance reviews — overwhelmingly through manual, paper-based workflows. AI adoption is in early pilot stages at best.[8]
Why it's underserved: Government procurement is slow, opaque, and relationship-driven. Sales cycles are 12–24 months. Founders prefer faster-moving enterprise markets. But the structural characteristics scream disruption: massive data volumes, repetitive rule-based decisions, chronic staffing shortages, and immense political pressure to modernize.
Who's building: Almost nobody at scale. Polimorphic ($18.6M Series A, 2025) sells AI chatbots and permitting tech to local governments.[8] EffiGov (YC) is building an "AI OS for local governments" starting with 311 call operators.[10] Deloitte published a major GovTech Trends 2026 report focused on "digital agents" for citizen services.[11] But compared to the ~$4.3B flowing into legaltech, government AI is essentially unfunded.
The opportunity:
Construction is the world's largest industry at $13 trillion globally, and one of the least digitized. Meaningful AI adoption is below 10%.[7] AI in construction is expected to grow from $1.3 billion in 2022 to $13.5 billion by 2030 — a 10x expansion that's still a fraction of the industry's size.[12]
Why it's underserved: Construction data is messy, fragmented, and physical-world. Unlike legal or finance, you can't just throw an LLM at a corpus of text. You need integration with sensors, drones, BIM models, project management tools, and safety systems. The technical bar is higher, and the customer base (general contractors, trade subcontractors) is notoriously resistant to new technology.
The opportunity:
The global agricultural market is $3.5 trillion, with a digital agriculture market of $7.8 billion growing at 10.4% CAGR to $17.2 billion by 2033.[9] AI in agriculture specifically is projected to reach $8.5 billion by 2030.[13] But adoption remains concentrated in large-scale operations, leaving the vast majority of the world's 570 million farms untouched.
Why the gap matters: The AI-as-a-Service model is finally making precision agriculture accessible to smaller farms via cloud-based subscriptions rather than heavy hardware investments.[13] This is an inflection point — the technical capability to serve the long tail of agriculture is arriving, but the startups to capture it haven't scaled.
The Heathrow substation fire (£40M cost) and the Iberian Peninsula blackout (€1.6B lost economic output) in recent months have laid bare how fragile energy infrastructure is.[7] Every $1 invested in AI-powered resilience and disaster preparedness saves $13 in post-event costs.[7]
Schneider Electric's partnership with NVIDIA on AI-driven digital twins for energy management is a leading indicator, but the vast majority of utilities — particularly smaller municipal and cooperative utilities — have no AI capability whatsoever. This is structurally similar to the government sector: large, slow-moving incumbents with immense data and clear ROI for AI, but a procurement culture that repels startups.
For context, here's where capital is concentrated — these are validated categories, but entry windows are compressing:
Harvey: $800M+ raised, $8B valuation, $100M ARR, serves 8 of 10 top-grossing US law firms.[2][14] Clio: $500M Series G, $5B valuation.[15] Total legaltech funding hit $4.3B across 356 deals in 2025, with AI-powered tools driving 70% of investment.[2] The growth is real — Harvey reached $100M ARR in roughly two years — but the top of the market is claimed. Seed opportunities exist in legal sub-verticals (regulatory compliance, contract lifecycle for mid-market, litigation analytics) but not in general-purpose legal AI.
Clinical documentation (Abridge, Ambience, Nabla) and medical imaging are well-funded categories. But specific clinical specialties, revenue cycle management for smaller practices, and clinical trial operations remain less contested. Jack's defense and deep-tech focus may find overlap in healthcare AI for military/VA applications.
Insurtech market projected to reach $239.2 billion by 2033 at 27% CAGR.[7] Agentic AI in underwriting and claims is becoming standard — by mid-2025, industry publications documented real-world deployment in claims processing, fraud detection, and underwriting.[16] FurtherAI ($25M Series A from a16z) is automating underwriting, claims, and compliance in the $7T insurance industry. Seed opportunities remain in specialty lines, parametric insurance, and the intersection of AI + climate risk.
The Big Four are collectively investing $10B+ in AI (Deloitte $3B by 2030, KPMG $5B, PwC $1.5B).[6] Thomson Reuters launched agentic AI for tax and audit workflows in December 2025.[17] 64% of accounting firms plan AI investments in 2025, up from 57% in 2024.[6] But the mid-market and SMB segments remain underserved — the Big Four investments serve enterprise, leaving millions of smaller firms without AI-native tools.
The best leading indicator of where startups will cluster in 18 months is where accelerator batches are concentrated today. We analyzed recent cohorts from YC, Neo, and HF0 to map founder activity by vertical.
YC's recent batches represent the single largest dataset of early-stage founder intent. The pattern is unmistakable:
| Metric | W25 Batch | S25 Batch | Signal |
|---|---|---|---|
| Total companies | ~155 | ~160–169 | Steady at scale |
| AI-native / AI-centric | ~90%[20] | 88%[21] | AI is table stakes, not differentiator |
| B2B / Enterprise | ~80% | 80–85%[22] | Consumer is out of fashion |
| Dev tools / Infrastructure | ~30% | ~30%[22] | Picks-and-shovels crowding |
| Agentic AI | ~50% | 50%+[22] | Agent verticalization is the dominant pattern |
| Defense / dual-use | 5 companies | Growing[21] | Emerging but small |
| Government / GovTech | 1–2 (EffiGov, Permitify) | "Least common"[22] | ★ Massive gap |
| Construction | 1 (Permitify — straddles gov + construction) | ~0 | ★ Massive gap |
| Agriculture | 1 (Red Barn Robotics) | ~0 | ★ Massive gap |
| Energy / Utilities | ~0 | ~0 | ★ Massive gap |
Where YC founders are swarming: Dev tools (30%), AI copilots for sales/recruiting/finance (~20%), healthcare AI (~8%), fintech (~10%). The typical S25 company is "an AI agent that does [specific B2B task] for [specific persona]" — vertical agentic AI applied to information-worker workflows. Almost no one is building for physical-world industries.
Neo is smaller (12–15 startups per cohort, <1% acceptance rate) but ultra-high-signal — Partovi's track record includes Cursor ($30B valuation), Kalshi, and Bluesky.[23] Recent portfolio reveals:
Pattern: Neo's portfolio is heavily concentrated in fintech, healthcare AI, and developer tools — the same crowded lanes as YC. Notable absences: zero construction, zero agriculture, zero energy, zero government. This is consistent with Neo's focus on "exceptional young technologists" — founders who come from Big Tech and build for information-worker workflows they personally experienced. Physical-world industries don't attract this founder profile.
HF0 is the most selective accelerator (~15 teams per batch, <1% acceptance), focused on repeat founders with technical depth. Portfolio companies include:[24]
Pattern: HF0 is the only top accelerator with any meaningful physical-world representation. Roofer.com ($7.5M seed, Dallas) is genuinely AI-native construction — using drones and AI for roofing inspections and estimates, not just a lead-gen marketplace with an AI veneer. Smartroof appears construction-adjacent but with less verifiable AI depth. Even so, HF0's dominant pattern remains dev tools and B2B software — these physical-world companies are the exceptions that prove the rule.
| Vertical | YC (W25+S25) | Neo | HF0 | Total Signal |
|---|---|---|---|---|
| Dev Tools / Infra | ~95 companies | Cursor + others | Fileread, Delv | Saturated |
| Fintech | ~25 companies | Moment | Crossmint | Crowded |
| Healthcare | ~15 companies | Anterior | — | Competitive |
| Government | 1–2 companies | — | — | ★ Empty |
| Construction | 0–1 companies | — | Roofer, Smartroof | ★ Nearly empty |
| Agriculture | 1 company | — | — | ★ Empty |
| Energy / Utilities | 0 companies | — | — | ★ Empty |
Accelerator batches show what institutional gatekeepers are funding. Product Hunt and Hacker News show what founders are independently choosing to build — a grassroots signal of where energy and excitement concentrate.
Product Hunt's top AI categories in 2025–2026 reveal where indie founders and small teams are building:[25]
| Category | Activity Level | Examples |
|---|---|---|
| AI Coding Agents | 🔥🔥🔥🔥🔥 | Cursor, Claude Code, Kilo Code, Lovable, Amp |
| AI Writing / Content | 🔥🔥🔥🔥 | Dozens of launches weekly |
| AI Sales / CRM | 🔥🔥🔥🔥 | Apollo, lemlist, Karumi |
| AI Agents (general) | 🔥🔥🔥🔥 | Agentfield, workflow automation tools |
| AI Data / Analytics | 🔥🔥🔥 | Supabase AI, dashboarding tools |
| AI Design / Creative | 🔥🔥🔥 | Image generation, video editing |
| AI for Healthcare | 🔥🔥 | Clinical documentation, diagnostics |
| AI for Education | 🔥 | Tutoring, course creation |
| AI for Construction | — | Essentially nothing |
| AI for Agriculture | — | Essentially nothing |
| AI for Government | — | Essentially nothing |
| AI for Energy | — | Essentially nothing |
HN front-page AI discussions in late 2025 / early 2026 are dominated by:
Conspicuously absent from HN discourse: AI for construction, agriculture, energy, government, manufacturing. These industries simply don't exist in the developer/founder community's consciousness. When they do appear, it's typically in the context of "AI winter" skepticism — "AI hasn't actually changed anything in the real economy."
Plotting every sector from our analysis on disruption potential (vertical) vs. founder activity (horizontal). The top-left quadrant — high potential, low activity — is where seed capital has the most asymmetric opportunity.
Bubble size ≈ market size. Position based on FAIR Framework displacement scores + accelerator/launch data.
The visual confirms the thesis: the four green dots in the top-left quadrant (government, construction, agriculture, energy) are the clearest areas where disruption potential significantly exceeds current founder attention. Every red dot in the bottom-right (legal, finance, office/admin, retail) is a category where capital and founders have already arrived in force.
Cross-referencing the gap analysis with Jack's investment thesis (AI, energy, deep-tech, defense, robotics), three categories offer the strongest fit:
| Opportunity | Why Now | Ideal Founder Profile | Estimated Entry Window |
|---|---|---|---|
| AI for Government Services | DOGE-era political pressure to modernize; chronic staffing shortages; Deloitte GovTech 2026 signals enterprise readiness | Ex-government + technical; understands procurement; patient with sales cycles | 12–18 months before category crowds |
| AI for Construction | 10x growth trajectory ($1.3B → $13.5B); labor shortage crisis; infrastructure bill deployment creating demand | Construction domain expert + ML; understands jobsite realities, not just software | 18–24 months; physical-world complexity creates natural moat |
| AI for Energy/Utility Resilience | Recent infrastructure failures (Heathrow, Iberia); 13:1 resilience ROI; Jack's energy thesis alignment | Energy sector operator + AI; understands grid dynamics, regulatory environment | 12–18 months; regulatory push accelerating |
| AI for Agriculture (SMB) | AIaaS model makes precision ag accessible to smaller farms for first time; $8.5B market by 2030 | Ag-tech domain expert; understands farmer behavior and seasonal cycles | 24+ months; adoption is slow but inevitable |
The consistent pattern across the highest-gap industries (government, construction, agriculture, energy) is that they involve physical-world complexity. Most AI capital has flowed to information-first industries — legal, finance, healthcare documentation — where the data is clean and digital.
2nd Order Physical-world AI is harder to build, which means it's harder to compete against. A legal AI startup faces the risk that OpenAI or Google ships a competitive feature in a model update. A construction AI startup that integrates with drones, BIM systems, project management tools, and safety hardware has a defensibility moat that model improvements can't easily replicate.
3rd Order Physical-world AI unlocks GDP growth that information-only AI can't. Goldman Sachs says AI has added "basically zero" to GDP so far.[3] That's because information-worker productivity gains are real but small in economic terms. The big GDP unlock comes when AI makes physical processes more efficient — construction, manufacturing, agriculture, energy. The $7 trillion GDP boost Goldman predicted requires AI to reach the physical economy. The companies that bridge that gap will capture enormous value.
Generated by Galileo 🔭 · February 27, 2026