AI capital concentrates, governed deal-room AI rises, and VC underwriting turns project-finance disciplined
The gist
This week, venture work shifted from broad AI enthusiasm to tighter capital discipline, governed data access, and project-style underwriting.
This week’s developments
AI Capital Is Concentrating Into Fewer, Larger Bets
AI funding is not recovering evenly; it is concentrating. Seed activity remains alive, but the market is favoring larger entry checks over more deals: median AI seed financings are about $4.6M versus $3.1M across sectors, and the share of seed rounds at $10M or more has climbed from 2% in 2018 to 9% in 2025, while sub-$5M rounds fell from 93% to 75%.
The later stage is even more compressed. In Q1 2026, AI deal counts were 51% below the Q1 2022 peak even as average deal size more than doubled year over year to roughly $24M. For investors and operators, that means the edge is shifting from sourcing volume to speed, technical diligence, and reserve discipline. Your team will spend less time optimizing for deal count and more time fighting for access, protecting ownership, and deciding which companies deserve concentrated follow-on capital.
How should teams adapt to fewer, larger AI funding rounds?
If you're an individual contributor
- Access is tightening; your edge is speed and technical judgment.
- Build sharper diligence instincts and move faster on follow-ons; volume matters less than being trusted on the few deals that get done.
If you manage a team
- Your team wins by judgment now, not by cranking more deals.
- Coach for fast technical screening and reserve discipline; fewer shots mean each decision and follow-on call matters more.
If you lead the organization
- AI capital is concentrating, so ownership and reserves are the fight.
- Rework sourcing and reserve strategy around access, conviction, and concentration; the org needs fewer, higher-stakes decisions.
Sources
- Menlo's Venky Ganesan: Venture's Next Crack Will Come From Debt, Not Equity — BigGo Finance — BigGo Finance, October 5, 2026
Explains why venture returns now hinge on access, conviction, and concentrated ownership in a few outlier AI companies.
- 20VC: Is Seed Investing Dead Without a $1BN Fund? | Does Ownership and Price Matter When Companies Can Be $1TRN Exits | Are AI Revenue Numbers Real and What to Watch Out For with Venky Ganesan, Menlo Ventures — The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch, October 5, 2026
Explains how venture firms are concentrating capital, accepting less ownership, and scrutinizing AI revenue quality.
DFIN Puts Governed AI Inside the Deal Room
On Sept. 14, 2026, DFIN added AI Document Summaries to Venue for supported PDF and Microsoft Office files, then followed with AI Document Translation for 130+ languages and, on Oct. 27, 2026, the Venue MCP Connector so approved assistants like Microsoft Copilot can access live deal-room content through a governed gateway. That matters because summarization, translation, search, and Q&A are now moving inside the virtual data room instead of flowing through exported files and separate AI tools.
DFIN’s model is built around permission fidelity, auditability, and data isolation: the AI inherits each user’s existing access rights rather than expanding exposure. The shift is less about faster first-pass reading than about a durable operating model for sensitive diligence, where assistant-mediated review can happen without breaking confidentiality controls.
For analysts and associates, the work now extends from consuming documents to designing prompts, verifying outputs, and spotting exceptions. Teams that standardize review workflows around governed AI will build on the cycle-time gains already seen, while cutting risk by keeping legal, security, and counterparties within existing controls.
How should we adapt diligence workflows to governed AI in Venue?
If you're an individual contributor
- Deal-room reading is shifting to AI supervision — your edge is judgment.
- Learn to prompt, verify, and flag exceptions fast; the value now sits in catching what governed AI misses.
Sources
- How to create a good LLM judge — The System Design Newsletter, September 30, 2026
Learn to write focused judge prompts, use labeled examples, and validate outputs against human reviews.
If you manage a team
- Your team’s throughput will rise only if they can police AI output.
- Coach analysts on review standards and exception handling, not just speed; governed AI changes what good looks like.
Sources
- Risk Management in the AI Era: A Playbook for Leaders | FTI — FTI Consulting, September 9, 2026
Learn how leaders set AI governance, review standards, and exception handling for continuous risk management.
- I stopped asking my team to use AI. I asked them to manage it — CIO, September 24, 2026
Learn how governed AI, review checklists, and exception handling help teams deliver faster with higher standards.
If you lead the organization
- Manual diligence workflows are being replaced by governed AI inside the VDR.
- Rework diligence ops around permissioned AI and auditability now, or your team will keep paying for slow, fragmented review.
Sources
- The Missing Layer in Enterprise AI: A Governance Architecture for LLMs and Machine Learning | HackerNoon — HackerNoon, October 23, 2026
Learn how permissioned AI, audit trails, and automated controls streamline diligence without sacrificing compliance.
- Enterprise AI’s Next Test Is Accountability — AiThority, October 6, 2026
How permissioned, auditable AI speeds diligence while preserving human judgment in high-stakes M&A decisions.
- AI Is Now an Enterprise Resource. So Why Are We Still Managing It Like Software? | HackerNoon — HackerNoon, October 4, 2026
Shows why AI needs enterprise governance, ownership, and monitoring to drive measurable outcomes and control costs.
Evergreen Funds Push Liquidity Planning Into a Standing Operating Model
Tree Line Capital Partners closed a $680 million continuation vehicle backed by PGIM and LGT Capital Partners to buy senior secured first-lien loans tied to 60-plus U.S. lower-middle-market companies, giving existing investors liquidity while extending the hold. The deal is notable because it pushes the same liquidity machinery seen in recent GP-led transactions into a more repeatable program: Goldman Sachs and Evercore are testing 10–15-year “evergreen” vehicles, with Jefferies, Raymond James, and Lazard exploring similar formats. For venture teams, that means the work is moving beyond one-off continuation deals toward recurring partial exits, LP cash-or-roll decisions, and distribution planning for aging positions. As this layer becomes more standardized, the firms that can run it cleanly will be better positioned to manage liquidity without forcing a sale.
How should your team operationalize evergreen liquidity planning?
If you're an individual contributor
- Liquidity work is becoming repeatable — not a one-off deal skill.
- Learn the mechanics of partial exits, rollover choices, and distribution timing; that judgment will make you more valuable than pure deal execution.
If you manage a team
- Your team must coach liquidity decisions, not just close investments.
- Build reps' fluency in LP cash-or-roll conversations and aging-position planning, or they'll stay stuck in one-off transaction mode.
If you lead the organization
- Liquidity planning is becoming an operating model, not a special event.
- Design repeatable processes for evergreen vehicles and partial exits now; firms that standardize this will control liquidity without forced sales.
AI Infrastructure VC Adopts Project-Finance Underwriting
By Nov. 1, 2026, the AI infrastructure buildout was already cash-flow negative at several of the biggest buyers of compute and data-center capacity. Amazon posted $42.4B in operating cash flow against $77.3B in property and equipment purchases, a $34.9B gap. Alphabet’s gap was $9.1B, Oracle’s $5.4B, and Meta’s about $1.1B, while Microsoft stayed positive at $58.0B of operating cash flow versus $39.4B of capex. Across Amazon, Alphabet, Meta, and Microsoft, capex totaled $201.9B against $175.5B of operating cash flow; adding Oracle widens the shortfall to roughly $31.8B. Providers are filling that mismatch with GPU-backed loans, take-or-pay contracts, leases, and SPVs, including Lambda’s $926M term loan and Volta’s $5B financing package.
That pushes venture underwriting away from software-style growth assumptions and toward infrastructure-grade capital discipline. Gartner’s roughly $1T AI infrastructure spending forecast for 2026, plus Moody’s warning that many facilities are financed on expected demand rather than proven utilization, implies longer payback periods and higher refinancing risk. For investors and operators, the edge now comes from modeling cash conversion, stress-testing offtake, and judging power access, utilization, and asset-level unit economics—not just headline demand.
How should we underwrite AI infrastructure deals differently now?
If you're an individual contributor
- AI infra diligence now rewards cash discipline, not just growth stories.
- Learn to model cash conversion, utilization, and refinancing risk—your edge is judging whether demand can actually pay back the buildout.
If you manage a team
- Your team must underwrite like lenders, not software buyers.
- Coach analysts to stress-test offtake, power access, and asset economics; the old TAM-first memo is no longer enough.
If you lead the organization
- Your investment model is being rewritten around project-finance risk.
- Rebuild diligence and portfolio support around cash flow, contract quality, and refinancing exposure—or you’ll miss the real failure mode.
Sources
- Financing the AI buildout | Brookings — Brookings, September 23, 2026
Brookings warns AI infrastructure is shifting to opaque financing, demanding cash-flow, contract, and refinancing risk scrutiny.