VCs are becoming liquidity engineers, and AI infrastructure is being underwritten like industrial capacity

By DripPublished Updated

The short version

VC work is shifting from pure capital allocation to active portfolio engineering, while AI infrastructure investing is being priced more like capacity planning than software betting.

This week’s developments

  • Structured secondaries and continuation vehicles are now part of the VC job — liquidity management, not just deal selection, is becoming a core partner skill.
  • AI infrastructure is being underwritten against cheaper open models — investors now need technical benchmarking discipline, or they will overpay for shrinking API moats.

Liquidity Engineering Becomes a Core VC Portfolio Function

Structured secondaries moved deeper into venture mainstream this week as LPs and GPs responded to a prolonged exit drought. With IPO and M&A markets still sluggish, distributions are lagging even as capital calls continue, pushing firms toward engineered liquidity instead of waiting for traditional exits. Hamilton Lane called GP-led continuation vehicles an “all-weather” strategy and cited average realized multiples of 3.9x for single-asset deals and 4.5x for multi-asset deals across transactions reviewed from Q4 2022 to Q2 2025. Goldman advanced a different model: LP-level portfolio liquidity across buyout, growth, and venture interests. Strip sales are also gaining traction, typically letting LPs sell 10% to 30% of exposure at 90% to 95% of NAV while keeping upside.

The practical shift is that liquidity is becoming a repeatable portfolio-management tool, not an exceptional workaround. Aging portfolios, longer holding periods, and resistance to discounts on full LP stake sales are making strip sales, startup share sales, and continuation vehicles more attractive. Evercore, Jefferies Private Capital Advisory, and buyers such as family offices are helping normalize these structures.

For VC professionals, the edge is moving toward secondary process design, valuation judgment, LP communication, and governance execution. Expect more work upfront on liquidity planning, advisor coordination, and managing aging assets with explicit process discipline.

How should teams build secondary liquidity capabilities now?

If you're an individual contributor

If you can help a fund or portfolio company navigate secondaries, you become more valuable than a pure deal-sourcing generalist, because liquidity process, valuation judgment, and LP communication are now part of the job instead of edge-case cleanup.

Build fluency in continuation vehicles, strip sales, and NAV-based pricing so you can support real liquidity processes without over-discounting assets or mishandling stakeholder expectations.

If you manage a team

Your team’s value is shifting from only finding and diligencing new deals to also managing aging assets and liquidity events, so the people who can run disciplined secondary processes will start to outshine those who only know primary investing.

Coach your team to treat secondary execution as a repeatable workflow — with tighter valuation memos, LP-ready communication, and clear governance — rather than an ad hoc scramble when exits stall.

If you lead the organization

Liquidity engineering is becoming an operating capability, not a rescue tactic, which means your platform will be judged on whether it can create distributions and manage aging portfolios as deliberately as it sources new capital.

Your next operating-model discussion should include secondary strategy, advisor coverage, and portfolio liquidity planning, because firms that institutionalize this now will have a real edge in fundraising, retention, and portfolio management.

AI Infrastructure Is Being Underwritten Like Industrial Capacity

Open-source and decentralized models are now pressuring the monetization layer directly: closed-model APIs that once commanded 10–20x price premiums are being benchmarked against open-weight alternatives that web research says are roughly 80–95% cheaper at scale, while Llama 4, Qwen 3, and DeepSeek are increasingly viewed as “good enough” for coding and agentic workloads. At the same time, the physical bottleneck is tightening. U.S. market analyses point to a 9–18 GW power shortfall by 2027, hyperscaler delays of 18–36 months, and interconnection queues that can stretch to seven years in Northern Virginia, with Atlanta, Phoenix, Texas, and parts of the Midwest also exposed.

That shifts the valuation frame from “compute is scarce” to “compute must be financed like an industrial asset.” The emerging model for compute factories centers on tokens per dollar of capex, tokens per MW, utilization, and power-adjusted capacity, not software revenue multiples. Meta’s willingness to fund oversupply and reported $26B data-center bonds show where advantage is moving: contract quality, power access, and financing structure.

For VC teams, diligence now needs to include power procurement, interconnection risk, offtake terms, and capex efficiency. Practitioners who can underwrite SPVs, sale-leasebacks, and downside utilization will have an edge over teams still pricing AI on software margins.

How should teams evaluate AI systems beyond model quality alone?

If you're an individual contributor

If you can still only talk about model quality, you’re already behind — the people who become indispensable will be the ones who can evaluate cost, latency, power constraints, and when open-weight models are “good enough” for the job.

Build fluency in unit economics and infrastructure tradeoffs now: learn to compare tokens per dollar, utilization, and deployment constraints so you can spot which AI products are defensible and which are just expensive wrappers.

If you manage a team

Your team’s edge is shifting from picking the best model to underwriting the best system — the managers who coach people on power, capex, and contract risk will outgrow teams still selling a software-margin story.

Start splitting team capability between product judgment and infrastructure diligence, and make sure someone on the team can pressure-test power access, offtake terms, and downside utilization before you back a deal or a roadmap.

If you lead the organization

At your level, AI infrastructure is no longer a software bet — it is an industrial-finance bet, and firms that still price it like SaaS will misallocate capital and miss the real moat.

Rework your investment and operating model around power procurement, financing structure, and asset-level returns, and hire or partner for people who can underwrite SPVs, sale-leasebacks, and long-duration capacity risk.

Stay ahead in Venture Capital

Get the weekly Venture Capital brief in your inbox — the developments, what they mean by seniority, and what to do next.

Want this for the accounts and people you track? Explore Drip.