PRIM3 Brief #4: The AI Capex Cycle and Its Spillover Into Web3 Fundraising

The aggregate AI infrastructure capex committed for 2026 is now north of $400B across the four major hyperscalers plus the second tier. Microsoft, Google, Amazon, Meta, plus Oracle, CoreWeave, Lambda, and the newer sovereign-backed entrants, have published 2026 capex guidance that, by Bloomberg's running tally, crossed that threshold in late January and continues to rise. The size of this number is, in many ways, the dominant macro fact for technology investing this year.

Most Web3 commentary treats this as ambient backdrop. PRIM3 thinks that's a mistake. The AI capex cycle is materially affecting capital flow into Web3 in two specific, opposed directions — and founders who don't price the directionality are mispricing their own fundraising environment.

The Headline Number Hides The Allocator Reality

When LPs allocate to "technology" exposure, the AI capex story is now the gravitational centre of that allocation. In conversations PRIM3 has had with sixteen institutional LPs in the last 90 days, fifteen mentioned AI exposure unprompted in the first ten minutes of the call. Web3 exposure came up unprompted in four. That ratio is the substance.

But Web3 allocations aren't getting cut. Most LPs in our conversations are holding or slightly increasing Web3 exposure. What's changed is something subtler: the attention budget of the LP cohort is dominated by AI, and Web3 allocations are increasingly being underwritten as a defensible diversification slice rather than as a primary thesis bet. That shift is structurally important for how Web3 funds raise their next vintages.

For founders at the protocol level, the downstream effect is that the GPs you're pitching are themselves operating in an LP environment where Web3 has to defend its slice of the technology allocation against an AI thesis that is much louder and, in 2026, much better-performing on a near-term IRR basis. GPs aren't going to write you a smaller cheque because of this. They are going to underwrite that cheque with more discipline.

Where The Money Is Actually Coming Out

Where the AI capex cycle is meaningfully competing with Web3 fundraising is in two specific places.

Talent capital. Senior technical talent that, in 2022–2023, would have considered a Web3 founder role is now overwhelmingly being absorbed by AI-side opportunities — both as employees of frontier labs and as founders of AI-native companies. The competition for senior cryptography, distributed systems, and ML talent is being won by AI compensation packages that Web3 protocols can't match on cash. The mitigation Web3 has, equity-and-token packages, works for a subset of founders but not for the broader senior IC layer that protocols need to scale.

The 2026 reality is that a Web3 protocol seeking to hire a senior cryptographer or distributed-systems engineer is competing against AI cash packages that have roughly doubled since 2023. We've seen three direct examples in the last 90 days inside our portfolio. None of them were existential to the team, but each cost meaningful negotiation effort that wouldn't have been necessary in a different macro.

Then public-market attention. Tech-focused public market investors who, three years ago, would have been the natural source of crossover capital into late-stage Web3 plays are now overwhelmingly absorbed by the AI-side narrative. Crossover capital flowing into late-stage Web3 protocols in 2026 is, by our tracking, roughly 40% below the 2024 peak. That's not a market call — that's a flow call. The capital exists; it's allocated elsewhere.

Where The Money Is Actually Flowing Into Web3

Now the other side. And it's where the picture gets more constructive. Real, measurable tailwind for exactly one Web3 sub-sector emerges from the hyperscaler capex surge: decentralized AI infrastructure.

The specific tailwind: every hyperscaler GPU dollar that gets deployed creates a marginal customer for decentralized compute. The hyperscalers are not meeting all the demand they're seeing — AWS, GCP, and Azure all reported sustained GPU rationing through Q4 2025. Some of that unmet demand is leaking to the second-tier cloud providers (CoreWeave, Lambda). Some of it is leaking to decentralized compute networks (io.net, Aethir, Render, Bittensor, Akash). The leak to decentralized compute is small relative to the second-tier cloud leak but it's growing, and the customer profile is shifting.

In 2024, decentralized compute network revenue was overwhelmingly from crypto-native customers running model training for token-launch projects or inference for on-chain bot traffic. In Q1 2026, by the numbers we've seen across three portfolio-adjacent inference networks, a non-trivial share of revenue is now coming from off-chain AI companies — applied AI teams whose burn rate is dominated by inference costs, who arrived at decentralized compute through the cost arbitrage rather than the ideology. We covered this transition in PRIM3 Brief #1's discussion of hedge fund flow, and we'll cover it again in more depth.

The implication: if you're building in the decentralized AI compute layer in 2026, the AI capex cycle is the most important macro tailwind in your sector, and it composes cleanly with the institutional-allocator interest. PRIM3 is overweight in this sub-sector and continuing to look for new positions.

The Trap For Other AI-Web3 Founders

The trap, and we've seen this in roughly 30 pitches in the last 90 days, is the assumption that AI-Web3 as a category benefits from the AI capex cycle.

It doesn't. Most AI-Web3 plays are not in the compute supply layer. Most are in agent economies, decentralized inference markets, AI tooling for crypto users, or AI-native consumer crypto. Those sub-sectors do not benefit mechanically from hyperscaler capex. Their customer base, their unit economics, and their fundraising backdrop are determined by other factors — and pitching them as if they were benefitting from the AI capex tailwind is a positioning error that sophisticated LPs immediately notice.

The founders winning in the non-compute AI-Web3 sub-sectors are the ones positioning honestly: we are building consumer infrastructure that uses AI as a wedge, not infrastructure that competes with hyperscalers on supply. That's a defensible positioning. The "we're going to ride the AI wave" framing is not.

ChainGPT Labs, disclosed: portfolio, Tomer is Investment Director, has been clear-eyed about this framing internally for the last 18 months. The category it's building in is consumer infrastructure for crypto users, with AI as a wedge. That's a different category than the compute layer, and it's underwritten on different grounds.

What This Means For Founder Strategy

A few specific things PRIM3 is telling portfolio founders right now:

For decentralized compute founders: ride the tailwind, but be disciplined about which customer mix you're winning. Inference-side enterprise customers compound differently than crypto-native customers. The 2026 advantage is in winning the off-chain customer profile while keeping the on-chain customer base for diversification.

For agent-economy and consumer-AI-crypto founders: position the wedge honestly. Do not pitch as if the AI capex cycle is a tailwind for your category. The LPs and GPs underwriting Web3 in 2026 are sophisticated enough to know it isn't.

For DeFi and RWA founders not in AI: the AI capex backdrop is structural noise for your sector. Don't waste pitch surface area on AI references. Pitch your sector on its own merits.

For everyone: the technology-allocator's attention budget is, in 2026, dominated by AI. The Web3 pitches that win attention are the ones positioned to compose with that reality, not to compete with it.

Where We're Allocating

PRIM3 has increased its weight on decentralized compute and decentralized inference plays in Q1 2026, with active diligence on three new positions. We have not changed our weight on agent-economy or consumer AI-crypto plays. We have selectively reduced exposure to plays whose fundraising thesis depends on AI-narrative timing rather than AI-substance integration.

If you're building in any of these sub-sectors and the framings here resonate, or don't, push back. Pitch us via prim3.vc.