Pete Warden’s bubble signal: nobody is funding the people who’d make AI cheaper

Pete Warden — founder of Jetpac’s early deep learning framework, former TensorFlow mobile lead at Google — published an essay arguing that the AI industry is in a bubble. His evidence is unusual: it’s his own difficulty raising capital for Moonshine, his efficiency-focused startup. He writes: It’s been an uphill battle to raise the investment we’ve needed for Moonshine.

The paradox he names: companies are spending hundreds of billions of dollars on hardware while GPU utilization is usually below 50%. The obvious play would be funding the engineers who can lift utilization from 50% to 90%, doubling effective compute for free. Instead, the capital flows to more GPU purchases.

Why Warden reads this as a bubble signal

For most of computing history, efficiency improvements get richly rewarded because they directly translate to either lower cost or higher capacity at the same cost. The current AI market has decoupled from that pattern. Founders pitching GPU counts get funded; founders pitching 2x utilization improvements get politely declined.

Warden’s diagnosis: Nobody ever got fired for investing in OpenAI. VCs are pattern-matching on safe-sounding deal flow rather than on technical fundamentals. This is the kind of dynamic that historically marks late-stage bubbles, where capital stops differentiating between the activity and the underlying economics. The dot-com parallel he draws: Sun Microsystems dominated unnecessarily — until Google proved cheaper commodity hardware worked better. Sun was the OpenAI-equivalent of the late 1990s; the Google-equivalent for this cycle isn’t visible yet but the structural conditions for one to emerge are present.

The validate-stage signal for indie founders

This matters for early-stage founders for a non-obvious reason: it tells you what the market is currently bad at funding, which is also what’s likely to be underbuilt and therefore opportunity-rich.

If nobody is funding efficiency tools right now, the efficiency tools that ship will face less competition than they should — because the funding cycle is gating the supply. The same is true historically of devtools during the dot-com bust (Atlassian launched into a market with very little well-funded competition), and of B2B SaaS infrastructure during the 2008 downturn (companies like Stripe, Square, and Twilio benefited from less crowded incumbent fields than they would have faced in 2007).

The inverse signal: if VCs are pouring money into a category, that category is structurally more competitive for indie operators because well-funded incumbents are racing for the same customers. Warden’s essay is, in this read, a free market signal for where indie operators should look for opportunity in 2026: anywhere efficiency, optimization, or do more with what we have matters and where capital is currently turning away.

Warden himself secured funding eventually and expects Moonshine cash-flow positive in Q1 2026 — which both validates the underlying market and is the kind of result that, scaled up, would mark the bubble’s correction.