Nvidia is paying AI coding startup Poolside roughly $7 billion combined — a $6 billion license fee for its model-training platform plus a $1 billion direct equity stake — in a deal first reported by Newcomer on August 20 and corroborated by Bloomberg, The Information, and PYMNTS. It’s the latest, and one of the largest, in a string of Nvidia bets on AI labs that has industry watchers asking whether the company is just recycling its own chip revenue back to itself.
The Nvidia-Poolside Deal, Broken Down
Two separate pieces make up the roughly $7 billion package, according to Newcomer’s sourcing:
A $6 billion license fee for Poolside’s “Model Factory” — the internal platform and tooling the startup built to train its own AI models — rather than for the models’ output alone. That fee is expected to be paid out to Poolside’s existing investors by the end of 2027.
A $1 billion direct equity investment, valuing Poolside at $12 billion pre-money. Bloomberg, TheNextWeb, and Dealroom all report consistent figures for both pieces. One outlier headline from Gurufocus cites a “$10 billion” investment, but that number doesn’t match any other outlet’s reporting and appears to be an error — treat the $6B license plus $1B equity structure as the reliable figure.
As part of the arrangement, 109 Poolside engineers and researchers — detailed in Newcomer’s original reporting — have received offers to join Nvidia, where they’re expected to work on the company’s open-weight Nemotron model effort. Poolside’s co-founders — Jason Warner, GitHub’s former CTO, and Eiso Kant — are staying on to run what remains of the independent company. In a note to shareholders, Poolside was explicit: “This is not an acquisition and it is not an acquihire.”
Why Structure It This Way
The backstory explains the odd shape of the deal. Poolside had tried and failed to close a conventional funding round and needed roughly $2 billion to stand up a 40,000-GPU cluster built on Nvidia’s GB300 chips, scheduled to come online in January 2027. Rather than a straight acquisition — which would draw antitrust scrutiny — Nvidia licensed the training platform, took a minority equity stake, and hired the bulk of the technical team. It’s a playbook Nvidia has run before: a similar license-and-hire structure covered chip startup Groq in December 2025 and networking startup Enfabrica for roughly $900 million in September 2025.
Poolside itself builds proprietary coding models — its open-weight “Laguna” family, following earlier “Malibu” and “Point” releases — trained with an execution-feedback method the company calls RLCEF, which uses the deterministic output of running code as a training signal. That puts it in competition with Cursor, Cognition’s Devin, GitHub Copilot, and Anthropic’s Claude Code, though The Information notes analysts are still unclear on exactly what justifies a $12 billion valuation for the license alone.
The Circular Financing Question
The Poolside deal lands in the middle of a much bigger story: Nvidia is reportedly assembling more than $750 billion in AI-related commitments through 2026, including a roughly $500 billion arrangement with SK Group, up to a $250 billion compute-lease backstop for OpenAI, a $30 billion direct equity stake in OpenAI, and up to $10 billion committed to Anthropic. Critics — including investor Michael Burry — have described the pattern as a form of vendor financing: Nvidia hands out capital to AI labs, and much of that money flows straight back into orders for Nvidia GPUs. Nvidia disputes the “vendor financing” label, and CEO Jensen Huang has framed the strategy more simply: “We don’t pick winners. We need to support everyone.” The Poolside deal, being non-exclusive and structured around a license rather than a straight cash-for-chips arrangement, is a smaller and different-shaped version of that broader pattern — but it’s the same underlying logic playing out again.
The same appetite for AI infrastructure capital is showing up elsewhere in the supply chain — see our piece on the AI memory shortage driving up hardware prices for how that demand is rippling into RAM pricing for everyone, not just AI labs.
What’s Next
Neither Nvidia nor Poolside had given an on-record statement to press as of publication, per PYMNTS’s reporting — so treat the deal as “reported,” not formally confirmed by either party, until one of them speaks on the record. Watch for two things over the next few months: whether Poolside’s remaining team can ship on its GB300 cluster timeline now that its engineering headcount has been cut roughly in half by the Nvidia hires, and whether regulators start paying closer attention to Nvidia’s license-and-hire pattern the same way they would a formal acquisition. It’s not the first AI chip-adjacent startup to land a multibillion-dollar backer this year, either — Etched raised at a $21 billion valuation just weeks earlier, a sign of how much capital is still chasing AI infrastructure bets even as the “circular financing” criticism gets louder.
Deixe um comentário