Nvidia has struck an unusual deal with AI coding startup Poolside. Instead of buying the company, it will non-exclusively license Poolside's model-development software for $6 billion, plus a separate $1 billion equity investment that values the startup at $12 billion pre-money. The arrangement was first reported by the newsletter Newcomer, which obtained a letter to investors, and was subsequently confirmed by Bloomberg and The Information.
What Nvidia Is Buying — the 'Model Factory' and 109 People
What Nvidia gains is Poolside's Model Factory, a platform for churning out generative AI models purpose-built for software development. Throughout 2026, Poolside has been growing its "Laguna" family of models, designed to write, debug, and optimize code. Nvidia gets access to that entire model-production pipeline.
The talent comes too. Around 109 Poolside employees have received job offers from Nvidia. In effect, a large chunk of the team moves over — but the company itself remains.
Equity investment $1 billion ($12 billion pre-money valuation)
Employees offered jobs 109
Deal structure Not an acquisition — Poolside stays independent
Not an Acquisition, Not an Acquihire
Poolside's founders were explicit that this is neither an acquisition nor an acquihire. The co-founders stay in place and the company continues to operate independently. The non-exclusive nature of the license is central: Poolside can still sell the same technology to other buyers. The original entity survives, now holding $6 billion in cash inflow plus $1 billion in fresh capital.
| Aspect | Traditional acquisition | This Poolside deal |
|---|---|---|
| Entity | Absorbed / dissolved | Stays independent |
| Technology | Ownership transferred | Non-exclusive license |
| People | Employment transferred | 109 individual job offers |
| Regulation | Subject to merger review | Room to avoid review |
Why This Structure
This is not the first time. It extends the "reverse acquihire" lineage that runs through Microsoft–Inflection, Amazon–Adept, Google–Character.AI, and Meta–Scale AI. Rather than buying a startup outright, Big Tech licenses the technology, hires much of the team, and leaves the shell company alive. That provides access to scarce AI models and engineers while creating room to avoid the antitrust merger scrutiny that a conventional acquisition would trigger.
For Nvidia, the deal pushes it beyond chips and infrastructure into the software and model layer, particularly AI coding. The company that sells the GPUs now wants direct control of the ability to build the coding models that run on them.
What It Means
Nvidia keeps using its enormous balance sheet to hoover up AI technology and talent. Days earlier, it was reported to be in early talks with South Korean AI inference-chip startup Rebellions over collaboration, investment, or acquisition. The Poolside deal shows that appetite reaching into the software layer.
The next question is how long regulators leave this "license-and-hire" pattern outside merger review. Critics have repeatedly argued that while the form is not an acquisition, the substance is close to one. Until the deal's fine print and regulatory response are settled, the structure itself — whether it becomes an industry standard — is worth watching more than the numbers.
- Nvidia non-exclusively licenses Poolside's Model Factory for $6B, plus a $1B equity investment ($12B valuation).
- 109 employees offered Nvidia jobs — most of the team moves, but the company stays independent; not an acquisition or acquihire.
- Extends the "reverse acquihire" pattern of Microsoft–Inflection and Google–Character.AI, with antitrust avoidance as the backdrop.
- Signals Nvidia expanding vertically from chips and infrastructure into building AI coding models.
Newcomer — Poolside Strikes $6 Billion Licensing Deal with Nvidia (first report)
The Decoder — Nvidia acquires Poolside's Model Factory and 109 employees for $6 billion
The Next Web — Nvidia pays Poolside $6bn to license its model factory
PYMNTS — Nvidia Pays $6 Billion to License Poolside AI Model-Development Software