Nvidia pays Poolside $6bn — the company selling the chips will now build the models
Nvidia is paying $6 billion for a license to Model Factory, the model-development software built by AI startup Poolside, plus $1 billion in equity at a $12 billion pre-money valuation and hiring offers to the 109 engineers behind Poolside's open-source model Laguna. Nvidia states this is neither an acquisition nor an acquihire; Poolside continues as a company and its three co-founders remain. The technology goes into Nvidia's Nemotron open-weight model family
The three lines
- Size — $6bn license plus $1bn investment. Poolside was valued at $12bn pre-money
- People — hiring offers to 109 engineers who built the open-source model Laguna. The three co-founders stay
- Meaning — the technology enters Nvidia's Nemotron open-weight family. The chip supplier is now a model builder
Key questions
- Did Nvidia acquire Poolside?
- **No, and Nvidia has said so explicitly — it is neither an acquisition nor an acquihire.** The deal has three parts. ① A **license to Poolside's model-development software, Model Factory — $6 billion**. ② An **equity investment of $1 billion** at a **$12 billion** pre-money valuation. ③ **Hiring offers to 109 Poolside engineers** who worked on its open-source model **Laguna**. Poolside continues to exist as a company, **its three co-founders remain**, and it raised a further $1 billion for the remaining business. This brief covers the structure itself in 'What a licensing deal is — buying the technology and the people, and leaving the company standing.'
- Why would a chip company want to build models?
- The stated destination is **Nemotron**, Nvidia's family of open-weight models, which is where Poolside's technology and staff are headed. Two broader motives overlap. First, **guaranteeing hardware demand** — every capable open model that exists is a model somebody runs on GPUs, which is why Nvidia has consistently promoted open-source AI. Second, **competition with Chinese open-source models**. DeepSeek and Qwen-family releases have been taking the default slot in developers' toolchains through free, open-weight distribution, a dynamic this brief covered in 'Five models in eight weeks — the real reason China is giving AI away.'
- Why did Poolside give up its own business?
- Poolside's letter to investors put it in terms of **hardware access**: continuing to compete in open-source model development would have required **more Nvidia hardware than it could realistically obtain**. That sentence is the whole shape of the deal. Building models requires GPUs; there is effectively one supplier at the scale required; and that supplier is now building models too. **Before you can be a competitor, you are a customer** — and the relationship runs through the same company either way.
The company that sells the chips has decided to build the thing its customers use the chips for.
Nvidia is paying $6 billion to AI startup Poolside. It is not buying the company.
1. Three parts
| Part | What | Amount |
|---|---|---|
| License | Poolside's model-development software, Model Factory | $6 billion |
| Investment | Equity at a $12 billion pre-money valuation | $1 billion |
| Hiring | Offers to the 109 engineers behind the open-source model Laguna | — |
Nvidia states that this is neither an acquisition nor an acquihire.
Poolside remains a company. Its three co-founders remain with it. It raised a further $1 billion for the business that stays behind.
This is not the first time Nvidia has used this shape. This brief covers the structure — and the objections to it — in "What a licensing deal is — buying the technology and the people, and leaving the company standing."
2. What $6 billion actually bought
Not a finished model. A way of making them.
"Model Factory" is what it sounds like: the pipeline. Preparing data, running training, evaluating results, feeding the findings back in, and repeating. This brief covers what that front half of the work consists of in "AI training vs inference — why the same GPU gets used twice."
Along with the pipeline come 109 people who have actually run it.
Buy a model and you own that model. Buy the pipeline and the people, and you can make the next one. That is why the price sits on the process rather than the product.
The technology is destined for Nemotron, Nvidia's family of open-weight models.
3. Why now
Nvidia occupies an unusual position: almost every company building AI models is its customer. Building models itself means competing with the people it sells to.
Two motives are cited for doing it anyway.
| Motive | Logic |
|---|---|
| Guaranteeing hardware demand | Every capable open model is a model somebody runs on GPUs. This is why Nvidia has pushed open-source AI for years |
| Chinese open-source competition | DeepSeek and Qwen-family releases have been claiming the default slot in developer toolchains through free, open-weight distribution |
The second is cited more often as the immediate trigger. This brief covered the dynamic in "Five models in eight weeks — the real reason China is giving AI away" and "What an open-source AI model is — grades of openness and how to read MoE."
The stake is not price. It is position. Whichever camp produces the open model developers reach for by default also determines the hardware underneath it.
4. Poolside's side of it
The clearest statement of what this deal is came from Poolside's own letter to investors: staying in the open-source model race would have required more Nvidia hardware than the company could realistically obtain.
There is a loop inside that sentence.
- To build models, you need GPUs
- At the required scale, there is effectively one supplier
- That supplier now builds models
Before you can be a competitor, you are a customer. And this structure does not end with one deal — it applies in the same shape to every other company building open models.
This brief covered another version of the same leverage in "Nvidia's $500 billion computing finance platform — GPUs became collateral."
5. How the market took it
Nvidia shares finished the week of the reports 5 percent lower.
Semiconductors were weak in the August 24 US session as well — Micron -5.8 percent, AMD more than -3 percent, Broadcom more than -2 percent, the SOXX ETF -2.7 percent. This brief covered the session in "S&P 500 closes at 7,652.86 on August 24, 2026."
Whether the Poolside deal caused any of that is not established. Tariffs and the Iran sanctions campaign were on the same tape.
6. What we could not confirm
- Headline total — reported as both "$7 billion" (license plus investment) and "$6 billion" (license only). Listed separately here.
- Further spending — reports of billions more for in-house model work could not be sourced to a figure.
- Actual moves — how many of the 109 will transfer was not disclosed.
- Antitrust review — not established at announcement.
- License exclusivity — the agreement itself was not available.
- What to watch next — the next Nemotron release. Whether Poolside's pipeline and people genuinely landed inside Nvidia is a claim only a model can settle.
Sources
- PYMNTS — Nvidia Pays $6 Billion to License Poolside AI Model-Development Software
- DIGITIMES — Nvidia strikes US$6B licensing deal with Poolside to accelerate Nemotron models
- Forbes — Nvidia Pays Poolside $6 Billion To License Its Model Factory And Workers
- The Next Web — Nvidia pays Poolside $6bn to license its model factory and hire 109 staff
- Newcomer — Poolside Strikes $6 Billion Licensing Deal with Nvidia & Raises $1 Billion at $12 Billion Valuation
- Yahoo Finance — Nvidia reportedly strikes $7B licensing and investment deal with AI startup Poolside
- The Daily Caller — Top AI Chipmaker Looks To Make Its Own AI Models