Daily Tech Briefing: Nvidia Strikes $6B Licensing Deal with AI Coding Startup Poolside & Nvidia's Investment with Perplexity

The CODEW Daily Tech Briefing | August 25, 2026

When Nvidia agreed to pay AI coding startup Poolside $6 billion for a license to its model-building software, hire 109 of its employees, and invest another $1 billion in whatever remained, Poolside's own investor letter was careful to specify what the deal was not: "not an acquisition and it is not an acquihire." Regulators, competitors, and Poolside's own rivals are likely to read that sentence rather differently. This is the second time in eight months Nvidia has run exactly this structure — the first was AI chip startup Groq, for a reported $20 billion — and the repetition is the actual story. A workaround that looked like a one-off in December 2025 now looks like a deliberate, reusable playbook: a real third path for AI startups, sitting somewhere between a traditional acquisition and a traditional funding round, engineered specifically to avoid looking like either.

The Poolside Deal, In Detail

Poolside, a two-and-a-half-year-old startup founded by Jason Warner and Eiso Kant, builds AI models for software development — Malibu, aimed at multi-file code generation and refactoring, and Point, a smaller model built for sub-200-millisecond code completion inside IDEs. The company had raised roughly $626 million prior to this deal, including a $500 million round in October 2024 that valued it at $2 billion, with Nvidia itself already among its backers. Estimated annualized revenue was around $50 million on a roughly 256-person team as of last year.

Under the new arrangement, Nvidia pays $6 billion for a non-exclusive license to Poolside's "Model Factory" — the methodology and tooling Poolside uses to build its models — and separately invests $1 billion in the company at a $12 billion pre-money valuation, roughly six times its 2024 mark and, on last known revenue, something like 240 times sales. Nvidia will extend job offers to 109 Poolside employees, close to half the company's headcount. Poolside's three founders are staying; the company says it will keep operating independently, and the $6 billion license fee is reportedly earmarked for distribution to Poolside's existing investors by the end of next year — functioning, in practice, as a partial liquidity event routed through a licensing contract rather than a stock sale.

The Template: How the Structure Actually Works

Strip away the specific dollar figures and the Poolside deal reduces to three moves executed simultaneously: license the technology (paying for rights to use it, not owning it), hire a large share of the people who built it (getting the talent without a change-of-control transaction), and take a minority equity stake in whatever's left (retaining upside without consolidating the company onto Nvidia's books). Individually, none of these three moves is new — licensing deals, acquihires, and strategic investments have existed for years. What's new, as one analysis of the deal put it, is doing all three at once, at multibillion-dollar scale, on a company that stays legally independent afterward. 

That combination is the template, and it's specifically engineered around a regulatory trigger: the Hart-Scott-Rodino Act requires antitrust pre-clearance for transactions above a certain size, but that requirement is built around acquisitions — a change in corporate control. A non-exclusive license plus a minority investment plus voluntary job offers doesn't transfer control of anything, which means it doesn't obviously trigger the same review, even when the practical outcome — Nvidia gains the technology and the team — looks a great deal like one.

Where This Playbook Came From

Nvidia didn't invent this structure. Microsoft ran an early version in 2024, paying Inflection AI roughly $650 million in licensing fees while hiring co-founders Mustafa Suleyman and Karén Simonyan along with most of Inflection's staff — the deal widely credited as the template everyone since has been adapting. Google followed with two of its own: a reported $2.7 billion non-exclusive license from Character.AI that brought co-founders Noam Shazeer and Daniel De Freitas to Google, and a separate $2.4 billion licensing arrangement with coding startup Windsurf that brought over its leadership team. By the time Nvidia signed its own version with Groq in December 2025, "license the technology, hire the people, leave a shell behind" was already a recognized pattern across the industry — Nvidia adopted an existing playbook rather than authoring a new one, which is itself worth noting, since it means the regulatory question this raises extends well beyond any single company's deal-making.

The Groq Precedent

Nvidia's own first run at this structure, with Groq in December 2025, shows how far the template can be pushed. Nvidia paid a reported $20 billion — roughly 2.9 times Groq's prior $6.9 billion valuation — for a non-exclusive license to Groq's LPU inference architecture, with founder-CEO Jonathan Ross, president Sunny Madra, and the bulk of Groq's engineering team joining Nvidia directly. Reporting on the deal's mechanics described roughly 90% of Groq's staff receiving offers, with payouts structured across three tranches (85% upfront, 10% around mid-2026, the remainder by year-end), unvested shares converting to Nvidia stock at the deal's implied $20 billion valuation, and even employees who hadn't reached their one-year vesting cliff getting it waived for immediate liquidity. 

What remained of Groq continued operating under a new CEO, eventually raising fresh capital of its own — $650 million in June, another $350 million in August — to rebuild as an AI inference "neocloud," at a reset valuation of $3.5 billion, roughly half its pre-deal peak. Groq today is a genuinely different company than the one Nvidia's deal was structured around: same name, same cloud product, almost none of the original leadership or the technology Nvidia licensed away.

The Regulatory Question: Is This Actually a Merger?

Regulators and lawmakers have already flagged the Groq deal specifically as a test case. Senators Elizabeth Warren and Richard Blumenthal opened a formal inquiry in March 2026, characterizing the arrangement as a potential "reverse acquihire" structured to evade Hart-Scott-Rodino filing requirements, and pressed Nvidia for a response along with a request that the DOJ and FTC investigate. 

The FTC separately signaled it was examining a broader pattern of what it's termed "merger in disguise" arrangements across the technology industry, explicitly citing the Microsoft-Inflection deal as an earlier example of the same category. As of the most recent public reporting, no formal enforcement action had been announced against Nvidia over either the Groq or Poolside transactions — the legal question remains open rather than resolved, and the outcome will likely shape how the next wave of AI deal-making gets structured across the industry, not just at Nvidia.

The core tension is straightforward to state and hard to resolve cleanly: a non-exclusive license technically preserves a target company's ability to sell the same technology to Nvidia's competitors, which is the feature regulators would look for as evidence that real competition survives the deal. In practice, once a company has licensed its core technology to the dominant player in its market and handed over close to half its workforce, how much competitive threat it still poses to that same dominant player is a fair question — one current merger law wasn't built to answer, because it was built around transactions that change who owns a company, not transactions that hollow out what a company can still do while leaving the ownership structure nominally untouched.

What This Means for AI Startups Facing the Same Choice

For a well-capitalized but compute-constrained AI startup, this structure is a genuinely rational offer to accept, not just a raw deal dressed up in friendlier language. Poolside's own investor letter reportedly said the company would have needed significantly more Nvidia hardware access than was realistically available to keep competing independently in open-source model development — meaning the actual bottleneck forcing the deal wasn't a lack of capital or talent, but compute access that only Nvidia itself could fully unlock. Framed that way, trading technology and half a workforce for guaranteed hardware access, a cash payout to existing investors, and a continued (if much smaller) independent existence is a defensible outcome compared to running out of runway trying to compete for scarce GPU allocation against the company that makes the GPUs.

That calculus is exactly why this is becoming a genuine third exit path rather than a one-time workaround. Traditional acquisition means giving up the company entirely and surviving a review process that, for a frontier AI deal in 2026, can run a year or more. A traditional funding round doesn't solve a hardware-access problem no amount of capital alone can fix, since GPU allocation, not cash, is often the actual constraint. The license-hire-invest structure solves the hardware problem, delivers a real payout to existing shareholders, and — on paper — leaves the company standing. 

For founders and investors evaluating their options, that's no longer an unusual arrangement worth a paragraph of caveats. It's a template with two Nvidia-specific precedents now on the public record, plus at least three more from Microsoft and Google before it, and every AI startup currently squeezed on compute access has a concrete example of what asking for the same deal might look like.

THE CODEW TAKEAWAY

Nvidia running the same license-hire-invest structure twice in eight months, on two different categories of AI company, confirms this isn't a bespoke solution to one awkward negotiation — it's a repeatable corporate development tool, and probably the most consequential form of AI industry consolidation most people aren't tracking closely, precisely because it's engineered not to look like consolidation. 

The structure lets Nvidia expand its reach into open-source model development and inference silicon, absorb the specific engineering talent it wants, and functionally remove independent competitors from the field, all while avoiding the antitrust review a formal acquisition of either company would have triggered. Whether regulators eventually decide this deserves the same scrutiny as a traditional merger remains genuinely unresolved. What's no longer unresolved is whether other compute-constrained AI startups will take the same deal if Nvidia offers it — Poolside's own reasoning suggests the answer is yes, and every founder watching this pattern now has two detailed precedents to weigh against the alternative of trying to out-compete Nvidia for the hardware access their own roadmap depends on.

---

Source Attribution

1. Bloomberg — Nvidia to Pay AI Startup Poolside a $6 Billion License, Newcomer Says
2. Newcomer — Sources: Poolside Strikes $6 Billion Licensing Deal With Nvidia & Raises $1 Billion for Remaining Company at $12 Billion Valuation
3. The Next Web — Nvidia Pays Poolside $6bn to License Its Model Factory and Hire 109 Staff
4. Value Add Pulse — Nvidia Pays $6B to License Poolside's AI Coding Models
5. MachineBrief — Nvidia's $6B Poolside Deal: Buying Talent Without Buying the Company
6. PYMNTS — Nvidia Pays $6 Billion to License Poolside AI Model-Development Software
7. Constellation Research — Nvidia's Groq Deal: Acquisition, Acquihire or Creative Licensing Deal?
8. IntuitionLabs — Nvidia's $20B Groq Acquisition: Why It Paid 2.9x Valuation for LPU Tech
9. TECHi — Nvidia's $20 Billion Groq Deal: What the Acqui-Hire Means for AI Investors
10. SoftwareSeni — The Nvidia Groq Deal Explained
11. Nate's Newsletter (Substack) — What the Nvidia-Groq Headlines Missed + The 3 Bottlenecks That Actually Explain the Deal
12. WCCFTech — No, Nvidia Isn't Acquiring Groq, But Jensen Just Executed a "Surgical" Masterclass

THE CODEW · DAILY TECH BRIEFING

Editorial Note

The CODEW Daily Tech Briefing is a fast morning read on the day's most important technology signal, plus the handful of other stories worth knowing — built to be read in minutes, with the deeper analytical work reserved for The CODEW's Watch series and Weekly Tech Roundup.

Coverage is based on company announcements, public disclosures, industry reporting, and other publicly available information. Reported figures and sourced-but-unconfirmed details are noted as such. Analysis reflects the reporting period and should be considered in the context of the sources and developments cited.