Newsletter · · Ashutosh Agarwal
NVIDIA's Acquihire Deals Spark a Rescue or Antitrust Heist Debate - The VC Read - Week of September 17, 2026
The VC Read for the week of September 17, 2026. Podcast synthesis on the hackquisition, the AI deal structure that hires a startup's team, licenses its code, and leaves the shell behind, plus OpenAI delaying its IPO while Anthropic sprints toward a NASDAQ listing, NVIDIA anchoring that offering, and a university endowment CIO on why cash back beats paper markups.
The VC Read
Week of September 17, 2026: NVIDIA's Acquihire Deals Spark a Rescue or Antitrust Heist Debate
How the biggest players in AI figured out a new kind of exit: one that hires the people, licenses the code, and quietly leaves the actual company behind. A lifeline for founders who couldn't raise, or a legal shortcut that guts everyone else? The week of September 10–17, 2026. Some quotes come from just outside the seven-day window; the date is on each one.
The Big Debate: Is the "hackquisition" a rescue or a heist?
Here is a deal structure that barely existed two years ago and is now the defining move in AI: a giant doesn't buy a startup, it hires the startup's people and licenses its code, and leaves the legal shell of the company sitting there, half-alive, for its remaining employees and later investors to sort out.
NVIDIA has done this three times in about a year. As Nicholas Rossolillo laid out on the Chip Stock Investor Podcast, "Nvidia's Biggest Acquisition Ever: $12.9B for Hugging Face (Can It Survive Regulators?)" (September 10, 2026), NVIDIA "didn't actually acquire Infabrica or Grok in December 2025 or most recently Poolside just last month in August 2026. What they did was essentially license the technology and started moving over large parts of the technical team in each of these cases, especially the founders at Infabrica and Grok also moved over with the licensing deal. Thus the name Aquihire." His verdict on why: "This has come under some criticism as well as NVIDIA skirting that regulatory process that would normally take place if it's an outright acquisition."
The contrast that makes the point is the one deal NVIDIA is doing the old way. Its proposed straight-up purchase of Hugging Face (the GitHub-for-AI code repository, with "more than 18 million developers, half a million data sets, a million applications, 200,000 companies" on the platform) is a $12.9 billion acquisition, and Rossolillo thinks "there's more than a small likelihood this one gets shot down," because a big chunk of Hugging Face's team sits in France and "there's going to be a lot of EU regulation besides regulators in China and the US." He notes NVIDIA already walked away from buying Arm at $40 billion in 2022 under exactly that kind of pressure. When a full acquisition risks a two-year regulatory fight or an outright block, the acquihire is the escape hatch.
The bear case: it's antitrust arbitrage dressed up as a licensing deal. The sharpest framing of the week came from Alex Niemann on Moonshots with Peter Diamandis, "Sam Altman: Singularity Slow-Down, Emad Runs 18 Grokbots, Waymo Slashes Hardware 83%" (EP #283, August 27, 2026). He drew a line most people miss: there are ordinary talent acquihires, "which are largely about getting talent," and then there are "acquihires with an H" ("hackquisitions") that are "about, at least ostensibly, avoiding antitrust scrutiny." His description of the game is worth quoting in full:
"If you're NVIDIA and you want to acquihire, say, Poolside, you're going the acquihire route rather than just doing an honest-to-goodness either asset acquisition or conventional acquisition of Poolside because you want to argue, no, actually, we're just a licensee of Poolside rather than the acquirer. No, we're leaving a competitive open-source model layer, blah, blah, blah. This is not tying, blah, blah, blah. That's the argument in principle for acquisition."
In other words, the deal is engineered backwards from the legal defense. Everything (the non-exclusive license, the "we're leaving a competitor behind" language) exists so the buyer can tell a regulator it didn't really buy anything.
And the people left holding the shell can get hurt. When a widely-telegraphed deal collapses, it can wreck the company that was counting on it. On The Twenty Minute VC (20VC), "Jensen Huang Declares AGI Has Arrived… & Anthropic Pulls From Descartes Acquisition" (September 10, 2026), the same week Anthropic walked away from acquiring Descartes after deeper diligence, Harry Stebbings and Rory O'Driscoll worried about exactly this. Stebbings recalled Ben Chestnut of Mailchimp describing how an earlier deal that fell through "basically destroyed the company… and it kind of haunted me." O'Driscoll's point about the wave of "neo-labs" now hoping for this kind of exit was blunt: many "are all doing interesting stuff, but it's not clear if there's a commercially viable standalone business here at scale… there's no fundamentals, there's no massive revenue stream… to support the company and the valuation today. So once belief goes, it can be quite scary down there." A startup whose entire plan is "get acqui-hired by a lab" is one failed leak away from having nothing left.
The bull case: it's a genuine rescue, and demand for it is real. Poolside is the case in point. On the same Moonshots episode, Emad Mostaque explained that Poolside (founded by GitHub's former CTO) "tried to raise $2 billion for a massive Blackwell cluster" at the end of last year "and they couldn't, so they lost that cluster." They had built "a really great, solid open-source model for its size" but couldn't fund the compute to compete. NVIDIA's roughly $6 billion move (which the hosts framed as building a domestic open-weight answer to China's Alibaba, DeepSeek and Kimi, under NVIDIA's "Nemetron" banner) gave that team a home and a mission. NVIDIA's incentive is clean: great open-source models drive demand for its GPUs.
And the demand for teams is, if anything, frantic. David of the Moonshots panel said his own inbound is "an all-time high in acquisition," with buyers like Mercor calling: "I need groups of three, 10, 15 people that work really well together. I don't care what it costs. Like, send them to me tomorrow."
Here's the nuance that keeps this from being a simple heist story. Because the IP is only non-exclusively licensed, the shell isn't necessarily a corpse. Niemann admitted he's changed his mind: "My original take on this was this is just an attempt to avoid regulatory scrutiny… But I've started to see now some of the other hackquisition targets come back to life." The leftover piece "may actually have life to it… could actually be in some sense even more interesting than the part that goes over to NVIDIA."
The swing factor is exactly that: does the shell the founders leave behind still own something worth building on, or is it a "1x shell," enough to hand early investors their money back and nothing more? If non-exclusive licensing genuinely leaves a live business, the hackquisition is a clever way to fund teams that markets won't. If it doesn't, it's a structure that rewards founders and the biggest labs, gives regulators the slip, and leaves late-stage backers and rank-and-file employees with the scraps. Either way, "get bought for parts by a lab" is now a real line item in venture strategy, and that alone tells you how strange the exit market has become.
Signals
OpenAI hits the brakes; Anthropic floors it. The two giants split hard this week. Sam Altman confirmed OpenAI will not go public this year, telling Fortune that current safety concerns make it an ill-advised moment, a claim Scott Galloway flatly rejected on Pivot, "AI Panic: Dario's Warning, Trump's Dismissal, and OpenAI's IPO Delay" (September 15, 2026): "This is just an excuse. They can't go public." He replayed his own February prediction that the OpenAI IPO wouldn't happen, and argued "the momentum is really negative around this company." Meanwhile Anthropic is sprinting: chosen NASDAQ, targeting a listing that "could happen as soon as October." Galloway's read on why the founders are suddenly cautious in public while racing to list: "their job is to get their A, B, C, D, E, and F round investors who have invested hundreds of billions… a return. And a lot of these guys, the VCs and the employees have already started spending that money."
NVIDIA is now underwriting its own customer's IPO. On the Elon Musk Podcast, "Nvidia anchors Anthropic's $2 trillion IPO" (September 14, 2026), the hosts unpacked Anthropic seeking to raise up to $100 billion at roughly a $2 trillion valuation, with NVIDIA in talks to anchor with up to $10 billion. The problem they kept circling: NVIDIA is the chip supplier buying a giant stake in its biggest buyer. "It functions sort of like an automaker buying a massive stake in a rental car company… The primary goal is just moving metal." Or more pointedly: "NVIDIA is buying guaranteed future revenue and calling it an equity investment." They spelled out the loop that has the whole complex on edge (venture funds an app startup, the startup spends it on Anthropic's API, Anthropic spends that on NVIDIA GPUs, NVIDIA reinvests in Anthropic) and noted the tell: at a $2 trillion price, "they're pricing in perfection… If they miss earnings by a fraction of a percent once they're public, hundreds of billions of dollars in market cap will just evaporate in a single afternoon." (For scale, Anthropic's annualized revenue reportedly leapt to $65 billion by July, up sevenfold from $9 billion seven months earlier.)
The founder who owns almost nothing is about to control everything. The governance setup for Anthropic's listing is unusual on every axis, per Tech Brew Ride Home, "Anthropic Doing 2x OpenAI?" (August 19, 2026). CEO Dario Amodei "owns only roughly 2% of the company," one of the smallest founder-CEO stakes of any company going public in decades, because Anthropic has raised hundreds of billions in outside capital and diluted its seven founders down to a small minority. So Anthropic is preparing a supervoting share class to hand the founders control anyway (the SpaceX/Meta/Snap playbook: Zuckerberg controls over 50% of Meta's votes with under 15% of the shares). Layered on top: it stays a public benefit corporation (which would make it by far the largest ever on the U.S. market), and it keeps a long-term benefit trust (non-employee, non-investor trustees, including former Fed Chair Ben Bernanke) that can elect a majority of the board through special shares with no economic value. Public shareholders would be buying in three rungs down the control ladder.
The LP truth nobody in venture wants to hear: cash beats markups. The freshest VC-native conversation of the week was David Morehead, CIO of Baylor's endowment, on 20VC, "How LPs Allocate to Venture in 2026… Why Velocity of Cashback Is the Most Important Thing" (September 14, 2026). His office has a rule: "you're not allowed to talk about returns without also talking about time." The math he laid out is a quiet indictment of how venture funds are built. A 15-to-18-year fund that returns a "phenomenal" 15x still loses to laddering shorter growth-equity funds: 3x in six years, redeployed twice, compounds to 27x over the same 18 years, "better than 15X by like a factor of two." His line for why it matters: "Students can't pay their tuition with returns. They have to pay with dollars." He's blunt that the incentives don't line up: GPs hold winners longer because "it actually looks better on your marketing if you're up 6X instead of 3X" and helps them raise the next fund, "but I actually don't care about any of that." For Baylor, venture is now "a pure diversification play"; the real work is growth equity and buyout. (Notably, Baylor has about 2.5% of the endowment in Anthropic, through its managers, and zero in SpaceX or OpenAI.)
"The SpaceX slump" is the ghost at the Anthropic feast. On Rich Habits Podcast, "Trump's $5K Bribe, Anthropic's $2T IPO, & Meta's Muse" (September 11, 2026), the hosts pointed to the live cautionary tale: SpaceX IPO'd in a frenzy, peaked near $200 a share in mid-June, and three months later trades around $140–150, down roughly 30%. "Bloomberg flagged this directly. They said banks are explicitly hoping Anthropic can avoid the SpaceX slump… because their own earnings depend on it doing well." Their skepticism on the headline number was direct ("actual market cap could be closer to $1 to $1.25 trillion" once "real investors get in and set the price"), and their advice to retail was to wait and watch "what happens 60 to 90 days after lockups start expiring, which is exactly when SpaceX started to crack." (Underscoring the retail angle: the same show carried a paid pitch for a product selling everyday investors "the public ticker for private tech." Pre-IPO AI exposure is now being marketed straight to the crowd, right as the pros are warning them off the open.)
A contrarian "SaaS isn't dead" bet, made on human behavior. Buried in the Baylor conversation was a sharp counter to the software-is-finished narrative. When software stocks were "getting killed early in 2026" (down 50-60% from October 2025), Morehead went the other way, and his reasoning wasn't technical, it was behavioral. He called family-business owners he knows and asked if they'd rip out their CRM for something an in-law vibe-coded: "not in a million years." His frame: AI might be "93% right," but "software is 100% right, so if you need your books to match up, that's not going to happen." His prediction is that vertical SaaS becomes the delivery mechanism for AI (the trust built up with software vendors "is going to translate into, hey, could you add AI bits for me on the back of this software") and the incumbents aren't "going to sit there and let this go to zero." A useful reminder that the loudest venture narrative and the best trade often point in opposite directions.
Quote of the Week
"Students can't pay their tuition with returns. They have to pay with dollars."
David Morehead, CIO, Baylor University endowment, on 20VC (September 14, 2026), on why the speed of getting cash back matters more than a big paper multiple.
Runner-up, for coining the word of the week:
"Acquihires with an H… are about, at least ostensibly, avoiding antitrust scrutiny."
Alex Niemann, on Moonshots with Peter Diamandis (August 27, 2026), naming the "hackquisition."