alt Hacker NewsNvidia's $20B antitrust loophole
Comments
I wonder how the startup scene will adjust to this if it becomes mainstream. can employee contracts be modified to force compensation even in this case? seems difficult to write one up without weird second order effects.
if this does end up being something that is legal and successfully circumvents anti trust, does it mean antitrust actually is a failure in practice?
2026 hasn't even begun and more shenanigans are in flight.
Mr Kwok already analysed these deals and gave them a nice little acronym: HALO
Related:
Nvidia to buy assets from Groq for $20B cash - https://news.ycombinator.com/item?id=46379183 - Dec 2025 (400 comments)
Nvidia just paid $20B for a company that missed its revenue target by 75% - https://news.ycombinator.com/item?id=46403041 - Dec 2025 (133 comments)
IANAL, and am especially weak on US law, but I suspect this is only an antitrust loophole if the administration chooses not to act. Substance over form must apply? Pretty sure this wouldn't fly in European law.
So many companies doing non-"acquisitions" during this AI boom! Though this one is at least more comprehensive than say, Google simply hiring back Noam Shazeer from Character.AI or OpenAI taking Windsurf
What is keeping Google/Amazon/Microsoft from licensing Groq’s tech? Sans people to be sure but at substantially reduced price. The Groq Cloud people should owe no allegiance to Ross.
BTW, the FA says Nvidia bought the patents. That’s probably an overstatement. Grow said non-exclusive licensing and Nvidia hasn’t said anything.
I think Nvidia licensed the IP and ‘bought’ (handcuffed) the people.
> Groq built the region's largest inference cluster in eight days in December 2024. From that Dammam facility, GroqCloud serves "nearly four billion people regionally adjacent to the KSA." This isn't API access. This is critical AI infrastructure for a nation-state, funded by the Public Investment Fund, processing inference workloads at national scale.
Maybe I'm just completely out of touch, and hardware has never been my expertise, but does it take O(days) and not O(years) to build data centers these days? I know Grok DCs in Memphis were built under a year cutting many corners and using plenty loopholes, but even by those standards, bringing up a full data center in just over a week sounds impossible without some insane construction automation to me.
I actually managed to find video of Jonathan Ross discussing how GroqCloud was moving forward with Simon Edwards. It puts everything in proper perspective:
This behavior is extremely damaging to the startup scene. Who would join a startup these days unless it’s run by a close friend or relative? At least in that case, the scorned junior employees would have social recourse.
> GroqCloud will wind down over 12-18 months. They'll either get laid off or jump ship to wherever they can land. They built the LPU architecture, contributed to the compiler stack, supported the infrastructure, and got nothing while Chamath made $2B.
This is depressing.
I think it's important to note that there's nothing forbidding LPU style determinism from being used in training. They just didn't make that choice.
Also tenstorrent could be a viable challenger in this space. It seems to me that their NoC and their chips could be mostly deterministic as long as you don't start adding in branches
Read the article and where it talks about accelerated vesting of Groq shares for both the leadership team that goes to Nvidia and the regular employees that stay at Groq. Is that even guaranteed? It's not an IPO or an acquisition, so why would vesting schedules change?
> The "non-exclusive" label is legal fiction. When you acquire all the IP and hire everyone who knows how to use it, exclusivity doesn't matter.
I have some doubts about this point. IP is IP, independent of the people who invented it. If a different hardware company were to also pay for a non-exclusive IP license, maybe it will just take a few months to catch up. It’s like inheriting a codebase written by another team, and there will be some pain and some time needed to integrate it.
In fact if GroqCloud wishes to survive, it should very well just attract licensees for its IP and collect license fees for the foreseeable future.
A la carte in AI is going to be the name of the game for a couple reasons:
- Avoids regulatory scrutiny (for now at least)
- Nobody is actually entrenched enough for customers to matter
- Weird "celebrity" culture in tech, and AI especially. Everyone is looking for a "whisperer" or a "godfather" or whatever.
- Investors still get paid out
Smart operational talent will probably adapt by demanding higher salary, signing bonuses, severance packages in lieu of equity. Distribution of the true "lottery tickets" will get more uneven.
So... What will the actual impact on groq services be?
I'm a fan, and I use Groq a lot for systems I build. I think they offer something different to most other providers (cheaper, faster, and until recently "we don't store your data by default") and it will be sad to see that fade.
I'd like to call back to yesterday's discussion on IP law [1] sparked by recent comments from Rob Pike.
There was a major thread on the issue of regulatory regimes and the dysfunction that can arise. How is this acquisition not a textbook example of said dysfunction? This non-acquisition acquisition does not happen at all in a world without IP law.
I think we're seeing a culmination of the dysfunction that results from IP law. The sheer amount of capital has given unbelievable momentum to the forces of consolidation. I still can't foresee the endgame (who can?) but it's even harder to see how it'll turn out well.
The analysis here is excellent. However, Groq's success was not obvious or straight forward. It require huge amounts of investment to keep it alive for many years long before any real positive cashflow. It was on its death bed for years before the ChatGPT moment in 2023. It has been over 9 years since its founding. If you know anything about VC funds and exit timelines, you will know that investors in Groq needed an exit by this time.
I'm happy that the investors and founders got a long and well deserved exit. I'm happy that the tech here will continue to see development and investment under Nvidia so we may one day get to use Claude Opus at 500 tokens per second.
Does it suck that certain employees got screwed over? Yes. Does this happen ALL THE TIME in startups? More often than you think. The expected value for employee options for this type of company is very very close to zero. Anyone who thinks otherwise is lying to themselves.
Does it suck that it didn't happen via a normal M&A process? As someone who used to work on tech M&As as an attorney, I would be first one to say that I hope this DOES become the norm. M&A sucks for the employees, the investors, the founders, the acquirers - it sucks for EVERYONE. The only people who it doesn't suck for are the lawyers and bankers who earn more fees the more complex and longer the process is. Best M&A I ever witnessed was the FB acquisition of Instagram that happened over the weekend (my old law firm was part of that deal).
Ask yourself: do you want to spend 2% of your funding round and 2 months on lawyers when you raise a $5m Series A? Then why do you want to do the same when you exit?
This has nothing to do with antitrust. Not like the current administration is going to enforce it anyways. Nvidia simply wants Groq’s tech and leadership without the burden of 500+ employees.
I wonder how much of the cap table knew
Matthew Berman (youtuber) mentioned he's invested in groq and found out same time as everyone else. Guessing he's a small/indirect investor but still telling
Also worth thinking a about the private equity market scene, groq was afaik tradable be it thinn liquidity on platforms like equityzen. What did those shareholders get?
Feels like a frenzy! To me, the willingness to spend 20bn for a different architecture speaks to the competition in the market, across the big players, which doesn’t seem to be factored into their valuations.
Nvidia bought groq? I am out of the loop but what does groq have that they want?
So I guess this talent hire + tech license is the new way to acquire startups? Chatacter.ai, Windsurf, and now Groq.
Any employees of those companies lurking here? I'm curious how the morale is now.
Great read. Thanks.
No shade but most other coverage will focus on whether this signals an AI bubble. That's missing the story.
Nvidia explicitly did NOT acquire Groq. They licensed the IP and hired the talent. This structure dodges CFIUS review (Groq had $1.5B in Saudi government contracts), antitrust scrutiny, and years of regulatory delays.
The $13B premium over the September valuation was the cost of regulatory arbitrage. Announced Christmas Eve while Trump's AI Czar (Chamath's All-In podcast co-host) is in office. Chamath's Social Capital made AT LEAST ~$2B on this exit.
My article breaks down: what Nvidia actually bought vs what they left behind, why the deal structure matters, who got paid, and the political connections nobody's talking about.
It's only a "loophole" if you believe psychopaths like Linda Khan should also be able to dictate whether companies can hire particular employees or purchase products.
>Timing it for Christmas Eve ensures minimal media scrutiny of these connections.
Sounds like the media is truly the one in charge.
I'm going to assume this is a predominantly AI-written article since for some reason it's talking about GroqCloud serving Llama 2, which they don't.
It claims they serve Llama 2 7B @ 750 tokens/s with 2K context, but over on OpenRouter Groq is listed as serving Llama 3.1 8B @ 1300 tokens/s with 128K context. (And the official GroqCloud site says 840 tokens/s.)
>And 10x better energy efficiency because you're not constantly moving data across a memory bus.
Moving data is not that expensive, many times there's no overhead over the nominal consumption of your system.
The rest of the article is spot on, though.
nVidia's big problem right now is they have more money than they can productively spend and are way down the stupid money hole just looking for any kind of return. This is a failure of capitalism.
The part I don’t fully understand is: could this non-acquisition eventually make the deal less than ideal for Nvidia?
Is it really a given that GroqCloud is going to be sunset and the company will die?
Couldn’t this company hire talent and continue to operate and maybe even innovate? Couldn’t Groq even hire back some employees from Nvidia? If any of them live in California there’s nothing stopping them and they have a bunch of cash from Nvidia. There are all kinds of loopholes for that like contracting arrangements.
Nvidia doesn’t really have exclusive access to any part of the company. They didn’t necessarily remove a competitor, though I’ll grant that they likely did in practice.
It’s potentially possible that the regulations did their job and kept a competitor on the market, though again I imagine this is my naevity speaking and that the most likely outcome is that Groq will wither.
I also don’t fully understand if the Saudis are getting cashed out or not. Are they really going to roll over and allow their Saudi AI data center to become worthless? I would think they have a lot of motivation after this deal to make sure Groq still operates and serves their goals.
this entire article is seemingly written by an LLM
That's why we can't have nice things.
>"The question isn't just why Nvidia paid $13.1B more than market rate for technology they could build themselves (they have the PDK, volume, talent, infrastructure, and cash). The question is why they structured it this way.
Where the premium was spent:
Regulatory arbitrage:
Non-exclusive licensing avoids years of antitrust review. Structure the deal as IP licensing + talent acquisition, and regulators have no grounds to block it.
This alone is worth billions in time and certainty."
Isn't that fascinating!
Observation: For any given Deal (in business or in life in general) -- there may be one or more legal components -- to it...
But (equal-and-oppositely!) there also may NOT be one or more legal components to it!
What each deal has in some legal components -- it may lack in other legal components...
Conversely, what each deal does not have in some legal components -- it may have in other legal components...
Now, perhaps this may sound like a "self-evident truth", and as such, apparently may not be worthy of a deeper exploration, but it seems that there exists an:
Intersection of Set Theory and Legal Aspects -- applied to Deals
(AKA "Transactions", "Exchanges", "Barters", "Trades", "Exchanges Of Value", etc., etc.) in various jurisdictions (which can be thought about as how contracts, both legal and natural, arising from such exchanges are, or would be interpreted through its courts, through its regional statutes (aka "Laws") IF there are inter-party disputes which subsequently require a court for such interpretation...)
And that intersection -- could well be worthy of further study!
Phrased another way -- it (and this article!) are highly interesting from a legal perspective!
(And also a Set Theory / Set Theoretical one!)
> Most production AI applications aren't running 405B models. They're running 7B-70B models that need low latency and high throughput.
Really? At least for LLMs, most actual usage is concentrated on huge SOTA models. 1 trillion parameters or more. And LLMs seem to be the lion's share of AI compute demand.
[dead]
[flagged]
[dead]
[dead]
The chatgpt-ism and the gpt section headlines make this piece immediately unreadable. Why do you outsource your own blog’s thoughts to a machine? Terrible.
I don't think you can treat owners of the same shares differently in the way this is suggesting. The VC shareholders and the employee shareholders are probably on equal footing and getting the same price. VCs will own preferred but I doubt that is enough to windfall them at the expense of the common shareholders.
So if VCs are getting paid a certain share price, employees with vested stock almost certainly are getting the same price. And probably employees with vested options can either exercise now or will just get paid the net during the transaction.
Yes, the company is probably doomed so people staying there are not doing well, but they also just got paid a 3x premium on their vested equity.