We’ve all been watching the AI backlash build for months now. People are worried about job losses, deepfakes, misinformation, creative theft. Valid concerns, all of them. But while everyone’s arguing about whether ChatGPT is coming for their career, something arguably worse is happening in plain sight: the companies building AI are also rebuilding the rules of American capitalism to ensure nobody can ever compete with them.
Nvidia just showed everyone exactly how it works.
On Christmas Eve, the world’s most valuable company paid $20 billion for a chip startup called Groq. Not to be confused with Musk’s AI, Grok. Two different things – anyway. Nvidia took Groq’s CEO, they took the president, they took around 90% of the staff, and they took the intellectual property. They took the technology that Groq spent nine years developing.
But don’t call it an acquisition. Nvidia insists it’s just a “non-exclusive licensing agreement.”
The announcement was 90 words long. There was no press release. No public regulatory filing. No public acknowledgment of the price tag – that came from an investor who talked to CNBC. They closed a deal worth three times Groq’s September valuation during a holiday news blackout, and the stock barely moved. One analyst noted that Nvidia is now so powerful it can execute a $20 billion transaction on Christmas Eve and nobody bats an eye.
Here’s why this matters beyond the usual corporate consolidation story.
Groq wasn’t just any startup. The founder, Jonathan Ross, is one of the guys who created Google’s TPU – the tensor processing unit, Google’s custom AI chip. He left Google in 2016 specifically to build chips that could compete with Nvidia’s graphics processors. Because that’s what they are – graphics processors originally made to run video games.
His company developed language processing units that it pitched as a faster, cheaper way to run AI than anything Nvidia sells. Groq says they’re 10 times faster and use 10 times less power than Nvidia’s graphics processors. At first that seems like a really great leap in technology – but Ross said something in an interview that led me to believe that in the end these chips won’t really save much money or power. These chips are made for LLMs, so the goal is to just free up Nvidia chips to focus on training.
So, Nvidia’s response was simple: if you can’t beat the competition, structure a deal so creative that regulators can’t call it an acquisition.
The Hart-Scott-Rodino Act requires companies to report any acquisition valued above about $126 million to federal antitrust agencies. A $20 billion purchase of a direct competitor by a company that already dominates 80% of the AI chip market would absolutely trigger scrutiny. Nvidia learned this the hard way when their 2020 attempt to buy Arm Holdings collapsed in 2022 after regulators in the U.S., U.K., and EU raised objections.
So they found another way.
Instead of buying Groq, Nvidia licensed its technology. Instead of acquiring the workforce, they hired the employees individually. Instead of buying the company’s stock, they structured payments so investors get their returns by the end of 2026. Groq technically continues to exist – the CFO got promoted to CEO, and their small cloud business keeps running. On paper, competition survives.
In reality, the founder is gone. The brain trust is gone. The technology now belongs to Nvidia’s roadmap. Whatever Groq might have built to challenge Nvidia’s dominance will now be built inside Nvidia, for Nvidia’s benefit.
One Bernstein analyst wrote that structuring it as a license “may keep the fiction of competition alive.”
This isn’t new. It’s becoming the standard playbook for every major tech company trying to lock down the AI market.
In March 2024, Microsoft paid $650 million to “license” technology from Inflection AI while hiring nearly the entire 70-person team, including co-founder Mustafa Suleyman, who now runs Microsoft’s AI division. The FTC opened an investigation asking whether Microsoft deliberately structured the deal to avoid antitrust review. That investigation is still ongoing. The deal closed immediately.
Amazon did the same thing with Adept AI – hired two-thirds of the staff, paid a licensing fee, left the corporate shell behind. Google paid $2.7 billion to Character AI for a “non-exclusive license” while hiring the co-founders who built the technology. Same pattern every time: take the people, take the tech, leave the husk, call it a partnership.
The German Federal Cartel Office looked at Microsoft’s Inflection deal and concluded it functionally constituted an acquisition, but declined jurisdiction because Inflection didn’t have enough business in Germany. The FTC has been investigating these structures, but investigations take years and the Trump administration seems unlikely to prioritize enforcement. By the time anyone decides whether these deals violate antitrust law, the consolidation will be irreversible.
What makes Nvidia’s move particularly brazen is the scale. This isn’t a $650 million talent grab – it’s $20 billion, roughly triple what Groq was worth three months earlier. Nvidia is sitting on $60 billion in cash and short-term investments, up from $13 billion in early 2023. They can afford to overpay dramatically for the privilege of eliminating competitors without technically eliminating them.
The inference chip market was supposed to be where Nvidia’s dominance could finally be challenged. Training AI models requires massive parallel processing that Nvidia’s GPUs handle well, but running those trained models – inference – has different requirements. Groq’s chips were specifically optimized for that work. They were faster for certain applications. They used less energy. They cost less.
Now that technology gets folded into Nvidia’s product stack. The engineers who built it will spend their careers advancing Nvidia’s interests. The startup that might have given enterprises an alternative to Nvidia’s pricing power is functionally gone, even if a legal entity bearing its name technically survives.
I keep hearing people worry about AI itself – what it might do, what it might become, whether it will replace human creativity or judgment. Those are real conversations worth having. But the more immediate threat isn’t the technology. It’s the handful of companies consolidating control over that technology so thoroughly that by the time we figure out what AI should be used for, we’ll have no choice about who provides it.
Nvidia didn’t just buy a competitor. They demonstrated that the rules meant to prevent monopolies don’t apply if you’re creative enough with the paperwork. And they did it on Christmas Eve, betting correctly that nobody would care enough to stop them.
If you’re wondering when someone in power will notice, I wouldn’t hold your breath.
#ratcclips