Post actualizado el día octubre 26, 2026 by DeiviSanzPlay
Meta just flipped the AI chip game on its head. In a move that caught almost everyone off guard, the company announced it will sell AI compute power — basically renting out its massive GPU clusters — while Anthropic goes all-in on designing its own silicon. And Samsung? Samsung is sitting there rubbing its hands together like a cartoon villain, because this whole mess lands right in its lap.
Meta’s Pivot: From AI Spender to AI Seller
Let’s be clear about what happened here. For years, Meta was the kid at the candy store, buying up every Nvidia GPU it could get its hands on to train Llama models. Now it’s turning around and selling that same compute to other companies. That’s a massive shift in the AI infrastructure market, and it’s not just a side hustle — it’s a full-blown strategy change.
The ripple effects were immediate. CoreWeave, the cloud provider that built its entire business around renting Nvidia GPUs to AI companies, saw its stock drop 14% in a single day. Nebius, another AI cloud player, fell 17%. Why? Because Meta was CoreWeave’s biggest customer. When your anchor tenant suddenly becomes your competitor, investors don’t stick around for the fireworks.
Meta had a $21 billion agreement in place. That’s not pocket change. That’s a commitment that made CoreWeave’s entire financial model work. Now Meta is saying: why rent out our spare capacity to you when we can sell it ourselves at better margins?
The AI compute power market just got a new heavyweight player, and it’s one that already has the hardware, the data centers, and the engineering talent to compete with anyone.
Anthropic’s Chip Ambitions: No More Nvidia Dependency
Meanwhile, Anthropic — the company behind Claude — is making its own chips. This isn’t some vague rumor. They’ve been quietly building a semiconductor team, and the industry is taking notice. The days of every AI company bowing at the altar of Nvidia are numbered.
Think about the logic. If you’re Anthropic and you’re spending billions on Nvidia H100s and B200s, and you’re competing directly with OpenAI, Google, and Meta — all of whom are building custom silicon — you can’t just keep writing checks to Jensen Huang forever. Custom AI chips aren’t a luxury anymore. They’re a survival requirement.
The person leading this charge? Someone who worked on OpenAI’s chip team, no less. The same engineers who designed the chips that keep climbing in performance and efficiency are now scattered across the industry, and Anthropic got its hands on some serious talent.
This isn’t about saving a few bucks on hardware. It’s about controlling your own destiny in an industry where compute is the bottleneck for everything — training runs, inference costs, model scaling. If you don’t control your chips, you don’t control your roadmap.
Why Samsung Is the Real Winner Here
Here’s where it gets interesting. Anthropic needs a foundry to manufacture these chips. TSMC is the obvious choice — they’re the gold standard for advanced semiconductor manufacturing. But there’s a problem: TSMC’s 2nm production is sold out until 2027.
Apple has locked up more than half of that capacity for its M-series and A-series chips. Nvidia, AMD, and Broadcom have scooped up the rest. If you’re a new player trying to get 2nm chips made, you’re basically standing outside TSMC’s door with a sign that says «will work for wafers.»
Enter Samsung. The Korean giant has been investing heavily in its 2nm process, and unlike TSMC, it has available capacity. That’s a massive opportunity. Samsung’s foundry business has been struggling to attract big-name customers away from TSMC, but now the market dynamics are shifting in its favor.
When the biggest AI companies in the world can’t get their chips made by TSMC, they have to go somewhere. Samsung is that somewhere. And Samsung knows it.
The New AI Chip Landscape: Everyone Wants Their Own Silicon
The pattern is unmistakable. Google has TPUs. Amazon has Trainium and Inferentia. Microsoft has Maia. Meta has MTIA. Apple has been designing its own chips for years. Now Anthropic is joining the club, and OpenAI has been rumored to be exploring custom silicon too.
What does this mean for Nvidia? Short term, nothing catastrophic. Nvidia still controls the vast majority of the AI chip market, and demand for its GPUs continues to outstrip supply. But the long-term trend is clear: the biggest AI companies are slowly reducing their dependence on Nvidia.
That doesn’t mean Nvidia is going anywhere. It means the market is maturing. When you’re a startup, you buy off-the-shelf. When you’re a hyperscaler spending $10 billion a year on compute, you build your own. That’s just economics.
The companies that design their own AI chips gain three advantages: lower cost per operation, better performance for their specific workloads, and insulation from supply chain shocks. Those advantages compound over time.
What This Means for the AI Industry in 2026
Three big things are happening simultaneously, and they’re all connected.
First, the compute market is fragmenting. Meta selling compute means the cloud GPU rental business just got a lot more competitive. CoreWeave and Nebius aren’t going to disappear overnight, but their pricing power is evaporating.
Second, custom silicon is becoming table stakes. If you’re a serious AI lab and you don’t have a chip team, you’re already behind. Anthropic figured this out. Others will follow.
Third, Samsung is positioned to become the alternative foundry for AI chips. TSMC’s capacity constraints have created a window of opportunity, and Samsung is the only company with the scale and technology to exploit it.
The AI power struggle isn’t just about models anymore. It’s about who controls the compute, who designs the chips, and who can actually manufacture them. Meta, Anthropic, and Samsung are all making moves that reshape the board.
The Bottom Line
Meta selling AI power is a signal that the company sees more value in being an infrastructure provider than just a model builder. Anthropic designing its own chips is a signal that AI labs are tired of being dependent on Nvidia. And Samsung’s position as the foundry of last resort is a signal that TSMC’s dominance isn’t absolute.
The companies that understand these shifts — and position themselves accordingly — will be the ones still standing when the music stops. Everyone else is just watching the stock ticker and wondering what happened.
