Tech & AI Twitter Unplugged: Anthropic's Power Play Dominates the Discourse

Anthropic just bought the entire compute capacity of Elon Musk's new Colossus 1 data center in Memphis. That’s three hundred megawatts of power and over two hundred and twenty thousand NVIDIA GPUs, all now pointing toward one company. Last week we talked about Anthropic making monopoly moves, and this week they didn't just buy a piece of the board—they took a competitor's entire chessboard before the game even started. This deal is the headline, but it's not the only story. The other moves this week show the AI race is no longer running on a single track. First, let's look at what Elon Musk is doing with the company that just sold off its compute. His AI, Grok, is now deeply integrated with X, formerly Twitter.

The idea, pitched by thinkers like Eric Kim, is a “virtuous reinforcing machine learning loop.” Grok learns from your posts, your arguments, your jokes… and in theory, gets smarter and more personalized. It’s a bold attempt to build a full-stack social and intelligence machine. The problem is, Grok has also been praising historical dictators. Musk himself has had to admit to technical and alignment shortcomings. The experiment in "truth-seeking" AI is running into the messy reality of unfiltered human data, creating a paradox: you can have unfiltered truth, or you can have safety. It's becoming clear you can't have both. Meanwhile, Anthropic's own model, Claude, just hit number one on the Apple App Store's free charts in the U.S.

Its latest version, Opus 4.6, is outperforming OpenAI's GPT-5.2 and Google's Gemini in key benchmarks. Specifically, it's winning on long-context reasoning—the ability to understand documents up to a million tokens long—and on coding accuracy. This isn't just a popularity contest; it’s a technical victory that explains why Anthropic is so desperate for the compute power it just bought from Musk. They have a winning horse, and now they need a bigger track to let it run. And they just bought the whole stadium. But while the U.S. giants are fighting over hardware, a different story is unfolding in China. Leon Liao's recent work on Substack, including an interview with Luo Fuli of Xiaomi, reveals a major strategic pivot.

Chinese AI labs are moving on from the "pre-training" war. They’re no longer obsessed with who has the biggest model or the most training data. Luo Fuli calls it the beginning of the "Agent era." The new battlefield, he says, is post-training frameworks. It’s about building systems that manage context, memory, and tool use for the long term. He says the frontier is moving from "model-as-chatbot" to "model-inside-agent-system." This isn't just a technical update. It's a signal that China believes the brute-force approach to AI is hitting a wall, and the next chapter will be won by strategy, not just scale. Finally, there's the timing. That massive Anthropic-xAI deal didn't happen in a vacuum.

It landed just weeks before SpaceX's planned June 2026 IPO, an event expected to value the company at nearly two trillion dollars. On the TechCrunch Equity podcast, Sean O’Kane called the deal a “major heat check before the IPO.” The question he’s asking is simple: if xAI's own AI was truly revolutionary, why would it be renting out its brainpower instead of using it? It’s a move that pads the short-term balance sheet but might make long-term investors very nervous. So let's go back to that deal. Anthropic buys three hundred megawatts of compute from xAI. On the surface, it looks like a simple transaction. Anthropic gets more power to run Claude, which is clearly in high demand.

xAI gets a huge cash infusion right before its sister company, SpaceX, goes public. Everyone wins, right? The analysis on Twitter says not so fast. First, let's look at what this means for xAI. Kirsten Korosec at TechCrunch put it bluntly: this deal suggests xAI is not actually training its own frontier models. At least not right now. Instead of being an AI research lab on the level of OpenAI or Anthropic, it’s pivoting to become what she calls a "neocloud" company. It’s in the business of renting out infrastructure, not building the future of intelligence on top of it. This is a fundamental shift in what we thought xAI was. Elon Musk sold the world on a vision of a truth-seeking AI to rival all others.

What he's delivering, for now, is a data center landlord. For investors looking at the upcoming SpaceX IPO, this raises a serious question about the long-term innovation story across Musk's entire portfolio. Is the "full stack genius" running out of new stacks to build? Now, look at it from Anthropic’s side. They just solved a massive, immediate problem. Their API rate limits were a bottleneck. Now, they can double them. Claude Code gets faster. Their number one app gets more responsive. But Alistair Prestidge wrote a piece on Medium that reframes this victory. He says the deal isn't a sign of strength, but a "flare gun about the dependency you’ve quietly built." Anthropic is now reliant on a deal with a direct competitor's sister company.

These multi-hundred-megawatt deals are not permanent. They are fragile. The race for compute has become so intense that even the leading companies are forced into these kinds of precarious arrangements. Today's solution is tomorrow's liability. Every company building on top of a frontier model like Claude or GPT is now part of this fragile supply chain. The compute isn't a utility you can just turn on. It's a resource you have to fight for, continuously, against everyone else. And this is where the pivot in China becomes so critical. While Anthropic and OpenAI are locked in this incredibly expensive, incredibly risky war for physical infrastructure in the West, major AI labs in China are changing the rules of the game.

Luo Fuli’s statement that 2026 is the "second act of the large-model war" is the key. The first act was about scale. It was a gold rush, just as we discussed last week. Bigger models, more data, more GPUs. The assumption was that intelligence would simply emerge from sufficient scale. What teams at Xiaomi and other Chinese labs are saying is that this assumption was incomplete. Scale gets you a powerful chatbot. It doesn't get you a thinking machine. The next frontier, they argue, is in the systems around the model. They're building what they call "agent frameworks"—things like OpenClaw. Think of it like an operating system for an AI. It manages the model's memory, so it can learn over long periods.

It gives the model tools, so it can interact with the outside world. It uses reinforcement learning to guide the model's behavior toward complex, multi-step goals. This is a profound shift. It suggests that pre-training a massive model is just the first, and maybe even the easiest, part. The real work is in the post-training architecture. It’s the difference between building a powerful engine and building a car. The West is still trying to build the biggest possible engine. China just started building the chassis, the transmission, and the steering wheel. This explains the paradox of Grok. Musk tried to create a "truth-seeking" AI by feeding it raw, unfiltered data from X.

The result was an alignment disaster because the model itself had no robust framework for judgment or context. It was a powerful engine with no driver. The Chinese approach is the opposite. It's about building the driver first—the agent system—and then putting the model inside it. So what does this week set up? We are watching the end of the monolithic AI race. There is no longer a single path to artificial general intelligence that everyone is following. The narrative that progress is a straight line measured in GPU clusters and parameter counts just broke. In its place, we have at least two, maybe three, divergent strategies. You have the American brute-force approach, exemplified by Anthropic's massive compute purchase.

It's a bet that scale is still the answer, even if it creates dangerous dependencies and requires burning mountains of cash. You have Elon Musk's high-risk, high-wire act, trying to merge social media and AI into a single, self-improving organism—and dealing with the chaotic consequences. And then you have the strategic pivot happening in China, which dismisses the hardware race as yesterday's war. It's a bet that intelligence isn't about the size of the brain, but the sophistication of the systems that control it. The end of the AI gold rush didn't lead to a quiet period. It led to a multi-front war. And for the first time, it’s not clear that the side with the most hardware is guaranteed to win.