AI Power Plays: OpenAI’s Cerebras Deal Shakes Up the Hardware Race

OpenAI just entered a twenty-billion-dollar deal with chip startup Cerebras for its AI servers. This isn't just about buying more computing power — it’s a signal that the race for AI dominance has moved from developing software to controlling the physical hardware that runs it. The deal, reported by The Information, spans three years and could give OpenAI an equity stake in Cerebras, tightening its grip on a critical part of the supply chain. This move away from total reliance on Nvidia shows the scale of the game has changed. We're no longer talking about building better models.

We're talking about building the industrial base for an entirely new kind of economy. And that industrial base requires two things: immense power, and immense data. Which brings us to the second, and much darker, conversation happening right now. The other side of AI's massive appetite is the cost it's imposing on the open web. New analysis this week puts a hard number on the damage. For every one visitor the AI company Anthropic sends to a publisher's website... it scrapes that same site 8,692 times. That isn't a typo. Eight thousand, six hundred and ninety-two to one.

OpenAI’s ratio is a little better, at 179 scrapes for every one referral. The CEO of Conductor put it even more starkly: ChatGPT crawls a website sixty thousand times to send back just one visitor. This is the core of the crisis facing digital media. The traffic spigot that powered the internet for two decades is being turned off. AI chatbots are becoming the destination, not the gateway, and they are consuming the web's content to fuel their answers without meaningful reciprocation. Research from Stanford's Graduate School of Business confirms the trend: click-through rates from AI chatbots are a dismal zero-point-three-three percent, a fraction of the eight-point-six percent from traditional Google Search.

This isn't a disruption. It's an extinction event for ad-based media models. And while that economic model is breaking, a third conversation is emerging from a different corner of the internet, proposing a new one. Ethereum co-founder Vitalik Buterin has re-emphasized his position from earlier this year: the goal of the blockchain isn't to compete in the race to build Artificial General Intelligence. The goal is to build the trust layer for it. To become the operating system for autonomous AI agents. This isn't just a philosophical idea anymore. This week, we're seeing the first concrete protocols emerge to make it real.

Standards like ERC-8004 are being developed to give AI agents their own unique, on-chain identities, like a digital passport. Another protocol, x402, is being built under the Linux Foundation to give those agents a way to pay for services, like APIs or data, using stablecoins. The conversation on Twitter is coalescing around this idea: if AI agents are going to operate autonomously in the world, they need identity, reputation, and a way to transact. Blockchain is positioning itself as the only technology that can provide all three in a decentralized, verifiable way.

So you have three parallel stories. An unprecedented hardware deal to fuel AI's growth. An economic crisis caused by that AI's consumption of data. And the birth of a new set of rules to try and govern it. At a glance, they seem separate. A hardware story, a media story, and a crypto story. But they aren't separate at all. They are one story, told in three acts. The story of what happens when a new form of intelligence begins to exert its will on the physical and economic world, and how we are scrambling to build the systems to manage it. Let’s go back to that OpenAI deal.

Twenty billion dollars. For servers. From a company that isn't Nvidia. This is the most important part of the story. For the last few years, the entire AI boom has been built on the back of Nvidia's GPUs. They became the single point of failure, the one choke point that could slow everything down. Owning their chips became a measure of a company's—or a country's—seriousness about AI. OpenAI's deal with Cerebras is a declaration of independence. It's a move to diversify its hardware supply chain, to ensure that its path to more powerful models cannot be throttled by a single supplier.

This is what a maturing industry looks like. It starts to worry about logistics. About physical infrastructure. About geopolitical risk. The conversation is no longer just about the elegance of the algorithm; it's about securing the silicon. Cerebras builds massive, wafer-scale chips designed specifically for AI workloads. By committing twenty billion dollars, OpenAI isn't just buying capacity. It's funding a competitor to the market leader, ensuring a more resilient and competitive hardware ecosystem for itself in the long run. It’s a move straight out of the playbook of a 20th-century industrial giant, not a 21st-century software company.

They are securing the means of production. Because what they are producing is intelligence itself, and the demand for it appears to be infinite. And that infinite demand requires infinite fuel. Which is where the media crisis comes in. The data from TollBit's Q1 report, showing those scrape-to-referral ratios, is a bombshell. For years, publishers had a simple deal with search engines. You crawl our content, and in return, you send us traffic, which we monetize with ads. It was a symbiotic relationship. That symbiosis is now over. Large Language Models are not search engines.

They don't want to send you to a link. They want to give you the answer. To do that, they ingest, process, and synthesize the content of the entire open web. The 8,692-to-1 ratio for Anthropic is the most extreme example, but it reveals the underlying logic. The AI's goal is to absorb information and create a new product from it, a product that lives inside its own walled garden. The referral is a byproduct, a failure case. It happens only when the AI cannot confidently synthesize an answer. So the better the AI gets, the less traffic it sends. The value exchange isn't just imbalanced; it's fundamentally broken.

The zero-click search, which was already a problem in the Google era, has become the default interaction. Stanford's data showing zero-click searches rising from fifty-six percent to sixty-nine percent in just the last year tells the story. The web is becoming a read-only resource for machines. This is not a technical problem with a technical solution. This is an economic problem. The content that fuels the most advanced technology on the planet is being created by an industry that is being systematically defunded by that very same technology. Publishers are providing the raw material for their own replacement, and they are getting almost nothing in return.

The conversation on Twitter isn't just about anger. It's about a search for a new model. How do you create a system where value is exchanged fairly between the creators of information and the consumers of it, when the consumer is a machine? This is the question that makes the third conversation—the one about Ethereum and AI—so critical. It lands with the force of a revelation. Vitalik Buterin's argument is that trying to build a "decentralized AI" to compete with OpenAI is a mistake. It misses the point. The real opportunity is to use the blockchain's unique properties to solve the problems that centralized AI creates.

Problems of trust, verification, and economic exchange. He calls it becoming the "trust layer" for the AI world. And now, we see the code that backs up the concept. Let's look at these new protocols. ERC-8004 is for AI agent identity. Think of it like this: right now, when OpenAI's GPTBot crawls your website, it's anonymous and unaccountable. It's just a line in a server log. With an ERC-8004 standard, an AI agent could have a persistent, on-chain identity—an NFT that functions as its passport. You could see its history. You could see its reputation. You could decide whether to grant it access based on its past behavior.

Suddenly, the anonymous scraper becomes a known entity. Then you have x402, the payment protocol. This is the direct answer to the broken media model. An AI agent with a crypto wallet can be programmed to pay for the data it consumes. Imagine that agent arriving at a publisher's website. Instead of just scraping everything, it negotiates a transaction. It pays a fraction of a cent for the right to ingest an article. The publisher now has a new revenue stream, one that is native to the machine economy. The value exchange is restored. This isn't science fiction. The protocols are being built on Ethereum's mainnet right now.

Other chains like Solana are working on low-latency applications for AI agents. This is a multi-chain effort to build the financial and social plumbing for a world populated by autonomous AIs. The pieces click together. OpenAI builds a twenty-billion-dollar industrial base to scale its intelligence. That intelligence consumes the open web, breaking its economic model. And the crypto world responds not by trying to build a bigger AI, but by building the rules of the road. An identity layer to make agents accountable. A payment layer to make them economic citizens.

And other proposals for things like verifiable execution, using zero-knowledge proofs to confirm that an AI actually ran the computation it claimed it did. What's emerging is a stack for trusted AI interaction. It treats AI not as a piece of software, but as a new kind of participant in the world. A participant that needs what all other participants need: identity, reputation, and a way to transact. This week's conversations reveal two parallel races unfolding. The first is the one everyone sees: the race for AI supremacy, measured in parameters, compute, and capital.

It's a race of scale, and the twenty-billion-dollar deals are just the beginning. But the second race is the one happening in the threads and forums of the decentralized web. It's the race to build the social and economic protocols for that AI. It's a race measured not in scale, but in trust. The hardware deals and the media crisis show us the shape of the problem. AI is an agent with immense power and an insatiable appetite, but it currently lacks accountability and a native way to participate in a fair economy. The blockchain protocols are the first serious attempt to provide those missing pieces.

They suggest a future where AI doesn't just take... it participates. Where every interaction, every piece of data consumed, can be part of a transparent and equitable exchange. The race isn't just to build artificial intelligence anymore; it's to build an artificial economy, and the blueprints are being drawn up on-chain, right now.