AI is getting cheaper. Somebody's paying the bill.

I should be studying right now. My CodePath interview prep assessment is due tomorrow night, and I still have problems to grind through.

Instead I fell down a news rabbit hole. This one, though. I couldn't close the tab.

On Thursday, Anthropic signed a deal to spend $11.6 billion over seven years on Akamai's cloud infrastructure. That's the biggest contract in Akamai's history — the company that used to just deliver content on the internet. Now they're building AI compute for one of the biggest AI labs on the planet (Reuters, TechCrunch).

The detail that stopped me: the capacity Anthropic bought isn't GPUs. It's CPUs. Regular general-purpose processors. The kind of chips that run code and browse the web.

Because that's what AI agents do. When an agent researches something, runs your code, or clicks through a website, that's not a GPU thinking. That's a CPU doing a job. TechCrunch put it plainly: demand for CPUs has grown as AI agents take on more tasks.

And then there's the part of the deal I had to read twice. Akamai gave Anthropic a warrant worth up to 5% of its stock — about 7.7 million shares at $111.33 each. Roughly 2% unlocks with the initial $11.6 billion commitment, and every extra $3 billion Anthropic spends unlocks about another 1%. The total could grow to $20 billion. Akamai's stock jumped over 20% in after-hours trading on the news.

Think about that for a second. The supplier paid the customer in shares of itself. Committed demand is now so valuable it's worth equity.

To build it all, Akamai expects to spend about $5.5 billion, and they're raising this year's capital spending by another $1.7 billion just to buy components and memory in advance. They won't see a dollar of revenue from the deal this year. They estimate $150 to $300 million in 2027, growing to an annual pace of about $1.7 billion by the end of 2028. This is a bet that takes years to pay off.

Meanwhile, two days earlier, the opposite headline. OpenAI dropped GPT-6 Sol and GPT-6 Luna at half the price of their predecessors — Sol at $2 per million input tokens and $10 output, Luna at $0.10 and $0.50. The same day, Anthropic shipped Claude Opus 5.5, about 20% cheaper on the price tag and roughly 40% cheaper on a typical workload (explainx.ai, Particle).

Same week. Two stories pulling in opposite directions. The models you call are getting cheaper. The bill to run them at scale is getting enormous.

That's the part I keep coming back to. I'm going to graduate into an industry where intelligence is getting close to free on the price page, and the real business is whoever can build, power, and cool the machines that deliver it. Anthropic is stacking these deals — $45 billion with Nscale last month, $11.6 billion with Akamai this week. These aren't research expenses. They're rent.

And honestly, as someone applying to IT roles, it rewires what "AI jobs" means. Not everyone is training models. Somebody has to build the data centers, buy the memory, do the capacity planning, watch the bills. The agent future runs on CPUs, not just GPUs, and somebody has to keep those racks alive.

I'll admit I don't fully understand the finance — warrants tied to spend commitments, vesting schedules. I'm learning. But I understand this: when the customer is this hungry, they don't negotiate a discount. They take a piece of the store.

Now back to studying. Assessment's due tomorrow.