Florida wants a court to halt OpenAI over extinction fears, and the same morning Akamai books eleven-point-six billion dollars with Anthropic. Guess which one actually moves the industry. If you're just joining us: AI labs and hyperscalers have been signing years-long infrastructure commitments at a startling pace. Anthropic took its first Australian lease at Zerra DC's proposed A$32 billion Western Downs campus in Queensland. Meta pre-leased CleanSpark's 175-megawatt facility in Sandersville, Georgia, on a 20-year deal worth about $6.6 billion in contracted cash flows. And CleanSpark financed that campus with $2.276 billion in 7.875% senior secured notes. This is AI Daily Briefing. Today: a courtroom fight, a giant contract, Google and Microsoft planting flags abroad, and the question of whether any of this spending is real demand or a bubble. We start in Florida, where the filing gets biblical. We're staying with this story: Big Tech AI lease obligations. Follow the show and you won't miss what comes next. Ars Technica writes:
“It is only by the grace of the Almighty that one of Defendants’ AI agents hasn’t compromised a water supply or shut down a power grid—yet,” the state writes in the motion. OpenAI representatives have yet to respond to a request for comment from Ars Technica regarding the injunction motion.
"Only by the grace of the Almighty." That's the actual argument in Florida's injunction motion against OpenAI. Water supply, power grid, and not one measured failure rate anywhere near it. And look at what's stacked into one brief. Chemically sterilizing humanity, kids hooked on sycophantic models, chatbots ending replies with leading questions. Blend all that together and you never have to say which one your remedy fixes. Right, so what's the remedy? Freeze OpenAI's training runs? Every model already deployed keeps serving traffic, and every other lab keeps training. Ars basically says so. Here's the contrast I'd hold onto. A federal appeals court just put guardrail authority in Hegseth's hands, and now Florida wants a state judge to pull the emergency brake. Same week, two governance theories, pointing opposite directions. And the stuff from the original June complaint, the false-advertising and kids claims? That's the part a court could actually act on. Which is exactly the part the water-grid prophecy buries. From Dan Swinhoe at Data Center Dynamics:
The seven-year deal, valued at $11.6 billion, will see Anthropic to leverage Akamai Cloud's distributed infrastructure and software to support its CPU workloads. Akamai said the deal includes potential expansion by up to an additional $9bn, reaching a total potential commitment of approximately $20bn.
Run the division. Eleven-point-six billion over eighty-four months is roughly a hundred thirty-eight million a month. And Akamai's the one handing over equity: a warrant for up to five percent of the company, two percent vesting on this deal alone. The vendor is paying the customer to show up. Which tells you who had the leverage in that room. Our running tally of compute commitments, from Anthropic's Australian lease to Meta's Georgia campus, just got its biggest line item. And it answers the Memphis question, at least partly. Here's a partner outside Musk's orbit for some of the stack. Part. It's CPU workloads. Anthropic serves models on accelerators, so what's Akamai actually running? Edge routing? Tool calls? Preprocessing? Until somebody says, I'm reading this as a contract story more than a compute story. Wait, I'd push on that. Agents chew through CPU: sandboxes, code execution, browsing. And Akamai's adding one-point-seven billion to this year's capex just to pre-buy memory. Funny contrast with the Florida brief we just hit. One state wants a court to halt development, and meanwhile the money's locking in through 2033. Here's Srinath at Business News Today:
Starting October 15, Gemini Enterprise on the web will support in-country processing for Gemini 3.5 Flash in Brazil, complementing existing local storage capability for agent workloads. Google is also making G4 virtual machines powered by NVIDIA RTX PRO 6000 Blackwell Server Edition GPUs available from its São Paulo cloud region.
Okay, Brazil. The date I care about is October fifteenth. That's when Gemini Enterprise starts processing Gemini 3.5 Flash in-country, and it's the first actual go-live on the calendar in all this regional buildout talk. A clock I can check. And notice which model got the in-country treatment. Flash. Not the flagship. Google's betting São Paulo enterprises want the cheap, fast one close to home, and honestly? Probably right. Yeah, same read on the hardware. The G4 VMs run RTX PRO 6000 Blackwell Server Edition. Real chip, but it's the inference-and-graphics part, not the big training silicon. Somebody sized this for agents answering support tickets. What I'd push on is the headline. 'Double by 2030,' and Google hasn't put a capex figure on it. We just walked through Akamai's eleven-point-six billion with actual contract terms, so 'double' with no dollar sign attached? That's a vibe. The part with teeth is the data-residency pitch: who gets to touch the data an autonomous agent is chewing on. It's a sovereignty argument, and now an American hyperscaler's the one making it. Press Network of India writes:
The visit follows Microsoft’s announcement in New Delhi earlier this week that India South Central is now live and has been designated a strategic hub for Asia and the Global South. This is Microsoft’s fourth cloud region in India and features a resilient three-zone architecture designed to expand as customer demand grows.
Okay, Telangana has one thing the Brazil piece we just hit doesn't. It's already live. Fourth Microsoft region in India, three zones, and the pitch is it expands as customer demand grows. Build to demand is the boring, correct way to do this. Boring and correct, and then they put 'strategic hub for Asia and the Global South' on the sign. That phrase was written for the minister holding the ribbon scissors, not for anybody provisioning VMs. Ribbons are cheap. Three availability zones aren't. My question is who's actually filling it. The minister's bragging about GCCs growing fast, so is this Indian enterprise demand, or Microsoft parking its own AI load somewhere with cheaper power and a friendly state government? Either answer lands on the same map, though. Google's promising in-country processing in Brazil, Microsoft's calling Hyderabad a regional hub, and on the same morning Anthropic's signing seven years with Akamai. American hyperscalers are pitching data residency in exactly the language governments want to hear, one country at a time. So why are AI companies locking themselves into these huge, long-term cloud and data-center deals? And how do we tell if it's real demand, or just everyone building way ahead of the market? Short version: they need the compute. Frontier labs need enormous amounts of capacity, and “take-or-pay” contracts reserve it even if the customer ends up using less than expected. CNBC's Tobias Burns reports these agreements have become a focal point for Wall Street, because they can be the weak link in the financial chain of the AI buildout, and he points to the Akamai-Anthropic deal as one example. The scale is wild. MIT Technology Review reports AI hyperscalers are likely to spend more than $1 trillion on data centers next year, and McKinsey estimates global data-center spending could reach $7 trillion by 2030. So companies are basically betting demand for training and running AI holds up long enough to justify committing the capital now. The catch is those contracts push risk onto the customers. If AI revenue or usage disappoints, they still owe for the capacity they reserved. So what would count as proof this is paying off, and not just a bigger and bigger financing machine? First, whether AI companies can bring in enough revenue to sustain the infrastructure spending. That's the central question MIT Technology Review keeps raising. Then the practical constraints McKinsey flags, especially access to capital and energy, because a signed contract isn't the same thing as a profitable, fully running data center. The healthiest signal would be demand that keeps supporting those commitments without customers straining to meet their take-or-pay obligations. You might also enjoy The Data Center Daily, a daily briefing on AI compute, hyperscaler capex, the power grid, semiconductor supply, and energy markets reshaped by intelligence at scale. Find it wherever you listen to podcasts.
Looking ahead, October 15 is the checkpoint for Gemini Enterprise to start in-country processing for Gemini 3.5 Flash in Brazil. And Akamai expects no Anthropic revenue in 2026, then $150 million to $300 million in the second half of 2027, and roughly $1.7 billion in annualized run rate by the end of 2028.
Links to every story are in the show notes, so follow up on whichever ones caught your attention. That's AI Daily Briefing for today. This is a Lantern Podcast.