Three multibillion-dollar compute bets in one morning—and the ownership math is getting uncomfortable. If you're joining us mid-arc, here's the short version: major AI companies are signing huge data-center and GPU-capacity commitments. Compute access is becoming as much a balance-sheet story as a model-building one. So far, that includes big buildout plans tied to AWS, Alibaba, ByteDance, and DeepSeek, plus reporting that Anthropic signed $517 billion in computing-capacity contracts over 11 months—even as it trails OpenAI's planned 30 GW by 2030. This is AI Daily Briefing. Today, who gets the keys when the chip vendor, the landlord, and the customer start looking like the same machine? Forkast, with Lena Park:
While Anthropic and Lambda have yet to officially confirm the details, reports from the Wall Street Journal and subsequent verification by Reuters indicate a $35 billion cloud deal centered on 350 MW of capacity at the Hut 8 Beacon Point campus in Texas. This arrangement, Anthropic’s ninth major compute corridor, reveals a shift where the primary hardware supplier, Nvidia, effectively acts as the landlord, ensuring its influence permeates every layer of the stack.
The Anthropic compute story moves again: Lena Park reports a $35 billion Lambda corridor for 350 megawatts in Texas, and Nvidia holds the lease on the site. So Lambda gets to operate the cluster, Anthropic gets the capacity, and Nvidia sells the GPUs while collecting the building rent. That is a very cozy negotiation for the company on both sides of the utility bill. This is corridor number nine for Anthropic. That $517 billion headline gets more concrete with Beacon Point's 350 megawatts—and a lease Nvidia controls. Nvidia gets paid for the GPUs and the building. Anthropic can still choose a cloud intermediary—but the site is already spoken for. Here's Andrew Samu at Disruption Banking:
That is the story. Not that Jane Street wants more compute. It has wanted more compute for years. The story is that $13 billion of it now sits in a single cloud contract, on a five-year clock, with a firm that used to mine Bitcoin.
Jane Street just signed a $13 billion, five-year deal with Crusoe—a company that started in Bitcoin mining. That's a huge bet that Crusoe's compute costs will still look good in year three, not just this quarter. And it's a single-vendor contract. After the Nvidia-landlord piece we just hit, that changes the read: big buyers may be diversifying away from hyperscalers, but they're still concentrating their exposure somewhere. Jane Street's rumored 100-to-200-megawatt self-build still has no confirmed site or grid connection. The Crusoe invoices are real, though. Apparently the data hall can wait; the GPU bill cannot. From Ann Mary Peter at Medianama:
Tata Consultancy Services’ data centre subsidiary, HyperVault, has acquired 264 acres in Hyderabad to develop an artificial intelligence data centre campus, with a planned investment of up to Rs 70,000 crore, as announced by the company and the Telangana government on September 5. Telangana Chief Minister A. Revanth Reddy requested the company to complete the project within 21 months, targeting an inauguration on June 2, 2028, which marks the state’s formation day.
A 264-acre, one-gigawatt campus in Hyderabad, targeted for June 2, 2028— very tidy on the ceremonial calendar. But a 21-month greenfield build with no announced anchor tenant is a utilization bet as big as a small power market. Rs 70,000 crore puts HyperVault's Future City plan in the same league as the Texas capacity we just covered. Telangana's chief minister has attached a public deadline, so permits, power deals, even cooling claims, are all on the clock. TCS seeded HyperVault with Rs 75 million, then TPG put up Rs 18,000 crore, and now the headline is a Rs 70,000-crore build. The spreadsheet has entered its optimistic phase. Liquid cooling and water-neutral design sound great; I want the phased megawatts and the customers who'll keep those racks hot. South Asia's firmly on the map this week. There's a real site and developer, plus a date voters can remember. The 1-gigawatt ceiling matters—but the story only gets there if those 264 acres turn into delivered compute. If you’re finding this briefing useful, please subscribe or leave us a review wherever you’re listening. Reviews help other people discover the show, and they mean a lot to our team.
We’ll be watching whether TCS’s HyperVault can complete the Hyderabad AI data-centre campus in time for its June 2, 2028 inauguration target. Links to every story are in the show notes, so take a look at the ones you’d like to explore further. That’s AI Daily Briefing for today. This is a Lantern Podcast.