The AI buildout has found leverage. Now: who gets stuck holding the bill? If you're just joining us, Crux AI is the TPU-focused cloud venture Google and Blackstone have brought into public view. Blackstone's supplying the equity, and Google's custom chips are at the center. The company hired Meta data-center engineering veteran Alan Duong as chief development officer, while the internally named Project Braid has reportedly faced site delays. This is AI Daily Briefing. We're looking at whether these AI infrastructure commitments survive a falling inference-price curve. Start with Crux. We'll keep tracking this story — Crux AI TPU neocloud launch. Follow the show so the next update finds you. From Amit Chowdhry at Pulse2:
Crux AI, the new cloud computing venture tied to Blackstone and Alphabet, is lining up a $22 billion loan from a group of 10 banks to finance purchases of Google’s custom artificial intelligence chips, according to Bloomberg. The transaction represents one of the largest debt financings assembled specifically to fund the processors needed for AI computing infrastructure.
Crux went from hiring and site questions to a proposed $22 billion TPU-backed loan. Ten banks are lending against Google chips and customer contracts, so I'd want to look very closely at those contracts. This clears the “is Crux real?” bar. Goldman, Barclays, SMBC, BNP Paribas—serious lenders are underwriting it. But those contracts only cover debt service if customers still want the capacity after inference prices fall. Exactly. The chips have resale value. The contracts are where the risk gets passed around. If those commitments are short, flexible, or tied to demand forecasts that miss, the banks may find they've financed a very expensive inventory problem. And they added a separate $1 billion revolver, because apparently twenty-two billion dollars of TPU debt needed a little walking-around money. The planned bond-market refinance gets easier only if Crux can sell compute at a margin that survives the next price cut. Laura Bennett, writing in Shattered.io:
Yotta’s plan splits 80,000 GPUs across two sites. The first, a new D4 data center under construction in Greater Noida, will house 40,000 Vera Rubin GPUs inside a 120-megawatt facility. The second, an NM2 campus expansion in Navi Mumbai, will run 40,000 Nvidia GB300 Blackwell Ultra GPUs across 80 megawatts.
Yotta split a $12 billion order into two very different production bets: 40,000 Vera Rubin GPUs in Greater Noida and 40,000 GB300 Blackwell Ultras in Navi Mumbai. Calling that one deployment hides the hard part. And Noida is 120 megawatts for the Rubin fleet. India has apparently landed an early spot in Nvidia's Vera Rubin allocation—well ahead of the usual “we'll take a few racks” crowd. The Noida schedule is where I get nervous. Vera Rubin is Nvidia's first HBM4 platform, and 40,000 units means Yotta needs supply, integration, power delivery, and customer demand to all line up at once. Crux is putting bank debt behind TPU purchases and contracts. Yotta is putting $12 billion of capex behind two campuses and two chip generations. The headlines are getting more concrete, and delivery risk is getting more expensive. Aakruti Thakkar, writing in Compute Forecast:
AirTrunk is committing RM12 billion, or about $3 billion, to two hyperscale data center campuses in Johor, extending one of its largest infrastructure investments in Southeast Asia. The JHB3 and JHB4 campuses in Iskandar Puteri will provide more than 280 megawatts of combined IT load and sit near the company’s existing JHB1 and JHB2 facilities.
AirTrunk's putting $3 billion into JHB3 and JHB4 for 280 more megawatts in Johor. And unlike a lot of this week's grand-build language, its existing 420-megawatt pair was almost fully contracted before it greenlit the next phase. That's the sequencing lenders should want to see: customers filling JHB1 and JHB2, then expansion beside them in Iskandar Puteri. “High-density AI” is the brochure phrase. Contracted load is the useful one. Johor's headed past 700 megawatts across four AirTrunk campuses. After the Crux debt stack and Yotta's GPU order, this is the less flashy part that decides whether those chips have somewhere practical to run. And AirTrunk is now at $6.8 billion committed in Malaysia. At that scale, it's a serious claim on power, water, and delivery schedules. Compute Forecast, with Kiara Mandavia:
Anthropic has signed its first Australian data center lease, marking a major step in the AI company’s expansion across the country’s digital infrastructure market. The agreement covers capacity at the proposed Western Downs Digital Park in Queensland, according to Australian media reports. Singapore-based Zerra DC, an AGP-backed data center developer, is developing the project and expects capacity to begin coming online in 2027.
Anthropic's first Australian lease is at a proposed Queensland park that starts coming online in 2027. That reserves future capacity; it doesn't mean 2.16 gigawatts are sitting there waiting for Claude. And it's for inference, which makes the math nastier. Inference prices keep falling, so Anthropic and Zerra DC need utilization that holds up long after the press release has evaporated. Zerra's four-phase plan spans 725 hectares near Dalby, tied into the Braemar substation and nearby generation. That's a serious power plan, but the 2.16-gigawatt figure is for full build-out, not Anthropic's first lease. Add a lab reserving inference capacity on a whole new continent. The concrete is still two years out. If you want to go deeper on the infrastructure behind AI, try The Data Center Daily—a daily briefing on AI compute, hyperscaler capex, the power grid, semiconductor supply, and how intelligence at scale is reshaping energy markets. Find it wherever you listen to podcasts.
We'll be watching whether banks bring additional lenders into Crux AI's $22 billion TPU financing through syndication, and whether Zerra DC's Western Downs Digital Park gets Western Downs Regional Council approval for Anthropic's planned 2027 capacity.
Links to every story are in the show notes, so check out the ones you'd like to explore further. That's AI Daily Briefing for today. This is a Lantern Podcast.