Data center news this week: October 22–28, 2026

Data center news this week, October 22–28, 2026, was mostly about who supplies compute to whom. Anthropic’s IPO filing put a ceiling of $120.5 billion on what it may pay SpaceX for GPU capacity through 2029, Oracle said it will host Google TPUs and expand NVIDIA Vera Rubin on its cloud to multi-gigawatt scale, NAVER signed a $9 billion infrastructure agreement with Brookfield, and Baker Hughes booked about 1.1 GW of gas turbines for an unnamed US technology company’s AI site.

Two new buildings also broke ground in Asia: kt cloud’s 50 MW Ansan AI data center in South Korea and L&T’s 52 MW Vyoma.AI facility in Chennai.

The week at a glance

Anthropic’s S-1 puts a number on its SpaceX compute

Anthropic filed its registration statement for a proposed Nasdaq listing on October 26, and the S-1 itself is the most detailed public account yet of its compute contracts. Between May and October 2026 it signed and amended several agreements with SpaceX.AI LLC, the SpaceX subsidiary formerly called X.AI, to use compute built on NVIDIA GPUs. Anthropic says it agreed to spend up to $120.5 billion over terms running into 2029, about $3.1 billion a month once all capacity is delivered, which it expects in December 2026. After initial non-cancellable periods that end in January and March 2027, either side can cancel each agreement on 90 days’ written notice.

The relationship started in May, when Anthropic said it would use all of the capacity of the Colossus 1 data center in Memphis, more than 300 MW and over 220,000 NVIDIA GPUs. The same filing lists a minimum commitment of $110 billion with Amazon between May 2026 and April 2036. The Memphis site is covered in our xAI Colossus explainer.

NAVER and Brookfield sign a $9 billion AI factory deal

On October 27, NAVER said it had signed an agreement with Brookfield to secure $9 billion of computing infrastructure, which it says meets the condition for NVIDIA’s planned equity investment in NAVER. Brookfield will set up a special purpose vehicle, provisionally called Asset Co., that will spend up to $9 billion on GPUs and data centers and provide the computing power to NAVER Cloud, which runs the AI factory business.

The plan is to start operating 55 MW in the first half of 2027 and 200 MW in 2028, with NAVER’s GAK Sejong hyperscale data center as the main hub, before expanding abroad. The three companies describe the wider project as 1-gigawatt scale. NAVER also agreed to guarantee NAVER Cloud’s usage-fee obligations for up to eight years.

Oracle adds Vera Rubin capacity and Google TPUs to OCI

Two announcements came out of Oracle AI World in Las Vegas on October 27. In the first, NVIDIA and Oracle said that Vera Rubin NVL72 rack systems and NVIDIA Vera CPUs are now generally available on Oracle Cloud Infrastructure, and that OCI Superclusters built on Vera Rubin NVL72 will expand to multi-gigawatt scale. Oracle also plans HGX Rubin NVL8 servers with eight Rubin GPUs and two 64-core Intel Xeon 6 CPUs. The release says OpenAI trained and served GPT-6 Astra on an OCI Supercluster in Abilene, Texas, with more than 100,000 NVIDIA Blackwell GPUs; the Abilene campus is part of the project in our Stargate explainer.

In the second, Oracle and Google announced that Google TPUs will be offered through OCI, the first time TPUs will be available on another hyperscaler’s cloud. OCI supplies the power, cooling and data center networking. Availability starts in calendar 2027, and Oracle’s own blog says the TPU capacity will ramp to gigawatt scale over multiple years. Neither release names sites. What TPUs are and who can use them is in our Google TPU explainer, and the Rubin generation is covered in Nvidia Vera Rubin explained.

Baker Hughes books about 1.1 GW of on-site gas power for AI

Baker Hughes said on October 27 that a major US technology company, which it did not name, ordered 34 Frame 5 gas turbines, nine NovaLT16 turbines, BRUSH generators and gears to provide about 1.1 GW of power for high-performance computing used for AI training, inference and physical AI. The power order was booked in the third quarter of 2026. A separate order for Chart dry coolers, for a closed-loop cooling system, was booked in the fourth quarter. The release gives no site or delivery date.

kt cloud starts its 50 MW Ansan AI data center

kt cloud said on October 22 that it is starting construction of the Ansan AI data center, a company-owned site in the Seoul metropolitan area with 50 MW of total power capacity, nine floors above ground and two below. Completion is targeted for 2029. Every server room is designed for liquid cooling, and the structure is built to carry heavier next-generation GPU racks such as NVIDIA Vera Rubin. The project is part of kt cloud’s plan to supply 1 GW of AI data center capacity by 2031.

L&T breaks ground on 52 MW in Chennai

On October 28, Larsen & Toubro’s AI cloud and data center business Vyoma.AI held the groundbreaking for a 52 MW AI-ready data center in Chennai with direct-to-chip liquid cooling. It is part of L&T’s planned gigawatt-scale AI Factory campus in the city, which already has an operating 30 MW facility and 150 MVA of power infrastructure ready. L&T says Vyoma.AI is targeting more than 250 MW of data center capacity across India. The release gives no completion date.

What the week adds up to

Oracle’s two gigawatt-scale plans name no sites or dates beyond 2027, and NAVER’s 1-gigawatt figure is a long-term ambition. The concrete numbers are in contracts: a monthly rate Anthropic expects to reach in December, a $9 billion vehicle with a 55 MW start date, and a turbine order sized at 1.1 GW. The two buildings that broke ground, in Ansan and Chennai, are each about 50 MW. Last week’s roundup, Data center news October 15–21, 2026, was dominated by power and leases in Canada, Indiana and Japan. Announced campuses are tracked in AI data centers, chip specs are in AI chips, and the biggest sites are ranked in our largest AI data centers explainer.