Anthropic commits $35 billion to Lambda for 350MW AI compute at Hut 8 Texas campus
The AI compute build-out continues with major capacity agreements, shifting focus to rapid deployment of dedicated AI factories.
The story
The AI infrastructure landscape saw significant capacity commitments this week, underscoring the ongoing demand for compute power. Anthropic, the AI lab behind Claude, agreed to a $35 billion cloud computing deal with Lambda, an AI cloud provider.
This agreement secures roughly 350MW of capacity at Hut 8's Beacon Point campus in Nueces County, Texas, with NVIDIA supplying the underlying compute. Separately, European real estate investor SWI launched its new AI cloud division, SWI Compute, backed by 3.6 gigawatts of secured power capacity across 24 sites in the US and Europe.
SWI Compute's first large-scale AI factory in Cambridge, UK, is planned to start with 30MW, deploying nearly 7,000 Nvidia Rubin GPUs and 3,500 Vera CPUs by the end of 2027, with targets to reach 2.3GW of Nvidia AI factory capacity by 2029. These developments highlight a trend where AI labs and specialized cloud providers are making substantial, long-term commitments to secure the physical infrastructure necessary for scaling AI models.
Silicon
New-generation Mali GPU architecture
Maker: Arm
What: Upgrades matrix compute units within the graphics pipeline to consolidate AI compute into GPUs for unified graphics and large-model inference workloads.
For Whom: Broad range of devices, including automotive, mobile, and embedded terminals.
E-series
Maker: Imagination
What: Showed a 4.4x improvement in convolution performance and a 4.8x improvement in matrix multiplication, with first chips expected to tape out within the year.
For Whom: On-device AI compute.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Texas AI Cloud Agreement | Anthropic, Lambda, Hut 8, NVIDIA | 350MW of capacity, $35 billion contract value | Beacon Point campus, Nueces County, Texas |
| Mikkeli AI Capacity | Cerebras Systems, Compute Nordic Finland | 165MW contracted IT capacity, first 50MW under construction, estimated €1.0-1.7 billion total build. | Mikkeli, Finland |
| Agios Dimitrios Data Center | PPC, Amazon Web Services (AWS) | 240MW IT load within 300MW total power demand, estimated €1.2 billion. | Agios Dimitrios, Western Macedonia, Greece |
| ADC3 Alcalá Data Center | Nabiax | 100MW planned IT capacity, estimated €800 million. | Alcalá de Henares, near Madrid, Spain |
| UK AI Factory | SWI Compute (Preferred NVIDIA Cloud Partner) | Initial 30MW with 7,000 Rubin GPUs and 3,500 Vera CPUs, planned to grow to 530MW, targeting 2.3GW by 2029. | Cambridge, UK |
Supply & policy signals
Rapid expansion of AI data centers risks squeezing power chip supplies for auto and electronics sectors.
Implication: Increased competition for specialized power delivery components, potentially impacting other industries.
Lumentum Holdings Inc., a supplier of optical links, is operating 70% below demand and is booked through 2029 for indium phosphide laser chips.
Implication: Optical interconnects remain a bottleneck for scaling AI infrastructure, driving investment in fabs and packaging.
New memory fabs take three to four years to come online, leading to most memory from top suppliers being sold out through 2028.
Implication: Continued high demand and potential price increases for high-bandwidth memory (HBM), impacting AI system costs.
Over 500 data center moratoria have been recorded across 42 US states due to community opposition and grid constraints.
Implication: Local opposition and power availability are increasingly limiting new data center development, especially for hyperscalers.
Gas turbine manufacturers are sold out through 2029, with orders extending to 2031, for projects requiring more than 1 GW.
Implication: Securing large-scale, dispatchable power generation for AI data centers faces significant lead times and supply constraints.
What we'll be watching
- Alphabet and Microsoft report third-quarter results on October 28.
- Amazon reports third-quarter results on October 29.
- Vertiv and GE Vernova report third-quarter results around October 28.
- Intel plans to launch a new AI data center chip by year-end 2026.
- Nvidia plans to launch its CPO (Central Processing Optimization) initiatives by late 2026 and 2027.
- TSMC to announce fourth-quarter results and initial 2027 capital expenditure in January 2027.
- Volume production at TSMC's next Arizona fab is expected in the third quarter of 2027.
Reporting + analyst voices: grounded via Google Search at publish time.