Venture Global to build gas-fired power plants for 10 GW of AI data center capacity.
The global AI compute build-out accelerates with significant investments in power generation and datacenter capacity, while chipmakers unveil new accelerators and supply chains face persistent demand.
The story
The AI infrastructure build-out continues its rapid expansion, with new players and technologies addressing the surging demand for compute and power. Venture Global announced agreements with GE Vernova and Baker Hughes Energy Services to construct gas-fired power plants, aiming to provide up to 10 gigawatts of computing capacity for AI data centers at its owned sites.
This move by the natural-gas exporter signifies a growing trend of direct power generation for compute, with initial power availability projected for late 2028 or early 2029. Meanwhile, AI labs are securing massive compute resources; Anthropic is reportedly sprinting to switch on data centers, having signed approximately $530 billion in leasing deals for an estimated 15 gigawatts of capacity over the coming years.
On the silicon front, at the OCP Global Summit on October 14, Intel detailed its upcoming GPU D-series, an air-cooled data center GPU designed for large language models, and showcased server racks with over 46,000 cores from Intel Xeon 6+ processors. AMD also presented a new flexible server rack reference design and an EPYC CPU with LPDDR memory interface. These developments underscore the industry's push for specialized silicon and integrated infrastructure solutions to meet the escalating requirements of AI workloads.
Silicon
Intel GPU D-series
Maker: Intel
What: Air-cooled data center GPU for large language models, long-context workloads, and concurrent services, supporting models up to 300 billion parameters on a single card.
For Whom: Data center operators building for agentic AI workloads.
Intel Xeon 6+ processors
Maker: Intel
What: Used in server racks with over 46,000 cores, designed to orchestrate workloads and manage I/O for accelerator chips.
For Whom: Data center operators needing high-density CPU capacity for agentic AI.
AMD EPYC CPU (with LPDDR memory interface)
Maker: AMD
What: First AMD EPYC CPU to include an LPDDR memory interface.
For Whom: Data center operators seeking improved memory performance for AI workloads.
The build-out
| Project | Who | Scale | Where |
|---|---|---|---|
| Gas-fired power plants for AI data centers | Venture Global, GE Vernova, Baker Hughes Energy Services | Up to 10 gigawatts of computing capacity | Sites owned or controlled by Venture Global |
| AI Data Center Leasing Deals | Anthropic | Estimated 15 gigawatts of capacity | Various cloud providers and data center firms, including a site outside Baton Rouge, Louisiana (Hut 8). |
| Prefabricated AI data center project | BrightRay (with unnamed US partner) | 10MW | Indiana, US |
| Semiconductor manufacturing project | Japan and the United States | Approximately $10 billion investment | New York, US |
| AI infrastructure investment | €13 billion over two years | Finland |
Supply & policy signals
US AI data center startups raised $33.5 billion in funding by early October 2026.
Implication: Investors are increasingly backing integrated energy and computing infrastructure solutions, with over 80% of 2026 funding targeting power supply and computing efficiency.
TSMC Q3 2026 revenue rose to $46.7 billion, reaffirming a $60-64 billion capital budget for 2026.
Implication: Strong demand for advanced AI chips continues, but gross margin is under pressure due to 2nm production ramp and overseas fabs.
A TSMC executive fields at least 10 calls daily from AI buyers competing for chip capacity.
Implication: Intense competition for advanced chipmaking capacity persists, with constraints potentially threatening AI infrastructure investments and revenue targets.
Siemens invested in Skeleton Technologies for supercapacitor technology on October 15, 2026.
Implication: Aims to manage rapid power fluctuations in AI data centers, potentially increasing compute density without extensive facility upgrades or grid capacity waits.
AI infrastructure and hardware stocks slumped on October 14, 2026.
Implication: Rising US interest rate expectations are increasing borrowing costs for data center investments and chip purchases, leading to investor caution on returns.
What we'll be watching
- TSMC's 2nm production ramp and its impact on gross margins.
- Deployment of Microsoft's Maia 300 chip in Azure data centers starting in 2027.
- Construction start for the Japan-US semiconductor manufacturing project in New York by the end of 2026.
- Venture Global's equipment orders for the first phase of on-site power generation by year-end.
- Anthropic's IPO, expected next month, and its impact on data center build-out financing.
- Lam Research's integration of NanoVDB into SEMulator3D, targeted for release by the end of 2026.
Reporting + analyst voices: grounded via Google Search at publish time.