Microsoft and Chevron Lock In 20-Year Power Deal for 2.67 GW Texas AI Data Center, Deepening AI Infrastructure Buildout
Microsoft Corp. (MSFT) and Chevron Corp. have signed a 20-year power supply agreement to build what could rank among the largest data centers in the United States, a deal that underscores the accelerating scale of AI infrastructure investment and its downstream demand implications for Korean memory chipmakers.
Chevron announced the pact on Monday, disclosing plans for "Project Kilby," a proposed natural-gas-fired power facility in West Texas. The plant is designed to ramp to 2.67 gigawatts (GW) of capacity over time — sufficient to power more than 530,000 Texas homes — with initial electricity production targeted for 2028. Chevron is collaborating on the development with activist-turned-ESG fund Engine No. 1, and the oil major is expected to reach a final investment decision (FID) later this year.
The agreement represents a notable expansion of Chevron's energy portfolio into dedicated AI power production and signals Microsoft's intent to lock in long-duration, large-scale electricity supply well ahead of broader grid constraints. Bloomberg News had previously reported the two companies were in exclusive talks.
Scale and Context
At 2.67 GW, Project Kilby would dwarf most conventional data centers. For comparison, the entire existing KOSPI-listed Korea Electric Power Corporation's (015760.KS) average annual domestic new capacity additions run below 2 GW. Microsoft has been among the most aggressive hyperscalers in securing dedicated power infrastructure — a necessity as generative AI and large language model (LLM) workloads push per-rack energy densities to new highs.
The deal structure — a natural-gas plant co-developed by an established energy major and a specialized fund — reflects a broader shift in which AI firms are moving from utility rate agreements toward long-term bilateral power contracts to guarantee supply certainty for the next decade and beyond.
Korean Memory Market Implications
The deal matters to Korean investors for reasons that extend beyond energy markets. AI data centers of this scale require dense memory subsystems, and the HBM (High Bandwidth Memory) and conventional DRAM markets are the primary beneficiaries.
Samsung Electronics (005930.KS) began mass-producing sixth-generation HBM4 chips in February 2026, reporting over USD 1 billion in HBM4 revenue within four months. SK Hynix (000660.KS), currently the leading HBM supplier to Nvidia, is preparing an ADR listing on Nasdaq that analysts value at up to USD 20 billion, in part based on sustained hyperscaler memory demand.
Each AI accelerator rack in a large data center typically requires 4–8 HBM stacks per GPU, and 2.67 GW of AI-specific compute capacity translates to tens of thousands of high-performance server nodes. At that scale, Microsoft's Project Kilby alone represents a material incremental demand driver for Korean foundries.
Broader context supports the connection: Microsoft's capital expenditure on AI infrastructure reached approximately USD 21 billion in Q3 FY2026, up more than 40% year-over-year, with management signaling the pace will continue through 2027.
Sources: Bloomberg (June 22, 2026), Financial Times (June 22, 2026)
This article discusses the macroeconomic and sector implications of the Microsoft-Chevron deal as reported by Bloomberg and the Financial Times. Forward-looking statements involve uncertainty; investment decisions should be based on independent research.
