Project Kilby / Microsoft + Chevron 2.67 GW behind-the-meter / Pecos TX Permian Basin / 20-yr PPA / $7B+ capex / FID YE 2026 / first power 2028 / OFF-grid (bypasses ERCOT)
Microsoft signed a 20-year power agreement with Chevron on June 22 for 'Project Kilby' — a ~2.67 GW natural-gas-fired plant co-located with a Microsoft AI data center campus near Pecos in West Texas, structured behind-the-meter so the generation sits inside the campus fence and does NOT deliver into ERCOT — co-developed with Chevron's Energy Forge One subsidiary and Engine No. 1's Joulent, $7B+ in capex, FID expected by end of 2026, first power 2028, phased modular build-out — the single most explicit hyperscaler bet yet that the cleanest answer to the grid + regulatory fight is to skip the grid entirely
Plain-English version of what Project Kilby is. Microsoft and Chevron jointly announced on June 22 that they are co-developing a roughly 2.67 gigawatt natural gas power plant in the Permian Basin near Pecos, Texas (about 200 miles east of El Paso), paired with a brand-new Microsoft AI data center campus on the same site. Microsoft signed a 20-year power purchase agreement (PPA) to buy the plant's output. The plant is being developed by 'Energy Forge One,' a Chevron-owned subsidiary, together with the investment firm Engine No. 1 through its Joulent platform. Total capital expenditure is reported above $7 billion. Chevron expects a final investment decision (FID) — the formal go/no-go board approval that releases the construction budget — by the end of 2026. First power is targeted for 2028, with the campus scaling up in phases. The technical structure — and why it matters. Project Kilby is explicitly designed as 'behind-the-meter' (BTM). In plain English, behind-the-meter means the generator sits inside the customer's fence and delivers electricity directly to the load, without that electricity ever flowing onto the public grid. So the 2.67 GW of new gas generation, by design, does not sell into ERCOT's wholesale market and does not show up on ERCOT's interconnection queue as a grid-connected generator. The data center, similarly, draws its power from the on-site plant rather than from ERCOT-delivered transmission service. The data center and plant are still physically located inside ERCOT's geographic territory, but the entire energy transaction happens privately between Chevron and Microsoft. Why this is a different category of deal. Most big hyperscaler power deals in 2024-2026 have been grid-connected: the hyperscaler buys a virtual or contracted power purchase agreement (a paper trade that helps the developer finance a wind, solar, or gas plant somewhere else on the grid), and the data center itself draws power from the existing utility. Kilby is the opposite — the generation is physically dedicated, on-site, and behind-the-meter, with no dependency on the ERCOT interconnection queue for the load or on a new transmission line to deliver the power. That structure is what is meant by 'bypassing the grid.' Why hyperscalers are reaching for this template right now. Three forcing functions: • ERCOT's Batch Zero process — covered in detail in earlier issues — requires every large load above ~75 MW (data centers, mostly) to be batched and studied together, with the first batch closed off by deadlines in July 2026 and roughly 438,000 MW of proposed demand stacked behind it. A grid-connected hyperscaler that misses Batch Zero waits years for the next batch. A behind-the-meter campus does not enter the batch process at all. • FERC's June 18 show-cause orders — also covered in detail — put all six FERC-jurisdiction RTOs on a 60-day clock to either justify or rewrite their large-load tariffs, and the entire federal cost-allocation question for data centers is now genuinely unsettled. Off-grid avoids that whole fight. • Gov. Abbott's June 10 directive (next item) explicitly tells PUCT and ERCOT to force data centers to pay their own infrastructure costs and to actively reduce residential transmission costs. Off-grid avoids the cost-shift fight by ensuring no Texas residential ratepayer is on the hook for any of Kilby's wires. What makes Pecos a viable site. Pecos sits in the Permian Basin, the largest oil and gas producing region in the U.S. The same geology that makes the area oil-rich makes natural gas effectively a co-product — local gas is cheap and abundant, and Chevron already operates Permian midstream infrastructure. Land is plentiful, the population density is low (community opposition is structurally weaker than in Loudoun County or the Research Triangle), and water for cooling is the main constraint. Chevron is providing the fuel feedstock; Engine No. 1 / Joulent is providing the structured-finance / energy-developer expertise; Microsoft is providing the 20-year demand certainty that makes the project financeable.
Why it matters
Three implications for Cliff. (1) Kilby is the most concrete proof-point yet for the BTM generation economics SKU Cliff has had queued behind the live-docket wedge. The strategy doc treats 'behind-the-meter generation economics' as one of the six structurally-empty software layers; Kilby is what a Cliff customer underwriting that template looks like at hyperscaler scale. Every variable that determines whether a Kilby-style deal pencils — Permian gas basis, on-site gas infrastructure, low-population-density siting, hyperscaler-grade 20-year offtake, $7B+ structured finance — is a discrete data input that is currently held in a single Tier-1 consulting team's head per project. A Cliff product that lets a second-tier developer or hedge-fund counterpart underwrite the same structure in a week instead of six months is exactly the AI-native moat the strategy doc describes, and it has just been priced by the market via Microsoft's signature. (2) Kilby reframes the moratorium streak. Every moratorium covered for the last two weeks (NC, WI, SC today) is structurally aimed at *grid-connected* data centers — they regulate land use, county-level permitting, and ratepayer cost allocation. A behind-the-meter campus in a low-density Permian county largely sidesteps that politics; it is not unaffected (water, noise, air permits still bind) but the cost-allocation argument that powers the moratorium movement has nothing to bite on. So the strategic split among hyperscaler-aimed sites is bifurcating: grid-connected sites in dense, high-political-pressure regions (Loudoun, the Triangle, Charlotte metro, MISO) face escalating regulatory risk, while remote behind-the-meter sites in oil-and-gas-friendly low-density counties become structurally cheaper to underwrite. That is the cleanest 'two-regime' map the corpus has yet seen, and Cliff should tag every project in its underlying corpus as grid-connected vs. BTM-eligible as a first-class field. (3) The Texas timing is not a coincidence. Kilby was announced on June 22 — twelve days after Abbott's June 10 letter to PUCT/ERCOT (next item) which explicitly tells the regulators to force data centers to pay their own infrastructure and to actively reduce residential transmission costs. Reading the timing as causal: Microsoft and Chevron locked in this structure because the politically credible 'data centers must self-fund' regime is now state policy, and the cleanest way to comply with 'pay your own infrastructure' is to literally build your own infrastructure inside your own fence. The next 6-12 months will likely see at least 2-3 more deals of this shape announced in the Permian, in the Eagle Ford, and probably in the Marcellus (Pennsylvania) where the gas-cost economics are similar. Cliff's diligence product should be ready to underwrite them when they appear.