Microsoft is picking up a Texas data center project OpenAI didn’t want, in a telling sign of how far they’ve drifted apart

Microsoft will build two new AI data center buildings plus a 900 MW on-site power plant in Abilene, Texas, bringing the campus to 10 buildings and about 2.1 GW of computing capacity after OpenAI declined to expand further. Crusoe is the developer and is still completing six buildings for OpenAI and Oracle, which currently has a 350 MW gas-fired plant as backup; the Microsoft project materially increases local on-site generation and industry concentration. The move underscores accelerating private investment in AI infrastructure and raises ESG/grid implications given the reliance on gas-fired capacity, while Microsoft and OpenAI continue pursuing increasingly separate buildouts despite Microsoft’s ~27% stake in OpenAI.

Analysis

Microsoft’s decision to secure adjacent physical compute capacity is a strategic move to convert optionality into an owned supply advantage: it lowers execution risk for large model training runs and creates local “data gravity” that increases switching costs for enterprise and model customers. That second-order lock-in can translate into outsized incremental Azure revenue capture per MW of compute available — a leverage point few investors model explicitly when valuing cloud growth.

The build-out also reshuffles upstream demand: manufacturers of data-center infrastructure (generators, transformers, switchgear) and chipmakers supplying accelerators will see lumpy, regional spikes in orders that outsize normal cloud cadence. That creates a two-way supply risk — accelerated order books can push spot prices and lead-times for critical components, but if GPU supply or labor bottlenecks emerge, compute utilization and time-to-production for models will lengthen, muting near-term revenue.

Policy and ESG pressures are the asymmetric tail risks: on-site generation and rapid industrialization invite faster permitting scrutiny, community pushback, and potential carbon accounting costs that can hit timelines and investor optics. Near-term catalysts that will resolve uncertainty are Azure contract disclosures, local permitting wins/losses, and quarterly signs of accelerated large-model spend; these play out across months-to-years, not days.

Contrarian lens: the market underprices the capex rigidity risk — owning physical capacity hedges against cloud price volatility but risks stranded assets if demand softens or renewables/storage economics accelerate. That makes Microsoft’s move a convex bet: modest near-term upside if compute demand materializes, but real downside if the industry shifts to smaller, more distributed footprints or tighter carbon rules.

Key Decisions for Investors

• Long MSFT (12-month call spread, small starter size ~1-2% NAV): buy a 12-month call / sell a further OTM call to fund premium; thesis: capture upside from Azure share gains and pricing power if large-model demand accelerates. Target +25-40% equity upside; cut to flat on two consecutive quarters of cloud rev/usage misses.
• Pair trade — long MSFT / short ORCL (6–12 month horizon, 1:1 notional, small sizing): Microsoft wins optionality and scale; Oracle has less to gain from neighbouring capacity consolidation. Aim for asymmetric 3:1 upside/downside (expect outperformance rather than absolute move); exit on Oracle re-acceleration or MSFT guidance deterioration.
• Play supply chain winners: initiate long positions in select industrials supplying on-site generation and transformers (select names, 12–24 months, tactical 1–3% NAV each): catalyst is visible order flow and margin expansion if build wave continues. Hedge with a small put in case permitting stalls or commodity inflation compresses margins.
• Convex maker play — long GPU exposure (NVDA 6–12 month calls, tactical size): If demand for large training runs scales as expected, GPU makers re-rate; risk is inventory correction and pricing volatility. Risk/reward: high volatility but large upside if multi-center builds convert into sustained bookings.