Stargate Phase 2 and the Texas compute corridor — when the OpenAI capacity commitment becomes a 4.5GW reality
Oracle and OpenAI's confirmation of Stargate Phase 2 — the Abilene, Texas campus expanding to 4.5GW total capacity with 2027 first-power timeline — is the operational moment when the multi-state compute-corridor strategy stops being a $500B PR figure and becomes a buildout commitment with first-power dates. The capacity scale changes what OpenAI can plan for and what competitors have to respond to.
The capacity commitment is the substantive piece. The Stargate Phase 2 confirmation expanding the Abilene buildout to 4.5GW total capacity is the largest single AI-compute buildout disclosure of the year. The structural shape — four additional hall complexes on the existing 875-acre site, 2027 first-power timeline for the expansion, $40B+ capital commitment from Oracle, OpenAI, SoftBank, and the various capital partners — translates the Stargate joint-venture's prior announcements into a concrete buildout schedule.
The workload-side justification is what makes the procurement scale credible. GPT-5.2's release the same day and the ChatGPT Agent Mode priority tier launch are the consumer-side workload-driver; the OpenAI Pro and Plus tier subscriber growth through 2025-2026 has been the structural demand-side commitment. The frontier-training workloads — GPT-5.3 and beyond, the next-generation Sora trajectory, the agent-platform infrastructure — are the future workload-side commitment Phase 2's 2027 first-power timeline addresses. The capacity-and-workload commitments are matched on a multi-year horizon rather than spot procurement.
The Microsoft-OpenAI relationship structure is the strategic shape the buildout operates inside. Microsoft retains primary capacity allocation through 2026 under the existing Azure-OpenAI commitment; Phase 2's 2027 first-power capacity is uncommitted and available to OpenAI directly. The structural shape lets OpenAI plan the post-2026 trajectory with capacity certainty that does not depend on Microsoft's renewal-and-expansion decisions. For OpenAI's competitive positioning, the capacity-certainty piece is what enables the multi-year product-and-research roadmap planning that competitors are doing on their own dedicated compute commitments.
The competitive-procurement implication is the multi-stack-multi-region pattern that frontier-lab compute procurement is consolidating on. AMD's MI400 roadmap acceleration and TSMC's 2nm capacity allocation are the supply-side commitments that match the Stargate buildout timeline. The combined picture — 4.5GW Texas capacity coming online through 2027, AMD and NVIDIA both shipping accelerated next-generation silicon on overlapping timelines, TSMC 2nm capacity allocated multi-year — is the most coherent supply-and-capacity outlook frontier AI has had through any 18-month horizon.
The Scale AI revenue trajectory is the data-and-RLHF side of the procurement-stack picture. Scale's revenue doubling to $1.6B annualized with Meta as the dominant single-customer driver reflects that the post-training data-and-RLHF spend is scaling alongside the compute capacity. The frontier-AI cost structure is shifting toward a more balanced mix of compute, data, RLHF, and infrastructure spend rather than the compute-dominated profile of prior years. The Scale IPO acceleration to mid-2027 is the public-market signal that the data-and-RLHF infrastructure category is a durable business model.
The geographic-concentration question is the longer-arc consideration. Concentrating 4.5GW of frontier-AI compute capacity in a single Texas campus is a structural commitment to the Texas compute-corridor strategy — water rights, power-transmission infrastructure, workforce-housing, and the broader regional development implications. The Texas Public Utility Commission's load-interconnection queue and the ERCOT grid-capacity outlook will be the binding constraints on whether the buildout proceeds on the announced 2027 schedule or slips. The political-and-regulatory work to maintain the Texas compute-corridor positioning is as important to the buildout as the capital commitment.
The line: the multi-state compute-corridor strategy used to be a $500B PR figure. In mid-2026 it is a 4.5GW Texas buildout with first-power dates, supply commitments from AMD and TSMC, and the workload-side commitment from GPT-5.2 and Agent Mode to justify the procurement scale.
Oracle — Stargate Phase 2 Abilene Texas expansion May 28 2026 → · OpenAI — Stargate compute corridor Phase 2 confirmation → · WSJ — Stargate Phase 2 4.5GW Abilene Texas 2027 timeline →