Oracle

Type: Company (ORCL) — Hyperscaler / demand side

Oracle was previously an acknowledged hole in this KB: named in the buildout (Stargate I at Abilene, 1.2GW; listed on NVIDIA's cloud map) but with no per-company capex figure in any ingested source. That gap is now partly closed. Oracle is guiding to $50B of 2026 capex, +136% YoY — the fastest growth rate of any name in the cohort, off the smallest base — backed by $523B in remaining performance obligations.

That RPO figure is the most striking number in the topic. At roughly 10x the annual capex it supports, it is the single largest disclosed demand commitment in this KB and the strongest available answer to "who pays the rent" — if it is durable. The KB has no data on its duration, cancellability, or customer concentration, and Oracle's RPO is widely understood to be concentrated in a small number of very large AI counterparties. A $523B obligation from one or two credits is a different asset from the same figure spread across a thousand. See Contracted Backlog & RPO.

Oracle is also the anchor participant in Stargate (the OpenAI / SoftBank / Oracle JV), which targets $500B in AI infrastructure by 2029 — $100B in initial deployment and more than $400B of commitments inside the first three years. That is a multi-year target, not committed spend, and should be labelled as such wherever it appears.

Key Facts

• [FORECAST] 2026 capex $50B, +136% YoY — closing a gap this KB previously flagged as open. Evidence: weak (single analysis source, forward guidance) (Futurum AI Capex 2026)
• $523B in remaining performance obligations — the largest disclosed demand commitment in this KB, ~10x the annual capex it backs. Evidence: weak (single analysis source) (Futurum AI Capex 2026)
• Stargate I (Abilene, TX): 1.2GW — a gigawatt-scale anchor site. Evidence: moderate (analysis — Data Center Knowledge) (Hyperscalers in 2026)
• [FORECAST] Stargate JV (OpenAI / SoftBank / Oracle) targets $500B in AI infrastructure by 2029 — $100B initial deployment, >$400B committed in the first three years. A target, not spend. Evidence: weak (single analysis source, long-horizon projection) (Futurum AI Capex 2026)
• Named on NVIDIA's supply-chain map under "Clouds/model builders." Evidence: moderate (technical-report — NVIDIA) (NVIDIA — Building America's AI Infrastructure)
• Oracle's inclusion changes the aggregate arithmetic: the $660–690B five-company 2026 total includes Oracle's $50B, so the implied big-four figure is $610–640B — well below the ~$725B this KB previously quoted. Evidence: n/a (derived arithmetic on stated source figures) (Futurum AI Capex 2026)
• RPO duration, cancellability, customer concentration; Oracle's own cloud revenue and margins: NOT in any ingested source. Evidence: n/a (gap)

Mentioned In

• Contracted Backlog & RPO — the $523B RPO is the concept's largest datapoint
• Hyperscaler Capex Cycle — Oracle's $50B is what makes the five-company and big-four aggregates non-comparable
• Power: The Binding Constraint — Stargate I (1.2GW) is one of the gigawatt-scale anchor sites

Related Entities

• Hyperscaler Buyers — Oracle sits outside the big-four grouping but inside the demand cohort
• Alphabet — the other large disclosed backlog in this KB
• CoreWeave — both rent AI capacity; the concentration risk is structurally similar

Outstanding Questions

• What is the duration, cancellability, and counterparty concentration of the $523B RPO? This is the most decision-relevant unknown attached to any single number in the topic.
• Is industry backlog being double-counted — can the same end demand appear as RPO at Oracle, backlog at a hyperscaler, and contracted revenue at a neocloud?
• How is Oracle financing a +136% capex step-up, and what does that do to its balance sheet?
• What portion of Oracle's RPO is Stargate/OpenAI-related versus diversified?
• Next action: ingest Oracle's own filing (10-Q / earnings release) — every figure on this page is secondhand from one analysis source.