Oracle has laid out plans to expand its AI compute capacity roughly tenfold, from about 1 gigawatt in fiscal 2025 to 10 gigawatts by the end of fiscal 2028, and says it can fund most of the buildout without heavy new borrowing.

The database and cloud company presented the roadmap at its October 2026 investor day, which also put a spotlight on a rapidly swelling contract backlog driven by artificial intelligence customers. Because Oracle's fiscal year ends in May, the FY28 deadline lands in the middle of calendar 2028 — less than two years after the plan was laid out.

A tenfold expansion in three fiscal years

According to the targets unveiled at the investor day, Oracle aims for approximately 10 GW of total AI data center capacity by the end of fiscal 2028. Around 7 GW of that is expected to come from incremental additions over the coming years, while the remainder reflects capacity the company already has in place or under way.

Gigawatts have become the industry's standard yardstick for AI infrastructure because a data center's useful output is capped by the electricity it can draw to run its chips — which is why power access, not chip supply alone, increasingly sets the pace of deployment across the sector.

The timing is significant because Oracle is approaching the most expensive stretch of the plan. The company expects capital spending to peak over the next two fiscal years.

Backlog surges on AI contracts

Oracle's remaining performance obligations, or RPO, reached $664 billion by the end of the first quarter of FY27. RPO, a standard measure cloud and software companies use to signal future revenue, captures contracted revenue that customers have committed to but that Oracle has not yet booked. The figure grew 3 to 4-fold year over year, with the company attributing the surge to AI contracts.

Deployment momentum accelerated as well. In Q1 FY27, Oracle brought 850 MW of capacity online, including more than 300,000 GPUs, across 38 projects in 11 countries. The previous quarter, the company delivered 400 MW, meaning output more than doubled in a single quarter.

Some quick arithmetic shows why that pace matters. At 850 MW per quarter, adding roughly 7 GW would take a little over eight quarters, which lines up closely with the FY28 window.

How Oracle plans to pay for it

The company had previously announced that it secured more than 10 GW of power and data center capacity for the next three years. Most of that buildout is financed through partnerships, customer pre-payments, or bring-your-own-hardware arrangements. In a bring-your-own-hardware deal, the customer supplies the chips, and Oracle provides the building, power, and operations around them.

Oracle projects that capital expenditures will peak during FY27 and FY28. Free cash flow is expected to turn positive in the second half of FY28, and the company expects it to exceed $30 billion by FY29.

On the financing front, Oracle estimates it will need at most $20 billion in additional external financing through FY30. Vendor financing, customer contributions, and pre-payments are intended to cover the remainder.

From software vendor to infrastructure landlord

The plan marks a shift away from Oracle's traditional identity as an enterprise software company and toward hyperscale data center operations, with AI at the center. Its role in high-profile collaborations such as the Stargate project with OpenAI — the large-scale AI infrastructure initiative Oracle joined alongside SoftBank in early 2025 — places it among the key suppliers of GPU and CPU resources for large AI workloads. That pivot also puts Oracle in direct competition for power, land, and chips with larger cloud rivals including Microsoft, Amazon, and Google, in a race where electricity access increasingly decides who can serve AI demand.

The company's investor relations messaging points to favorable utilization rates and growing multi-service spending from AI-focused customers as the engine for future self-funding.

Execution now the central question

For investors, the story now rests on execution more than demand. The $664 billion backlog indicates that customers want the capacity; the open question is whether Oracle can deliver it fast enough to convert those commitments into recognized revenue.

The constraints are physical rather than financial. Oracle still has to navigate power procurement and infrastructure build timelines — the same bottlenecks now shaping buildout plans across the AI data center industry.

Several checkpoints stand out: whether Oracle holds or improves on its 850 MW quarterly delivery pace; whether capex actually peaks in FY27 and FY28 as projected; and whether free cash flow turns positive in the second half of FY28 while external financing stays within the $20 billion ceiling through FY30.

Source: CryptoBriefing, coverage edited by Diego Almada Lopez.