Oracle's AI Bet Has a $70B Price Tag — and Wall Street Is Spooked

Oracle reported fiscal fourth-quarter results on June 10 that would have sparked a celebration just two years ago: cloud infrastructure revenue surged 93%, total revenue hit a record $19.2 billion, and adjusted earnings per share of $2.11 cleared analyst estimates by more than 7%. The company signed $67 billion in new AI contracts in a single quarter, with four individual customers each committing more than $8 billion. By every measure of demand, the story is extraordinary.

Wall Street hit the sell button anyway.

Oracle shares fell roughly 9% in trading on June 11, erasing more than $100 billion in market capitalization, as investors looked past the headline beats and fixated on a single alarming figure buried in the guidance: Oracle expects to spend approximately $70 billion in net capital expenditures in fiscal year 2027 — more than double what it generated in operating cash flow in FY2026. To fund the gap, the company said it would raise an additional $40 billion through debt and equity sales, on top of the $43 billion it already borrowed in the past year. The total debt load now sits at roughly $117 billion.

The reaction encapsulates the defining tension of the AI infrastructure era: demand is real, urgent, and accelerating. The question the market cannot yet answer is whether it is real enough, and fast enough, to justify the capital being consumed to meet it.

The OCI Growth Engine

Oracle Cloud Infrastructure has become the company’s most remarkable asset, powered almost entirely by AI compute demand. OCI revenue grew 93% year-over-year to $5.8 billion in Q4, and for the full fiscal year it reached $18.1 billion, up 77%. The company’s Remaining Performance Obligations — essentially its contracted-but-not-yet-recognized revenue — rose to $638 billion, up from $553 billion the previous quarter.

Those numbers reflect the scale of the AI infrastructure buildout that Oracle has positioned itself at the center of. The company is the anchor infrastructure partner for Stargate, the joint venture with OpenAI and SoftBank pledging $500 billion in U.S. AI infrastructure investment. Oracle said that 42% of capacity at the flagship Stargate data center in Abilene, Texas — reportedly one of the largest AI facilities ever built — has been delivered, with another 35% expected over the following three months.

GPU utilization across OCI came in at 97.5%, which CEO Safra Catz cited as the clearest proof that supply still lags far behind demand. “The demand is real, the contracts are signed, and the capacity we’re building will generate returns,” Catz told analysts. “We simply cannot build fast enough.”

Total cloud revenue for the quarter, including applications, grew 47% to $9.9 billion, and overall cloud applications revenue rose 10% to $4.1 billion — a more modest pace that reflects the maturity of Oracle’s older SaaS businesses relative to the OCI hypergrowth story.

The Free Cash Flow Problem

The math underneath Oracle’s ambition is what rattled the market. The company generated $32 billion in operating cash flow in FY2026, a 54% improvement year-over-year — an impressive number in isolation. But after spending $55.7 billion on data centers, which exceeded Oracle’s own $50 billion projection by nearly $6 billion, free cash flow came in at negative $23.7 billion for the fiscal year.

For FY2027, the situation is projected to intensify. Oracle has guided to roughly $70 billion in net capex, a figure management acknowledged could reach $90–95 billion on a gross basis once prepayments for hardware components are included. The additional $40 billion fundraising plan — split between $20 billion in share sales and $20 billion in new debt — will dilute existing shareholders and extend an already-stretched balance sheet.

Return on invested capital declined from 14.8% in FY2025 to 10.7% in FY2026, a deterioration that suggests the flood of capital is not yet generating commensurate returns. Oracle moved aggressively to offset some of the cash pressure in March 2026, cutting approximately 30,000 employees — 18% of its global workforce — and redirecting an estimated $8–10 billion in annual labor savings toward data center construction.

Visibility Questions

One of the more striking details in the earnings report is the conversion timeline embedded in that $638 billion backlog. Only 12% is expected to convert to recognized revenue within the next twelve months, meaning the vast majority is committed revenue that Oracle will recognize over many years. At that rate, working through the full backlog would take approximately eight years.

Bears argue that this is evidence of structural mismatch: Oracle is building capacity today at enormous cost, but the revenue from that capacity will arrive slowly. Bulls counter that the backlog represents a de-risked, contracted future, and that conversion rates will accelerate as AI workloads scale and new Stargate nodes come online.

The company’s four-customer concentration risk is also notable. Four individual clients each signed contracts exceeding $8 billion in Q4 alone. OpenAI is the most prominent of these, but the concentration means Oracle’s near-term fortunes are unusually dependent on a small number of AI hyperscalers whose own financial health and capital plans could shift quickly.

The Broader Reckoning

Oracle is not a unique case. Microsoft, Amazon, Google, and Meta have collectively committed hundreds of billions to AI infrastructure in 2026, and each is facing versions of the same capital-intensity question. What makes Oracle’s situation more exposed than its hyperscaler peers is that OCI’s growth is almost entirely AI-specific — it lacks the diversified enterprise base that cushions AWS or Azure if AI demand moderates.

There is also a structural dependency worth noting: Oracle’s largest customers are themselves raising enormous sums to pay for compute. OpenAI’s pending IPO, expected to close later this summer at a valuation above $1 trillion, is partly a capital raise to fund ongoing AI training and inference costs. The circular dynamic — where AI companies raise equity to pay infrastructure costs, and infrastructure companies raise equity to serve AI companies — creates a system that works brilliantly when confidence is high and becomes fragile if any link in the chain falters.

Wedbush Securities analyst Dan Ives maintained an Outperform rating on ORCL following the report but acknowledged the market’s reaction. “Oracle is building the foundation for a generational AI infrastructure platform, but the capital intensity required in the transition period is genuinely challenging,” he wrote. “Wall Street needs to see proof that the backlog converts into actual billings on a faster timeline.”

What Comes Next

For Q1 FY2027 (ending August 2026), Oracle guided to total revenue growth of 17–19% year-over-year, with OCI expected to continue growing at or above 90%. Those are still remarkable numbers for an enterprise tech company of Oracle’s scale.

The monitoring question for investors over the next four quarters is direct: does the $638 billion backlog begin to accelerate its conversion rate? If enterprise AI adoption continues at the pace Oracle’s customer roster suggests, that timeline compresses significantly. If spending normalizes or a large customer hits financial turbulence, the arithmetic becomes uncomfortable.

For now, Oracle enters FY2027 as perhaps the most important single witness in the ongoing question about the AI infrastructure buildout: whether the largest capital investment cycle in enterprise computing history will be remembered as the foundational infrastructure of the AI era, or as the most expensive industrial bet ever placed on a forecast that proved, at least in timing, to be wrong.