Oracle's AI Backlog Couldn't Stop a Post-Earnings Selloff

On June 10, Oracle reported a quarter that would have been a victory in any other era. Revenue hit $19.18 billion, beating consensus. Adjusted earnings came in at $2.03 per share against expectations of $1.96. Cloud infrastructure revenue surged 93% to $5.8 billion. Remaining performance obligations — contracted revenue not yet recognized — climbed to $638 billion, up 363% year over year. CEO Clay Magouyrk told analysts Oracle would bring nearly one gigawatt of computing power online in the current quarter, roughly matching its entire fiscal 2026 deployment. Bank of America analysts noted that more than half that backlog traces to OpenAI. The stock fell 10% in after-hours trading and kept falling. By mid-June, shares had dropped roughly 25% from their early-June levels, trading near $184 against a 52-week high of $345.72. Oracle did not miss. It delivered exactly what the AI trade ordered. The market punished it anyway.

The punishment was not about quarterly numbers. It was about the invoice. Oracle spent $55.7 billion on capital expenditures in fiscal 2026, a 162% increase over the prior year. It reported $23.7 billion in negative free cash flow. Management guided to roughly $70 billion in net capex for fiscal 2027 and said it expects to raise another $40 billion through debt and equity financing — including a previously announced $20 billion at-the-market equity program — on top of the $43 billion in debt and $5 billion in equity it already raised during fiscal 2026. Total debt stood above $153 billion. CFO Hilary Maxson, hired from Schneider Electric during the quarter, framed customer prepayments and GPU-supply arrangements as mitigating factors. Investors heard dilution, leverage, and a bet that AI demand will remain strong long enough to convert a $638 billion backlog into cash — a bet Oracle's balance sheet is less equipped to carry than Amazon, Microsoft, or Google.

This is the tension the AI IPO season is about to price. OpenAI filed confidentially on June 8. Anthropic filed a week earlier. SpaceX has already hit public markets. Private valuations stretch toward $850 billion for OpenAI, $965 billion for Anthropic, and trillions for Musk's empire. Carnelian Asset Management founder Manoj Bahety, writing in June, put the multiples in plain terms: SpaceX at 93 times sales, OpenAI at 35 times, Anthropic at 21 times. "My problem isn't the AI technology," he told ET Markets. "It's the price tag." Oracle is not a startup burning venture capital. It is a 47-year-old enterprise software company that just told shareholders it must issue equity to finance another company's infrastructure buildout — and the equity market said no thank you at the prior price.

Tom Essaye, founder of Sevens Report Research, connected the dots in a note published June 18. Oracle's selloff, he argued, is not an isolated earnings miss but an early signal that the AI infrastructure cycle could stall the way the dot-com buildout did in 2000. "While people connected to the internet, their connection wasn't nearly as profitable as quickly as everyone assumed," Essaye wrote. "Because of that, the buildout stopped." His scenario is simple: if a hyperscaler cancels ten data centers because returns fail to materialize, order cancellations cascade through Nvidia, Micron, Broadcom, and SanDisk — the same supply chain that has driven the S&P 500 for two years. Essaye noted that Nvidia trades at roughly 21 times forward earnings, Micron at 10 times, Broadcom at 24 times — cheap for companies that have delivered enormous stock gains. Cheap, he argues, means skeptical. Markets are no longer paying for the dream. They want the math.

Oracle's math is both impressive and uncomfortable. The demand signal is real: OCI revenue growing 93%, RPO exploding, customers prepaying for GPU capacity. OpenAI's dependence on Oracle infrastructure is not speculative — it is contractual. Related Digital and Blackstone secured $16 billion in funding for a Michigan data center site during the quarter. Oracle maintained fiscal 2027 revenue guidance of $90 billion and raised its adjusted EPS forecast to $8.05. TipRanks analysts called the post-earnings selloff a "knee-jerk reaction" that "misses" the strength of AI demand. Bull cases are not absurd. But Oracle is among the most leveraged names in enterprise software, and it is asking shareholders to fund a capex cycle that hyperscaler peers finance from cash flow. When a company beats earnings and raises guidance while its stock drops double digits, the market is repricing risk, not revenue.

The broader market is sending the same signal in softer terms. Yahoo Finance reported in June that analysts see 2026 as the year AI must "foot the bill" after 2025's euphoric run — with concerns about circular financing, where investors fund companies whose revenues flow back to the same ecosystem. Oracle's backlog is not circular on paper; customers are prepaying for real capacity. But the ecosystem rhymes: OpenAI needs Oracle data centers, Oracle needs capital markets, capital markets need OpenAI's IPO to justify the valuation stack. David Roche of Quantum Strategy warned on ET Markets June 19 that AI investment has become "excessive" and "unsustainable," even as the Federal Reserve's inflation fight supports risk appetite. The Magnificent Seven carried the S&P 500 to a 17% gain in 2025. Essaye's research describes the AI enthusiasm as "fractured" — a sorting between winners like Micron, up over 240% year to date, and former darlings like Oracle, punished for spending.

For enterprise buyers, Oracle's selloff is a procurement signal disguised as a Wall Street story. The company remains a credible cloud infrastructure vendor with explosive AI demand behind it. But vendor financial risk is now part of the stack: if your model provider's infrastructure partner cannot finance its buildout without diluting shareholders, capacity commitments, pricing, and delivery timelines become contingent on capital markets that are losing patience. The same week Oracle reported, 42 state attorneys general subpoenaed OpenAI over model behavior ahead of its IPO. Anthropic faced export-control enforcement that took frontier models offline. Washington is simultaneously accelerating AI through executive order and throttling it through trade controls. The industry is asking public markets to fund a dual-use infrastructure buildout while regulators, litigators, and bond markets ask whether the returns justify the risk.

Convina's view: Oracle's crash is the AI trade's first honest moment. For two years, the market rewarded any company that announced AI capex — the bigger the spend, the higher the multiple. Oracle followed the script: beat earnings, grow backlog, pledge gigawatts, raise capital. Shareholders responded by cutting the stock's value nearly in half from its peak. That is not a rejection of AI. It is a rejection of AI economics at any price. The infrastructure buildout is real, the demand is contracted, and the technology works. But the financing model — debt, equity, prepayments, and circular ecosystem bets — has a shelf life, and Wall Street just started reading the expiration date. Every enterprise betting on cheap, abundant AI compute should watch Oracle's chart as closely as OpenAI's benchmark scores. The bottleneck may not be chips or power. It may be whether investors will keep paying for the buildout after the IPO roadshows end and the invoices arrive.