Oracle and CoreWeave both saw their shares drop in late April 2026, punished by investors who looked past headline-grabbing deals and focused instead on a harder question: whether the enormous sums being poured into AI infrastructure will pay off before the debt comes due.
The timing made the sell-off especially striking. CoreWeave had just disclosed an expanded agreement with OpenAI worth up to $6.5 billion. Oracle had just completed an $18 billion corporate bond sale to fund AI data center construction. (Bloomberg initially reported the planned raise at $15 billion in September 2025; the final offering was upsized to $18 billion, according to subsequent company disclosures.) Neither announcement produced a rally. Instead, both stocks fell.
Massive bets, thin margins for error
The capital at stake is staggering even by Big Tech standards. Oracle’s bond sale loads its balance sheet with debt at a time when the Federal Reserve has kept interest rates well above pre-pandemic levels, making the cost of that borrowing meaningfully higher than it would have been just a few years ago. CoreWeave’s expanded OpenAI contract, meanwhile, ties the company’s near-term financial trajectory to the spending decisions of a single customer.
That “up to $6.5 billion” figure deserves scrutiny. The phrasing signals a ceiling, not a guarantee. Payment timelines, performance benchmarks, and cancellation terms have not been made public. For investors trying to model CoreWeave’s future cash flows, the contract value tells them what is theoretically possible, not what the company will actually collect.
CoreWeave has also been on an acquisition tear that adds both capability and cost. The company completed its purchase of Weights and Biases, a widely used platform for tracking AI model development. It agreed to acquire reinforcement-learning startup OpenPipe, launched a ventures arm targeting early-stage AI companies, and committed to a major expansion in the United Kingdom. Each move broadens CoreWeave’s footprint, but each also adds complexity at a moment when investors are already questioning the pace of spending.
CoreWeave’s one-customer problem
CoreWeave went public in March 2025 at $40 per share and has experienced sharp price swings since, a pattern that reflects how sensitive its valuation is to any shift in sentiment around AI infrastructure spending. The expanded OpenAI deal only deepens that sensitivity. A substantial share of CoreWeave’s projected future revenue now flows from a single client.
If OpenAI’s growth continues at its current pace and it keeps routing workloads through CoreWeave, the payoff could be enormous. But OpenAI has not publicly detailed how it plans to split compute demand across providers, and it has alternatives. Microsoft, OpenAI’s largest investor and cloud partner through Azure, already handles a significant portion of its infrastructure needs. Amazon Web Services and Google Cloud are competing aggressively for AI workloads as well. CoreWeave’s bet is that it can offer specialized GPU capacity and flexibility that the hyperscalers cannot match. Whether that advantage holds as the giants pour tens of billions into their own AI clusters is an open question.
Oracle’s debt-fueled data center push
Oracle faces a different version of the same dilemma. The company is not overly dependent on one customer, but it is taking on heavy debt to compete in a market where the incumbents have been building cloud and AI infrastructure for over a decade and operate at far greater scale.
No public Oracle filing has broken down how much of the $18 billion in bond proceeds will flow toward OpenAI-related projects specifically versus the company’s broader Oracle Cloud Infrastructure business. That ambiguity leaves investors guessing about Oracle’s true exposure to any single AI customer and the contractual terms governing those relationships. For a company historically known for its database software and enterprise licensing, the pivot to capital-intensive cloud infrastructure represents a fundamental shift in risk profile.
The gap between spending and revenue
The sell-off in Oracle and CoreWeave did not happen in isolation. Across the AI sector, a growing number of investors have started pressing harder on the gap between infrastructure spending and actual revenue generation. Many of the most compute-hungry applications, from frontier large language models to multimodal AI systems, are still in relatively early stages of commercial deployment. OpenAI has scaled its user base rapidly, but the company is also spending aggressively to train and run its models. The company has not disclosed a precise cost figure, though multiple press reports through late 2025 described annualized spending in the billions of dollars, with profitability still years away by the company’s own projections.
For infrastructure providers, the math is uncomfortable. They must build capacity now, before demand is fully proven, and hope that usage and pricing catch up before the bills come due. When borrowing was cheap, the penalty for overbuilding was manageable. With rates still elevated in spring 2026, the margin for error has narrowed considerably.
What could change the calculus for Oracle and CoreWeave
Several developments could shift sentiment in either direction. On the bullish side, detailed financial disclosures from Oracle or CoreWeave breaking out AI-specific revenue, data center utilization rates, or contract renewal metrics would give investors something concrete to model. Evidence of strong early returns on new data centers, such as rising margins or rapid payback periods, would validate the current spending spree. Successful integration of CoreWeave’s acquisitions and tangible traction from its UK expansion could also demonstrate that the company is building a diversified business, not just scaling raw GPU capacity for one client.
On the bearish side, any sign that OpenAI is slowing its infrastructure spend, shifting workloads to competitors, or building more of its own compute capacity would confirm the fears that drove the late April sell-off. Rising financing costs would compound the pressure on balance sheets already stretched by multibillion-dollar commitments.
For now, Oracle and CoreWeave are asking shareholders to trust that the AI infrastructure buildout will generate returns large enough to justify the risk. The market’s response this week suggested that trust, at least at these prices and these debt levels, has its limits.
