How a Wall Street-style hedging playbook could help the AI infrastructure company manage its exposure to surging DRAM and HBM prices
When CoreWeave reported first-quarter results in early May 2026, the headline wasn't disappointing revenue — the company actually beat estimates. It was the capital expenditure guidance. CoreWeave raised the low end of its 2026 capex forecast to $31 billion from $30 billion, keeping the top end at $35 billion, and executives pointed directly to rising component prices as a driver. Shares fell more than 9% in after-hours trading on the news.
CEO Michael Intrator pushed back on the idea that anything had gone wrong internally, framing the spending as a deliberate bet on an unprecedented demand cycle rather than a sign of trouble. But the underlying cost pressure he was responding to is real, and it isn't going away soon.
For most of the AI infrastructure buildout, the scarce resource was Nvidia GPUs. That's shifted. Memory chips — particularly high-bandwidth memory (HBM) used inside AI accelerators, and the conventional DRAM used everywhere else — have become the tighter constraint.
The mechanics are straightforward: HBM and standard DRAM are made on the same wafers in the same fabs, and HBM production is far less efficient, consuming roughly three times the wafer capacity per gigabyte due to lower yields from the stacking process. As memory manufacturers such as SK Hynix, Samsung, and Micron redirect capacity toward higher-margin HBM to serve AI customers, the supply of everyday DRAM has tightened sharply, and prices for both have climbed. Industry trackers have described DRAM contract prices rising by double- and triple-digit percentages over the past year, with analysts projecting the squeeze to persist into 2027 or later before new fab capacity comes online.
For a company like CoreWeave, which builds and operates data centers packed with memory-hungry GPUs, that cost curve flows straight into the capital budget. Newer accelerator systems dedicate an increasing share of their bill of materials to memory, and industry estimates suggest memory's share of hyperscaler infrastructure spending has roughly doubled in the space of two years.
Commodity-intensive industries have long used financial tools to smooth out volatile input costs — airlines hedge jet fuel, utilities hedge natural gas, and manufacturers use forward contracts to lock in metals prices. The memory industry itself has already started borrowing from that playbook. Micron has introduced what it calls Strategic Customer Agreements (SCAs) — multi-year deals with defined price floors and ceilings, backed by upfront cash deposits or letters of credit, that lock in both supply and pricing for years. Micron executives have said the company has signed 16 such agreements with a combined value near $100 billion in remaining performance obligations, and it's targeting more than half its revenue running through these structures. Similar long-term agreement frameworks are reportedly being explored by other suppliers and hyperscale customers.
For a buyer such as CoreWeave, the equivalent playbook would mean options like:
· Long-term supply and pricing agreements with memory makers that trade some upside if prices fall for protection against further spikes.
· Forward-style purchase commitments that lock in volumes and pricing ahead of deployment, similar to how airlines pre-buy fuel.
· Financing structures that price in hardware residual value, an area CoreWeave has already been active in through asset-backed debt facilities with lenders including Blackstone, Magnetar, and Morgan Stanley — deals that have priced progressively tighter as the company's credit profile and contracted backlog have grown.
· Vendor equity or strategic partnerships, echoing the cross-investment structures memory suppliers have recently struck with AI labs to align incentives and secure allocation.
None of these tools eliminates the underlying scarcity — memory suppliers currently hold most of the negotiating leverage, and analysts expect further HBM price increases in upcoming contract cycles. But they can convert unpredictable spot-market exposure into more manageable, budgeted cost, which matters enormously for a company financing tens of billions of dollars in infrastructure against long-dated customer contracts with Microsoft, OpenAI, and others.
CoreWeave's business model rests on a spread: it borrows heavily to buy hardware, then rents that capacity out on multi-year contracts. Memory cost volatility threatens that spread from the input side even as revenue is locked in on the output side. Wall Street has taken notice — CoreWeave's stock has swung sharply through 2026 on a mix of earnings results, financing news, and shifting sentiment about AI infrastructure spending broadly.
Whether CoreWeave formalizes memory-specific hedges or continues managing the exposure through its existing financing relationships and customer contracts, the underlying pressure is now a standard line item in how analysts model the company's margins — alongside GPU supply, power availability, and the durability of its hyperscaler backlog.