{"slug": "coreweave-s-long-term-contracts-never-hedged-against-memory-chip-prices", "title": "CoreWeave's Long-Term Contracts Never Hedged Against Memory Chip Prices", "summary": "CoreWeave locked in long-term cloud contracts with Microsoft and OpenAI through 2030 without hedging against memory chip prices, leaving it exposed as DRAM prices tripled. The company now operates on roughly a 1% adjusted EBIT margin, and its stock fell 31% to $81.75 by early July amid margin pressure from rising memory costs.", "body_md": "*CoreWeave locked in what its biggest customers would pay it through 2030. It never locked in what it would pay for the memory chips running its GPUs, and that gap is now the whole story.*\n\nCoreWeave signed its marquee cloud contracts, the ones with Microsoft and OpenAI that run through 2030, before DRAM prices tripled. That timing is now the company's biggest problem.\n\nDA Davidson analyst Luria cut his price target on CoreWeave from $175 to $100 in May and moved the stock to Neutral from Buy. The reason wasn't demand. CoreWeave's revenue backlog is enormous. The reason was margin. CoreWeave runs at roughly a 1% adjusted EBIT margin on close to $8 billion in annual revenue, and Luria flagged that its multi-year contracts may not have hedged memory prices for three to five years out. That detail sounds technical until you realize what it actually means. CoreWeave locked in what it gets paid. It never locked in what it pays for the memory going into the racks.\n\nMemory got expensive fast. Market researchers tracking DRAM pricing peg the current run at somewhere between 275% and 300% growth from 2025 through 2027, a spike that dwarfs the roughly 90% increase during the 2017-18 memory super cycle. High-bandwidth memory, the stacked chips that feed Nvidia's GPUs, is already sold out for 2026 and is expected to consume close to a quarter of all DRAM wafer production this year. SK Hynix, Samsung and Micron make it. None of them are discounting.\n\nCoreWeave felt the squeeze directly. In May, the company raised the low end of its 2026 capital expenditure guidance from $30 billion to $31 billion, keeping the top of the range at $35 billion. Chief financial officer Nitin Agrawal told analysts on the earnings call that the increase came from \"increases in component pricing,\" and he described the resulting margin pressure as \"timing-based, not economic,\" meaning the costs are landing on the books before the new capacity they fund starts generating revenue. That's a reasonable explanation for one quarter. It gets harder to make for four or five.\n\n## The market isn't waiting\n\nShares had fallen to $81.75 by early July, down more than 31% over the prior month. The stock took another leg down on July 13 alongside Nebius and IREN after SK Hynix, fresh off its Nasdaq debut, sold off sharply in Seoul on profit-taking. Investors read that as a sign that even the chipmakers pocketing record margins off memory scarcity aren't safe from a correction. If the sellers of memory aren't safe, the buyers of it certainly aren't either.\n\nCoreWeave has tools to buy itself room. It closed an $8.5 billion GPU-backed loan in March, arranged with Blackstone Credit and Insurance, MUFG, Morgan Stanley, Goldman Sachs and JPMorgan, pushing its debt and equity raises over the past year to roughly $28 billion. That financing is collateralized by the chips themselves and by contracted revenue, including a $14.2 billion compute agreement with Meta. It buys time. It doesn't buy cheaper memory.\n\n## What happens next\n\nHere's the thing that actually separates a mature commodity business from an infrastructure story still finding its footing. Mature businesses hedge their inputs. Airlines buy jet fuel futures. Food companies lock in wheat prices years out. CoreWeave, by its own analysts' account, wrote contracts running three to five years without doing the equivalent for the single input now driving its margin story. Whether that changes for the next generation of contracts, and whether Nebius, Lambda and the rest of the neocloud field are quietly doing the same math, is worth watching over the next two quarters of earnings.\n\nNone of this means the AI infrastructure buildout stalls. Demand for GPU capacity still outstrips supply, and CoreWeave's backlog reflects that. But a company running a 1% margin has almost no room to absorb a doubling input cost. Right now, that's exactly the bet CoreWeave is making, whether it meant to or not.\n\n**Also read:** [Hinge's Founder Just Bet $18 Million That Swiping Is Broken](https://startupfortune.com/hinges-founder-just-bet-18-million-that-swiping-is-broken/) • [OpenAI Is Building a Moveable Screen-Free Speaker as Its First Hardware Product](https://startupfortune.com/openai-is-building-a-moveable-screen-free-speaker-as-its-first-hardware-product/) • [Australia is rushing to approve AI data centers before backlash grows](https://startupfortune.com/australia-is-rushing-to-approve-ai-data-centers-before-backlash-grows/)", "url": "https://wpnews.pro/news/coreweave-s-long-term-contracts-never-hedged-against-memory-chip-prices", "canonical_source": "https://startupfortune.com/coreweaves-long-term-contracts-never-hedged-against-memory-chip-prices/", "published_at": "2026-07-15 00:13:55+00:00", "updated_at": "2026-07-15 00:26:52.395145+00:00", "lang": "en", "topics": ["ai-infrastructure", "ai-chips", "ai-startups"], "entities": ["CoreWeave", "Microsoft", "OpenAI", "DA Davidson", "Luria", "SK Hynix", "Nvidia", "Blackstone"], "alternates": {"html": "https://wpnews.pro/news/coreweave-s-long-term-contracts-never-hedged-against-memory-chip-prices", "markdown": "https://wpnews.pro/news/coreweave-s-long-term-contracts-never-hedged-against-memory-chip-prices.md", "text": "https://wpnews.pro/news/coreweave-s-long-term-contracts-never-hedged-against-memory-chip-prices.txt", "jsonld": "https://wpnews.pro/news/coreweave-s-long-term-contracts-never-hedged-against-memory-chip-prices.jsonld"}}