CoreWeave Wants to Hedge DRAM Like Airlines Hedged Jet Fuel. The Airlines Lost $519 Million and Quit.

Four point seven billion dollars. That is the conservative estimate of CoreWeave's annual memory chip exposure, derived from the company's $31 billion capital expenditure guidance and industry benchmarks placing DRAM and NAND at 15 to 20 percent of GPU-heavy data center builds. To secure supply, CoreWeave locked in long-term agreements with Micron and SanDisk that guarantee chipmakers a price floor, a smart move during a shortage that becomes a dangerous bet if prices crash.

Reuters reports CoreWeave is now exploring put options on semiconductor stocks to protect that downside, not DRAM futures, which do not exist on any liquid exchange, but equity puts on the shares of memory chipmakers like Micron and SanDisk, because nothing else trades at the scale a $31 billion capex budget requires.

This strategy has a name: cross-hedging, and it has a history that should make CoreWeave's board nervous.

Southwest Saved $3.5 Billion. Then Everything Broke.

From 2004 to 2008, Southwest Airlines ran the most celebrated commodity hedge in corporate history, locking in jet fuel contracts at roughly $51 per barrel while the crude oil market climbed past $140 and competitors grounded planes they could no longer afford to fuel. The savings over those four years reached an estimated $3.5 billion.

The catch was structural: no liquid jet fuel futures market existed at the time, so Southwest cross-hedged using West Texas Intermediate crude oil derivatives, a strategy that worked only as long as crude and jet fuel prices moved in tandem.

Then the correlation snapped, and by 2009, United Airlines posted $519 million in fuel-hedge losses, a figure large enough on its own to push the carrier into a $779 million net loss, while Continental booked $63 million in hedge losses and Northwest's hedge book helped send it into Delta's arms. Southwest's 10-K conceded the failure directly: "The correlation between WTI crude oil prices and jet fuel prices during recent periods has not been as strong as in the past." Premiums alone cost Southwest $150 million per year.

In 2025, Southwest killed the program entirely, and by early 2026, not one major U.S. airline maintained an active fuel hedge, so when Iran pushed jet fuel to $4.12 a gallon, every carrier absorbed the spike unprotected.

The technical cause of death: basis risk.

CoreWeave's Basis Risk Is Structurally Worse

Basis risk measures the gap between the asset you need to protect and the proxy you actually trade to protect it, and in the airline case, it was the divergence between jet fuel and crude oil, two commodities linked by the same refining chain, with historical correlations running between 0.85 and 0.95. Academic research published in the Journal of Risk and Financial Management found that even at those correlation levels, fuel hedging had no statistically significant effect on airline operating costs over the long run.

CoreWeave's proposed hedge is categorically worse because it jumps across asset classes altogether, substituting a commodity instrument, DRAM contract pricing, with an equity instrument in Micron stock, and when those two diverge, the hedge fails in exactly the scenario it was designed to protect against.

Micron's share price does not track DRAM spot prices; it prices a discounted cash flow incorporating NAND, HBM, and storage revenues alongside manufacturing margins across all product lines, analyst sentiment, market beta, and whatever narrative premium Wall Street assigns to memory stocks on any given Thursday. Recent history proves the mismatch in both directions: during the 2022-2023 downcycle, global DRAM revenue plunged 28.9 percent in a single quarter while Micron stock fell roughly 50 percent peak-to-trough, amplifying the commodity decline because equity markets front-run expectations with a velocity that commodities lack. In the subsequent recovery, DRAM contract prices tripled by mid-2026 while Micron merely doubled, lagging because investors remained skeptical that a cycle that produced a 7.7 trillion won loss at SK Hynix would not repeat.

Stocks amplify on the downside and lag on the upside, which means a hedge built on equity puts overreacts when you need precision and underreacts when you need magnitude, the opposite of what hedging is supposed to accomplish and a pattern that transforms risk management into a directional bet with a structural tilt.

The Numbers

CoreWeave's Q1 2026 filing showed $716 million in cost of revenue against $2.08 billion in revenue, a net loss of $740 million, and trailing margins of negative 22.8 percent, a financial profile where an unexpected $1 billion hit to input costs could become existential. The $31 billion capex, raised specifically because of rising component costs, puts annual memory procurement somewhere in the $4.7 to $6.2 billion range.

DRAM is cyclical with a brutality few commodity markets can match: in 2019, average selling prices dropped 44 percent in twelve months, and in Q3 2022, quarterly revenue across the entire industry dropped 29 percent as memory makers bled red ink for four consecutive quarters before the AI boom pulled prices back up threefold. If prices correct to $5 per gigabyte from today's $19 to $20, CoreWeave's floor contracts lock it into paying $7.80, a floor derived from leaked long-term agreement terms, while competitors procure at spot. On $5 billion of annual procurement, that spread is roughly $1.8 billion a year, a figure exceeding the company's entire trailing twelve-month net loss of $1.2 billion.

Delta Air Lines found a physical solution to the same problem, buying a 185,000-barrel-per-day refinery in Trainer, Pennsylvania for $150 million in 2012 and producing its own jet fuel, which eliminated basis risk at the source by removing the proxy instrument entirely. CoreWeave cannot buy a memory fab.

Strongest Counterargument

CoreWeave may be making a rational decision despite the structural problems, because the DRAM market is cyclical and the current supercycle has every hallmark of a peak: prices tripled in 18 months, long-term agreements are being signed at near-record ceilings, and new fabrication capacity from SK Hynix M15X and Micron's Idaho expansion is expected to come online in early 2028 with enough volume to tip the supply-demand balance. Doing nothing and absorbing a $1.8 billion annual overpayment during a correction could be fatal at negative margins. Even an imperfect equity hedge capturing half the downside saves $900 million. The problem is not the instinct to hedge; the problem is that the only available instruments carry basis risk that the airline industry already proved was lethal.

Limitations

CoreWeave does not disclose memory as a component of capital expenditures; the 15 to 20 percent figure derives from industry benchmarks for GPU-accelerated builds. Specific contract terms with Micron and SanDisk, including floor prices and durations, are not public; the $7.80 per gigabyte floor comes from separately reported long-term agreement data, not from CoreWeave disclosures. The DRAM-to-equity correlation analysis uses directional price data, not formal regression. Historical cycle magnitudes may not predict future corrections if structural AI demand decouples memory from its traditional commodity pattern.

What This Means

CoreWeave's hedge exploration signals that AI cloud economics have matured past the build-fast-worry-later phase, into a regime where memory is no longer a procurement detail but a board-level financial risk comparable in scale and cyclicality to jet fuel for airlines or iron ore for steel mills. The pattern repeats across industries: a commodity becomes critical, prices spike, long-term contracts get signed at inflated levels, prices threaten to fall, management reaches for derivatives to protect the exposure, and basis risk eats the hedge because the available instruments cannot track the underlying asset with enough precision to justify the premiums. Southwest spent two decades and billions of dollars completing that sequence. CoreWeave is at step five.