Key Takeaways
A scramble to secure the memory chips essential for artificial intelligence has led technology giants to propose paying billions upfront for production capacity, an unprecedented step in the historically cyclical semiconductor industry.
SK Hynix Inc., the leading producer of the high-bandwidth memory (HBM) used in AI accelerators, is fielding extraordinary offers from large customers to co-invest in new production lines and even help purchase extreme ultraviolet (EUV) lithography machines from ASML, according to a May 8 Reuters report citing six sources familiar with the matter. The chipmaker, which holds a dominant position in the HBM market, has seen its supply for 2024 and 2025 completely sold out.
"Whether it's one proposal or another, there's basically zero available capacity right now," one source told Reuters, highlighting the severity of the supply crunch. "There isn't even a small portion that can be allocated to a specific customer."
The proposals include funding dedicated production lines for a single customer and providing capital for SK Hynix's new $15 billion fab complex in Yongin, South Korea. This approach is a radical departure from the industry's traditional model of quarterly price negotiations and reflects the desperation of companies like Microsoft and Meta Platforms to secure a supply of the critical component. Microsoft has said its capital expenditures could rise to $190 billion this year, with $25 billion attributed to rising component costs.
For SK Hynix, the customer proposals are a testament to its market leadership but also a strategic dilemma. Accepting direct funding could lock the company into lower-priced, long-term supply agreements, sacrificing future pricing power in a market defined by rapid technological advancement. It would also risk alienating other major customers by appearing to favor one AI developer over another.
"They don't want to bet on a horse in the AI race and find out they bet on the wrong one," another source noted.
In response, SK Hynix and competitors like Samsung Electronics and Micron Technology are exploring new, more binding long-term contracts. These could involve "price range mechanisms" that set annual upper and lower price limits, or require customers to make pre-payments of 30 percent to 40 percent to lock in future supply.
The frenzy to secure hardware comes as the capital required for the AI buildout faces new scrutiny. While hyperscalers are projected to spend between $600 billion and $720 billion on capital projects in 2026, with three-quarters aimed at AI, signs of financial strain are emerging.
A recent report from The Information revealed that OpenAI may be unable to close an $18 billion financing tranche for its custom chip partnership with Broadcom, a deal core to reducing its reliance on Nvidia. This follows reports of Oracle raising $18 billion in bonds to finance its own AI commitments and a sharp rise in Nvidia's own unpaid customer receivables, which now stand near $33 billion. The incidents suggest that even for the most prominent AI players, the massive capital expenditures are becoming harder to fund, potentially tempering the blank-check environment that chipmakers currently enjoy.
This article is for informational purposes only and does not constitute investment advice.
