Big Tech’s AI memory crunch turns brutal as Microsoft and Google face problems 

Artificial Intelligence

The global race to build AI infrastructure is beginning to expose its weakest link, and it is not compute or software. According to industry analysts, memory supply has become so constrained that some of the world’s largest technology companies are resorting to extreme internal measures to secure it.

A report highlighted by analyst Jukan of Citrini Research suggests that procurement executives from Microsoft and Google have effectively been dispatched to South Korea with a blunt mandate: lock in additional memory capacity from Samsung Electronics and SK Hynix, or risk being removed from their roles.

The pressure is reportedly not hypothetical. Korean industry sources cited in the analysis claim Google has already dismissed procurement executives after failing to secure additional memory supply once demand for its AI accelerators exceeded internal forecasts. When Google attempted to negotiate more high-bandwidth memory from SK Hynix and Micron, it was reportedly told that no extra capacity was available under any terms.

That response, according to the report, triggered internal fallout. Executives responsible for procurement were blamed for not locking in long-term supply agreements earlier, exposing Google to shortages just as AI workloads surged. The episode has become a cautionary tale inside the industry, reinforcing the idea that memory access is now a board-level risk, not a routine supply-chain concern.

Similar tension is said to be playing out at Microsoft. One reported incident describes Microsoft executives walking out of a meeting with SK Hynix after being told the supplier could not meet Microsoft’s requested conditions. Industry insiders claim that negotiations have become so aggressive that some technology companies are effectively placing open-ended orders, asking suppliers to deliver as much volume as possible regardless of price.

At the heart of the crisis is the explosive demand for high-bandwidth memory and advanced DRAM used in AI accelerators and data centre servers. Only three companies globally can produce these components at scale: SK Hynix, Samsung Electronics and Micron. All three are reportedly sold out of HBM and advanced DRAM capacity through at least the end of next year, leaving buyers with little leverage.

The imbalance has begun reshaping how big tech approaches procurement. Traditionally, memory sourcing was handled by teams based in Silicon Valley or Seattle. Now, companies are increasingly hiring procurement managers with deep engineering backgrounds and stationing them in Korea, Taiwan and Singapore. The goal is to combine technical coordination and commercial negotiation in one role, while maintaining constant on-the-ground presence near suppliers.

The implications extend beyond short-term supply pain. Research firm International Data Corporation has warned that the memory shortage is likely to persist well into 2027. IDC has described the situation as the end of an era of cheap, abundant memory, a structural shift driven by AI workloads that demand far more memory per unit of compute than traditional cloud applications.

For Microsoft and Google, the reported ultimatums reflect a broader reality. In the AI era, access to memory is no longer a commodity advantage. It is a strategic necessity, and failure to secure it can derail multi-billion-dollar infrastructure plans. As the shortage deepens, the pressure is cascading down organisational charts, turning procurement into one of the most high-stakes roles in big tech.

If the analyst reports are accurate, the message from leadership is clear: in the race for AI dominance, excuses are no longer accepted, and even senior executives are expendable when memory runs out.