OpenAI CEO Sam Altman says DRAM in personal PCs is inefficient versus cloud data centers
The battle for semiconductor supremacy has moved beyond GPUs to the very memory that feeds them, sparking a clash between the vision of the "AI PC" and the needs of massive data centers. As global supply chains struggle with chronic shortages of memory and storage, OpenAI CEO Sam Altman has suggested that the DRAM currently residing in consumer laptops and desktops is a waste of precious resources.
During an appearance on the Cheeky Pint podcast in late October 2026, Altman argued that it is "quite terrible" for the world's limited supply of DRAM to be distributed across millions of personal computers. His primary contention is based on utilization; Altman estimates that memory in personal devices is used only about 5% of the time. From his perspective, this idle capacity is a luxury the industry cannot afford in a supply-constrained environment, suggesting that pooling this memory in cloud data centers would ensure maximum utilization.
This push for centralization is not merely theoretical, as OpenAI is aggressively securing the hardware necessary to build this cloud-centric future. The company has reportedly entered supply arrangements with Samsung and SK Hynix to procure approximately 900,000 DRAM wafers per month, a volume that could account for roughly 40% of global production. This massive acquisition is intended to help OpenAI push past the "memory wall," the technical bottleneck where processors outpace the memory feeding them data. To further optimize this efficiency, OpenAI has developed a custom AI inference chip dubbed "Jalapeño," which the company claims delivers 1.5 to 1.9 times better performance per watt than reference systems.
The Memory Supply Crunch
The scarcity Altman describes is felt most acutely by the manufacturers and consumers caught in the middle. While memory producers are currently seeing massive windfalls—with Altman projecting that some could generate $100 billion in annual free cash flow—they remain hesitant to flood the market with new capacity. This caution stems from a history of boom-and-bust cycles where producers invested billions in capital expenditures only to face bankruptcy after overbuilding fabs.
This industry discipline, combined with the insatiable demand from AI giants, has tangible consequences for the consumer market. Micron Technology, for instance, exited its Crucial consumer memory business in December 2025 to redirect resources toward larger strategic AI customers. The resulting shortage has contributed to rising costs across the board, affecting everything from PC components to gaming consoles. Intel has cited industry-wide memory and substrate shortages as a primary supply risk, while Microsoft has reported a 22% decrease in Windows OEM and Devices revenue, signaling a softening PC market.
Centralization vs. Local Inference
Altman’s comments highlight a growing ideological divide in the tech industry regarding local inference—the ability to run AI models directly on a user's device. While Altman views local DRAM as "hoarded" and underutilized, other industry leaders, including Apple, champion on-device AI. The primary driver for the local camp is privacy, as keeping data on a personal device eliminates the risks associated with transmitting sensitive information to external servers.
Ultimately, these competing visions reflect the divergent incentives of the players involved. A cloud-service provider like OpenAI benefits from a centralized model where users pay for access to powerful, pooled resources. Conversely, hardware manufacturers have a vested interest in making local devices indispensable. If the industry accepts Altman's logic that memory is too precious to leave idle in a home office, the momentum will shift decisively toward a few large cloud providers, further concentrating control over the hardware that powers the AI era.
