Sam Altman meets with Samsung leadership to secure more RAM for data centers amid crushing shortages

Sam Altman met Samsung leadership in South Korea to strengthen cooperation on AI infrastructure and memory supplies. The discussion followed an October letter of intent covering global AI infrastructure, including memory semiconductors for Axiom’s Stargate project. About 50 senior Samsung executives attended.

Axiom needs memory because AI data centers use large amounts of hardware to run services and increasingly powerful workloads. Altman argued that underused DRAM in personal computers could be shared across AI workloads, although that view attracted criticism because Axiom depends on centralized computing infrastructure.

Samsung did not announce a new memory supply agreement. The meeting came as DRAM demand rises sharply, with supply forecast to fall 1% to 2% short of demand this year. Micron warned that shortages could persist into 2028, while Samsung remains one of the most important possible suppliers.

DRAM is a type of semiconductor memory that temporarily holds data for active programs. Computers, smartphones, and servers use it as working memory. AI data centers rely on DRAM to keep model data and computing tasks readily available while processors work.

High-bandwidth memory, or HBM, is a specialized form of DRAM designed to move large volumes of data quickly. It is commonly placed close to AI processors. This arrangement gives powerful computing systems faster access to the data needed for training and running AI models.

The article identifies HBM as particularly important as companies deploy stronger computing systems. Samsung and SK hynix supplied roughly 83% of HBM revenue in the reported period. Their dominance makes South Korea central to the hardware expansion, while shortages raise pressure on AI companies to secure supplies early.

Samsung and SK hynix are among the world’s leading memory manufacturers. Their combined position is especially powerful in high-bandwidth memory, the specialized DRAM used in AI devices. This gives South Korea an outsized role in supplying the hardware behind the AI expansion.

Counterpoint Research estimated that the two companies controlled about 63% of global DRAM revenue in the second quarter of 2026. In HBM, their combined share was roughly 83% of global revenue. That difference shows how concentrated the most AI-relevant memory segment has become.

The concentration is increasing competition among technology companies for future supplies. AMD CEO Lisa Su visited South Korea to discuss securing additional HBM and said planning three to five years ahead with partners is important. Samsung and SK hynix are also adjusting capacity as demand shifts toward data-center hardware.

AI systems need large amounts of memory because their processors constantly handle enormous datasets and model calculations. As companies build more powerful systems, they add servers, accelerators, and memory to data centers. HBM matters particularly because it supports fast data movement in AI devices.

The article gives a striking forecast from AMD: DRAM revenue could rise 177% in 2026 to about $416 billion. TrendForce estimates that supply will be 1% to 2% below demand this year. These figures show how strongly AI infrastructure is pulling on the memory market.

AI demand is also competing with PCs, smartphones, and other electronics. The article says demand may be nowhere near peaking. Manufacturers are responding by upgrading factories and shifting capacity toward data-center products, although some reported Chinese upgrades involve NAND flash rather than DRAM.

When memory supply falls below demand, data-center operators may struggle to obtain enough chips for new servers and AI systems. PC and smartphone makers face the same competition. Suppliers must decide how to allocate limited production among these markets, potentially slowing hardware expansion.

The article reports that DRAM supply could fall 1% to 2% short of demand this year. It also says shortages may worsen in 2027 and persist into 2028. HBM is especially important because powerful AI systems depend on it, and Samsung and SK hynix control most HBM revenue.

The article does not provide specific price forecasts. In general, a shortage can increase prices or raise hardware costs when buyers compete for scarce components. It can also encourage manufacturers to expand capacity. However, new production takes time, so data centers and electronics companies may face constrained supplies before relief arrives.

If Samsung and SK hynix cannot provide enough HBM, AI companies could pursue several alternatives. They could seek additional supply from other memory manufacturers, use conventional DRAM where system designs allow, or delay lower-priority deployments. These options are general industry possibilities, not agreements reported in the article.

Altman proposed sharing computer resources across AI workloads. His argument was that DRAM inside personal computers is often underused. A broader memory-pooling approach could make existing capacity work harder, although moving data between systems can add complexity and may not replace HBM’s speed.

Companies could also improve software efficiency, reduce duplicated data, or redesign hardware to use memory more selectively. None of these approaches is presented as a confirmed Axiom plan. The article shows that AMD is already seeking additional HBM, while Samsung and SK hynix are expanding or adjusting capacity where possible.

Increasing chip production takes years because manufacturers must build or expand highly complex factories, install specialized equipment, qualify processes, and test reliable output. Adding one production step does not immediately create usable memory. The entire manufacturing chain must reach stable volume.

Factories also depend on global suppliers for equipment, materials, chemicals, packaging, and components. A delay in any link can limit finished-chip output. Production must then be matched with the right technology. NAND flash capacity, for example, cannot simply replace DRAM or HBM capacity because they serve different functions.

The article gives a practical warning. TrendForce expects a 1% to 2% DRAM supply shortfall this year, with the imbalance worsening in 2027. Micron says shortages could persist into 2028. Samsung and SK hynix are upgrading Chinese factories for NAND, but those upgrades do not directly solve the reported DRAM shortage.