OpenAI's chief executive, at a closed-door meeting with Samsung executives, unusually admitted that demand for AI infrastructure involves double counting, and proposed jointly developing next-generation AI terminal devices with Samsung.
According to South Korean media, during his visit to Korea at the end of last month, Sam Altman warned roughly 50 division-level executives of Samsung Group that demand for AI memory and data centers could fluctuate in the near term and that double counting of demand exists in the market.
At the same time, Altman extended an olive branch for cooperation at the same event, stating he hoped to jointly develop next-generation AI terminal devices with Samsung, aiming directly at the "post-smartphone" era. This was widely interpreted as a strategic move by OpenAI to join forces with Samsung against Apple's ecosystem dominance.
Short-term demand carries bubble risk
When responding to a Samsung executive's question about whether AI and memory demand would saturate or slow over the next 5 to 10 years, Altman made clear that "there will be ups and downs in the short term."
He specifically criticized this year's bubble controversy surrounding "emerging cloud service providers"—companies that specialize in renting out GPU and other AI compute infrastructure.
Altman pointed out that "double counting of demand exists," and stressed that "only OpenAI and Anthropic can truly drive large-scale demand growth, and a supply shock and a period of instability may arrive."
This statement has direct reference significance for memory chip suppliers such as Samsung Electronics and SK Hynix. The two companies announced in October last year that they would participate in the OpenAI-led "Stargate" project to advance cooperation on advanced memory supply.
Long-term bottlenecks lie in the chip supply chain and data centers
Despite warning about short-term demand, Altman remains highly optimistic about the long-term outlook for AI demand.
He said that as AI is applied to areas such as real-time writing of customized software, robots, spacecraft, and scientific research, the use of "tokens"—the units of information processed by AI models—will rise substantially.
Citing data, he said that in early 2020, the top user by token usage consumed about 100,000 tokens per month, while by early this year the highest-usage user had reached 10 billion tokens per month.
When asked about the biggest bottleneck to AI expansion, Altman listed four factors: the supply chain, fundraising, AI safety and social acceptance, and corporate transformation. He identified the "chip supply chain" and "data center infrastructure" as the most critical constraints over the next several years, calling for "more investment in the supply chain."
Analysts noted that Altman's special trip to Korea amid a busy schedule, with a tight one-night, two-day arrangement to meet Samsung executives, itself confirms the urgency of the shortage in memory chip supply.
The business logic of low margins and large-scale sales
When Samsung executives pressed on whether sustained large-scale investment would bring profit pressure, Altman said bluntly that "this is exactly the core of it," and explained OpenAI's competitive strategy: by building vertically integrated infrastructure at lower cost than any competitor, selling tokens at low margins and in large volumes, and winning through volume.
He said that investment in compute efficiency, including self-developed chips, has begun to show results, and that high operational efficiency is being achieved through long-term contracts, thereby lowering the processing cost per token.
Altman also stressed that "the market size is large enough, and the rules of ordinary commodity industries do not apply here," and that "simply meeting global token demand is enough to build an outstanding industry."
Seeking to build "post-smartphone" devices with Samsung
On the device strategy front, Altman took the initiative at the meeting to propose that he hoped to work closely with Samsung to jointly develop devices.
He described the form of next-generation AI terminals: not centered on applications, but with AI proactively intervening in and managing users' physical and virtual lives while minimizing interference with users.
It is understood that OpenAI is developing an AI terminal device targeting next year, and the most likely form at present is a screenless smart speaker smaller than a palm.
Regarding the currently much-discussed "AI glasses," Altman explicitly said he personally has reservations. "They may be excellent for personal productivity, but when someone wears glasses and looks elsewhere during a conversation, it feels like technology has intervened in the relationship between people, and I feel uncomfortable."
Strongly urging Samsung executives toward "full AI transformation"
At the meeting, Altman also issued a strong call to Samsung executives, advising Samsung to "go all in on AI" and drive full AI transformation across the company.
He pointed out that if AI is used only as a tool to improve employee efficiency, productivity gains are about 2 to 3 times; but if the entire business process is led by AI, productivity gains can reach 100 times. Citing OpenAI itself as an example, he said the company had fully introduced AI across all businesses and services three months ago, and employees only need to input work priorities into AI, which can then handle them autonomously by integrating context.
Using the mobile internet era as a warning, Altman said: "Companies that acted first in the mobile era survived, while most that were one or two years late were eliminated. In 3 to 5 years, companies not based on AI will be unable to remain competitive."
Comments