Elon Musk said that the memory demand for robots will outstrip that of all other devices, driving a sharp surge in the storage sector.

On local time October 27, Elon Musk, CEO of Tesla, stated on a social platform that robots' demand for memory will exceed that of all other things on Earth. This view not only once again refreshed the market's imagination of the computing power ceiling for "embodied intelligence", but also quickly spilled over to the secondary market, triggering a strong rally in the semiconductor storage sector.

Musk made this statement by directly citing and refuting the calculation of DRAM memory demand for humanoid robots released by Counterpoint Research on the social platform.

The institution previously released a research report forecasting that the total global DRAM demand from humanoid robots will increase from 8,300TB in 2026 to 172,000TB in 2030, representing an approximately 20-fold increase in 5 years, with the average memory capacity per unit rising from 19GB to 39GB. Some market participants made deductions based on this data and judged that the incremental demand brought by robots is relatively limited after comparing it with the current total annual global DRAM output.

Musk directly described the calculation as "Silly analysis" in his reply, and then put forward the assertion that the memory demand of humanoid robots will exceed that of all other devices on Earth. The reason is that humanoid robots need to perceive and complete physical operations in the real physical environment, and highly rely on local memory to support their operation. He added that although superintelligence has higher computing power demand, most of such computing power will be placed in space in the future, and the core demand of hardware on the Earth side will focus on humanoid robots.

As the news spread, on October 28, the A-share storage sector continued to rise from the morning session to the afternoon, with stocks in the sector rising in a differentiated pattern. Trading quotes show that Junzheng shares hit a 20% daily limit during intraday trading, closing up 18.37%; Longsys' increase exceeded 15% and closed up 12.01%; Puyuan Semiconductor, Biwin Storage, GigaDevice, Hengshuo Co., Ltd., Demingli and other targets rose in tandem, while supporting storage manufacturers such as Shannon Systems, Montage Technology and Dapuwei also attracted capital attention.

The humanoid robot sector also moved upward simultaneously, with the sector index rising in fluctuations, driving the index of the Tianhong Robot ETF (159770) to rise by more than 5% at one point, with a cumulative increase of more than 12% since October 19. Capital spread along the segmented industrial chains of complete robot machines, servos, sensors and controllers, and the main market trading line has partially shifted from the traditional cloud-side AI computing power to the end-side physical AI hardware supply chain.

Strategic comments from many brokerages mentioned that the "AI computing power chain" that was popular in the market before is highly tied to HBM (High Bandwidth Memory) dedicated to data centers, while the "physical world AI" pointed out by Musk has brought low-power DRAM (such as LPDDR) into the spotlight of long-term demand. Combined with the bottom recovery of the storage industry cycle and the sufficient previous valuation adjustment of the sector, the sector has the conditions for resonant rise.

With the opening of such demand expectation, it will also benefit the share prices of overseas giants such as SK Hynix and Micron Technology.

In addition to Musk, many technology leaders have recently put forward similar predictions on the "explosive growth of storage/computing power demand brought by AI and embodied intelligence".

Last month, Sanjay Mehrotra, CEO of Micron Technology, stated on the earnings call that with the simultaneous increase in the number of devices and the on-board storage capacity per unit, physical AI is expected to become an extremely important driving force for memory and storage demand by the end of this decade. Autonomous driving is the first landing scenario for physical AI, and humanoid robots will also have similar large-capacity memory demand later.

Jensen Huang also defined physical AI as the next wave, arguing that memory bandwidth is the biggest "physical wall" hindering the thinking speed of AI.

Kwak Noh-Jung, CEO of SK Hynix, also expounded the reconstruction logic of the AI wave on the storage industry at many international semiconductor summits and press conferences. He pointed out that the advent of the AI era has completely reconstructed storage chips from the traditional "standardized bulk commodity" to a "core customized carrier" closely bound to specific computing power architectures. Driven by both end-side embodied intelligence and cloud-side AI models, storage bandwidth and local capacity have directly determined the reasoning response speed of AI, which in turn triggers a long-term explosion of global memory demand and an extremely long cycle of supply shortage in the industry.

Right before this round of sector market rally, JPMorgan Chase released a research report on October 25, significantly raising its expectations for Unitree Robotics, upgrading the company's rating by two levels from underweight to overweight, and raising its target price for December 2027 from 300 yuan to 995 yuan. The bank kept its 2026 humanoid robot sales forecast unchanged, but raised its 2027 sales forecast from 26,226 units to 31,701 units, and significantly raised its 2028 sales forecast from 41,552 units to 111,554 units.

The core of the research report's adjustment of judgment comes from the statement of Wang Xingxing, founder of Unitree, on the industrial rhythm at the World Physical AI Conference in Hangzhou. He believes that the breakthrough window for physical AI is expected to be shortened to 1 to 3 years, far earlier than the previous maximum prediction of 10 years.

In addition, Tesla's Optimus under Musk has started initial mass production ramp-up at its Fremont factory, with a guaranteed shipment guidance of 10,000 to 20,000 units in 2026. Musk expects that the number of humanoid robots will exceed the number of humans in the future.

In September this year, Sam Altman, CEO of OpenAI, confirmed in a podcast that the company will independently develop humanoid robots, and also lay out other forms of robot products, planning to come up with a convincing demonstration prototype in 2027.

This article is from "Jiemian News", reported by Shen Xiaoge, and published with authorization by 36Kr.