The news of the (extremely) cheaper Chinese DeepSeek response to OpenAI’s ChatGPT — “40–50x more efficient than other large language models,” according to Goldman’s Rich Privorotsky in the days following DeepSeek’s launch earlier this year — ushered in a new theme: doing more with less and the mounting risks of an emerging AI data center bubble.
On Tuesday, Alibaba Group Holding Ltd. Chairman Joe Tsai told the audience at the HSBC Global Investment Summit in Hong Kong that the hundreds of billions of dollars in AI data center investments in the US appear to be the start of a bubble.
“I start to see the beginning of some kind of bubble,” Tsai said. He said some AI data center projects raised funds without securing “uptake” agreements, adding, “I start to get worried when people are building data centers on spec. There are a number of people coming up, funds coming out, to raise billions or millions of capital.”
At the same time as Tsai’s warning, ChatGPT creator OpenAI, along with SoftBank, Oracle, and other US tech firms, have planned $500 billion in AI infrastructure projects. Also, Meta recently announced $200 billion in data center projects, while Apple revealed new AI investments in the US.
“People are talking, literally talking about $500 billion, several 100 billion dollars. I don’t think that’s entirely necessary. I think in a way, people are investing ahead of the demand that they’re seeing today, but they are projecting much bigger demand,” Tsai pointed out.
Days after DeepSeek’s launch, we provided readers with enough color about the unfolding AI data center bubble:
deepseek better not be the real deal… pic.twitter.com/qv6vtINl2d
— zerohedge (@zerohedge) January 26, 2025
By late February, TD Cowen’s Michael Elias spooked the market and warned clients in a note that Microsoft had begun canceling data center leases. At the time, MSFT refuted those claims.
The last few months — during which Big Tech firms like Amazon, Alphabet, and Meta competed to outspend each other on AI data centers — may be ending (at least for now), as growing evidence suggests that peak AI data capacity could be much closer than previously thought.
On Monday, Goldman’s Allen Chang, Verena Jeng, and others revised down their “Rack-level AI Server volume forecast due to the combined reasons of product transitioning and uncertainties of demand and supply.”
Here’s their outlook for servers:
“Training” server will remain the growth driver given the increasing need for computing power to upgrade advanced AI models, but the volume ramp up is slower than we previously expected due to the combined reasons of product transitioning (Read more) and uncertainties of demand and supply. As the GPU platform is transiting to next generation in 2H25, shipment can potentially slow during the transition period. Uncertainties remain in production ramp up, given the complexity of full rack systems and there remains debates on the demand for intense computing power after the release of more efficient AI models like DeepSeek.
We expect revenues of AI training servers will grow at 24%/58% YoY to US$150bn/237bn in 2025-2026E, (vs. US$179bn/248bn previously). By segment:
AI training servers – full racks started shipment in 4Q24 and we expect will ramp up to a larger volume from 2Q25. We now model shipments of 19k/57k racks in 2025-26E (measured in 144-GPU equivalent), or a TAM size of US$54bn/ 156bn (vs. US$88bn/182bn previously).
AI training servers – high power (servers in traditional baseboard-based format) shipments will amount to 419k/418k units in 2025-26E (measured in 8-GPU equivalent), or a TAM size of US$97bn/81bn (vs. US$90bn/66bn previously).
“Inferencing” servers to follow up. We model AI inferencing servers to increase +41%/+39% YoY in volume and +105%/+30% YoY in value in 2025-26E, driven by increasing applications
“General” servers in gradual recovery. We expect volumes to grow 6%/4% in 2025-26E and revenues to grow 9%/7% YoY in 2025-26E, supported by the mile recovery of replacement cycles and the introduction of new CPU platforms.
Outlook for servers
In markets, Alibaba Group shares in Hong Kong fell 4% following comments from its chairman. Goldman’s China Data Center basket plunged 8% overnight. More broadly, Chinese technology stocks tumbled from a three-year high to the brink of a correction in just five sessions.
Saxo Markets chief investment strategist Charu Chanana told clients, “Alibaba’s caution around a potential bubble in AI data center buildouts has added pressure, hinting that the red-hot AI theme may face a short-term bump.”
The development of an AI data center bubble throws a massive wrench into President Trump’s Stargate AI infrastructure project.
Tsai concluded at the HSBC Global Investment Summit: “I’m still astounded by the type of numbers that’s being thrown around in the United States about investing into AI.”
Loading…
Read the full article here