UBS: NVIDIA's data center revenue is expected to double by 2028

According to the Zhitong Finance APP, this week, NVIDIA (NVDA.US) held its GTC 2025 annual conference. UBS stated in its third-day report of the GTC conference that NVIDIA's data center revenue is expected to at least double from 2025 to 2028; it also pointed out that there are still highlights in data center capital expenditures. Additionally, the development of AI Agents, Samsung's HBM4 technology, advancements in humanoid robots, and the advantages of liquid cooling solutions were also highlighted.

In the analyst Q&A session, UBS mentioned that while no specific data center capital expenditure forecasts were provided, the conclusion drawn from NVIDIA's comments at the conference is that it believes its data center revenue will at least double from 2025 (with UBS expecting $215 billion and Wall Street expecting $180 billion) to 2028, which implies that earnings per share during this period will be around $12.

The company also emphasized that many metrics for measuring data center capital expenditures often overlook some of the largest clusters currently under development, including xAl, OpenAl, sovereign, and many Neocloud companies. Other highlights include: 1) NVIDIA once again refuted the argument about computing power intensity and strongly reiterated that improvements in inference models actually drive increases in computing power intensity, as machines must "think for themselves" to solve problems and require rapid reasoning; 2) The ability to run these new models on cheaper hardware has made significant progress, but in many cases, NVIDIA believes this is not the most economical approach due to trade-offs in speed and performance; 3) NVIDIA emphasized that it is an infrastructure business and the only realistic choice that customers can rely on when planning large-scale deployments in advance (compared to self-developed ASIC chips).

With the participation of Randy Abrams, head of semiconductor securities research at UBS's Asia technology team, a very popular panel on artificial intelligence was highlighted, where executives from Meta (META.US), Microsoft (MSFT.US), ServiceNow (NOW.US), and Accenture (ACN.US) emphasized how AI Agents can now build their own software workflows instead of relying on scripts/plans previously written by SaaS providers. UBS wrote that following the thoughts of the Micron Technology (MU.US) panel, Samsung introduced its SOCAMM platform and announced some HBM4 specification data, including the use of HCB to reduce stacking height by about 33% (compared to TCB) and reduce thermal resistance effects by about 20% (compared to TCB).

The rise of humanoid robots once again highlights the popularity of NVIDIA's software stack, as the introduction of GRO0T means accelerating the development of humanoid robots while becoming a computing platform for physical artificial intelligence. Super Micro Computer (SMCI.US) emphasized in its meeting about AI data center liquid cooling that deploying liquid cooling solutions (compared to air cooling) has TCO advantages (including lower upfront capital expenditures) and significantly improves computing power density over the years. In UBS's meeting with Hon Hai Group, the company emphasized that its recent inventory increase was primarily driven by rack components rather than GPUs Hon Hai also confirmed that the stability issue of the GB200 has been resolved and expects to ship 30,000 to 50,000 units of racks in the next 12 months (with an expected share of over 40%)