MS: Market under-appreciates mismatch between AI-powered demand and supply growth

The market under-appreciates the discrepancy between artificial intelligence’s rapid growth in power demand growth and the slow growth in power supply, said Katy L. Huberty, analyst and global director of research at Morgan Stanley.

In a note, Huberty said that MS analysts concluded that the market is “under-appreciating” generative AI compute power cost, the mismatch between Gen AI’s growth and power grid infrastructure’s growth, and the time value for power solution providers.

There is a rapid drop in the cost of generative AI compute power. Data centers’ capital cost per teraflop -- which measures one trillion floating point operations per second -- dropped by 50% when moving from Nvidia’s (NVDA) Hopper GPU to its new Blackwell GPU, recently announced.

According to MS, the data center capital cost per teraflop is about $14, compared to $7 for a Blackwell data center.

Secondly, there is “a significant mismatch between the hyper-rapid growth in Gen AI power needs (notwithstanding continued efficiency improvements) and the slow growth in power grid infrastructure,” Huberty said.

The third dynamic at play is the massive magnitude of time value for power solutions providers that would result in the reduction of delays in powering new and expanded data centers.

Companies that are well positioned to serve this rapidly growing power needs of the Gen AI enablers include (all buy-rated stocks):

• The AES Corp. (NYSE:AES)
• Bloom Energy Corp. (NYSE:BE)
• Constellation Energy Corp. (NASDAQ:CEG)
• Engie SA (OTCPK:ENGIY)
• Fortum Oyj (OTCPK:FOJCF)
• Iberdrola S.A. (OTCPK:IBDSF)
• NextEra Energy Inc. (NEE)
• Orsted A/S (OTCPK:DNNGY)
• Public Service Enterprise Group Inc. (PEG)
• RWE Aktiengesellschaft (OTCPK:RWEOY)
• Sembcorp Industries Ltd. (OTCPK:SCRPF)
• Tenaga Nasional Berhad (OTCPK:TNABF)
• Vistra Corp. (VST)