Bank of America says the rapid expansion of the Nvidia-driven AI infrastructure is exposing a critical weakness in data-center design, one that most investors are overlooking.
In a new CNBC interview, Andrew Obin of BofA Securities says the Wall Street narrative around Nvidia’s growth is missing the deeper physical and electrical realities driving the next leg of AI spending.
While investors focus on chips and compute power, he argues the hidden bottleneck lies in the hardware underneath: power delivery and cooling systems that are no longer sufficient for next-generation AI racks. Obin says data centers once treated electrical and cooling systems as commodities. Those days, he argues, are over.
“So look, I think what the street is missing is that before, right, the electrical stuff, the cooling stuff, they were basically commoditized… Now, an average data center is around 100 megawatts, right? The cost per megawatt is $50 million, so that’s a $5 billion enterprise.”
According to Obin, Nvidia’s accelerated product cycles are forcing operators to redesign the rack itself, moving from traditional alternating current to direct current systems running at far higher voltages. The shift means massive retrofits, replacing the internal electrical architecture, HVAC systems, and cooling setups that were never engineered for the thermal load of AI GPUs.
“All of a sudden, the HVAC system, the cooling system that you had before, is not going to work… When they go to the Kyber rack, you will effectively have to rip out all your electrical stuff inside the data center and put in brand new electrical equipment.”
He estimates that for every megawatt of AI power, electrical and cooling costs make up roughly $3.3 million, but failure to upgrade could destroy $40 million worth of chips per megawatt “within 30 seconds.” That figure illustrates how thermal and power constraints have become the make-or-break layer in AI infrastructure.
“Rubin Ultra is not going to work if you don’t have the right cooling and if you don’t have the right electrical connection… So that’s the big difference. All of a sudden, this has become the technology bottleneck, without which Nvidia’s plan doesn’t work.”
Disclaimer: Opinions expressed at CapitalAI Daily are not investment advice. Investors should do their own due diligence before making any decisions involving securities, cryptocurrencies, or digital assets. Your transfers and trades are at your own risk, and any losses you may incur are your responsibility. CapitalAI Daily does not recommend the buying or selling of any assets, nor is CapitalAI Daily an investment advisor. See our Editorial Standards and Terms of Use.