Jigar Shah, the former head of the Department of Energy’s Loan Programs Office and co-founder of Generate Capital, used his recent Prof G Markets appearance with Ed Elson to throw cold water on one of the loudest narratives in tech investing. The claim under attack is that hyperscalers can actually build the data center capacity they keep promising.
His core claim, in his own words, is that the industry “can’t build more than 50 gigawatts of data centers between now and 2030”, yet announced commitments are “creating havoc at the level of 500 gigawatts across the country.” Shah named five binding constraints: the grid, transformers, GPUs, memory, and CPUs.
The Texas Tell
Shah’s go-to example is ERCOT. He cited filings suggesting “the load queue could be 300 gigawatts” against a current Texas system that runs around 70 gigawatts. His own forecast for what actually gets built in Texas is closer to 30 gigawatts. Ed Elson pointed out that OpenAI alone has promised “30 gigawatts worth of compute by the year 2030” and has built roughly 2 gigawatts so far. Reports also suggest OpenAI is missing internal revenue and user targets.
What Hyperscalers Are Actually Spending
Set Shah’s ceiling against the demand side. Microsoft (NASDAQ:MSFT | MSFT Price Prediction) just reported Q3 FY26 capex of $30.876 billion, up 84.39% year over year, with an AI run rate of $37 billion, up 123% YoY, and commercial RPO of $627 billion. CEO Satya Nadella detailed the buildout in the company’s Q3 8-K.
Amazon (NASDAQ:AMZN) spent $44.2 billion on capex in Q1 FY26, with Andy Jassy guiding to about $200 billion across 2026. Alphabet (NASDAQ:GOOG) raised 2026 capex to $175-$185 billion while Google Cloud backlog nearly doubled to over $460 billion. Meta Platforms (NASDAQ:META) just lifted 2026 capex guidance to $125-$145 billion, citing higher component pricing and additional data center costs.
Feeding all of it is NVIDIA (NASDAQ:NVDA), where Jensen Huang said “global demand for NVIDIA’s AI infrastructure is incredibly strong” and data center revenue hit $39.1 billion, up 73% YoY.
What It Means for Investors
Shah’s point is narrower than blanket AI skepticism. The announced gigawatt figures function as accounting fiction without transformers, interconnects, and silicon to back them. For retirement investors, that creates two practical questions. First, do hyperscaler revenue ramps assume capacity that physically slips by years? Meta shares fell 8.97% on April 30 after capex came in heavier than expected, and Microsoft is down 12.03% year to date. Second, who benefits from scarcity? NVIDIA, up 91.98% over the past year, suggests the market already has a working answer.
Keep an eye on transformer lead times, ERCOT interconnection data, and hyperscaler commentary on power-constrained sites. Those numbers will tell you whether 500 gigawatts is a plan or a press release.
Top Gaining Stocks
Top Losing Stocks