U.S. startups working to solve the power and computing bottlenecks behind the artificial-intelligence boom have raised $33.5 billion so far in 2026, according to a Morgan Stanley report that began circulating in mid-October and has continued to draw attention this week — nearly double the total raised in all of 2025, and a sign of how much investor money is chasing the infrastructure underneath AI rather than the AI models themselves.
Power, Not Chips, Is the Constraint
More than 80% of this year's funding has flowed to companies working on expanding power availability or squeezing more computing output out of each unit of electricity, according to the report, cited by NewKerala and other outlets covering the research. Looking back further, the bank's analysts found startups in this space have raised more than $50 billion since early 2024 across four categories, including new data-center designs and site locations.
The framing marks a shift in how investors are thinking about the AI buildout: rather than simply securing access to graphics processors, companies are increasingly expected to coordinate power, computing, networking and cooling as a single system. Recent individual rounds illustrate the scale of money involved — cloud and data-center company Crusoe raised $3.9 billion in September at a $30.9 billion valuation in a round co-led by Atreides Management, Mubadala Capital and Valor Equity Partners, while rival Firmus raised $2 billion in August from backers including Coatue Management, Nvidia-linked vehicles and Blackstone.
Morgan Stanley's analysts caution, however, that raising capital is not the same as building capacity. The report argues that validation for this category of startup will increasingly depend on how much power and computing infrastructure actually gets built and switched on — with grid connections, permitting and equipment-procurement timelines as the biggest risks, rather than access to funding itself.
That shift has also reshaped who is doing the investing. Traditional venture firms have been joined by sovereign wealth funds, private equity vehicles and chipmakers themselves writing checks directly into power and data-center infrastructure — a blurring of venture and infrastructure finance that is becoming one of the year's defining capital-markets stories.
The tally adds to a running theme across this year's financing data: AI's bottleneck has shifted from chips to electricity, and from building models to building the physical plants that run them.