The AI buildout's biggest bottleneck is shifting from chips to electricity, and converted Bitcoin mines are part of the proposed fix
Morgan Stanley’s Stephen Byrd calculated that US data center developers face a 70-gigawatt power shortfall through 2028.
The numbers behind the gap
The shortfall is a function of demand outrunning supply at a startling pace. Morgan Stanley now projects total US data center power demand of 97 GW through 2028.
That figure is a 43% jump from the bank’s earlier estimate of 68 GW. The upgrade reflects rising chip shipments and increasingly power-hungry rack systems, including Nvidia’s.
The shortfall estimate itself has climbed as the modeling evolved. An earlier pre-solutions calculation put the gap at approximately 57 GW. Later modeling pushed that figure to 70 GW.
Then comes the mitigation math. After accounting for onsite generation options, such as gas turbines, Bloom Energy fuel cells and nuclear co-location, the research puts the net shortfall at around 33 GW.
That remaining gap would represent about 34% of projected total demand.
The outlook does not improve with time, either. Forward modeling suggests the gap could widen to between 68 and 72 GW by 2029.
From chip shortage to power shortage
Morgan Stanley’s research marks a pivot in that story. The binding constraint is moving from chip supply chains to electricity availability.
This matters because tech companies’ future capital spending is estimated to reach into the trillions. If power cannot keep pace, some of that money could end up funding buildings that sit waiting for electrons.
The research also notes that Nvidia appears relatively insulated from these shortages compared to almost all other semiconductor providers. Its hardware is a major driver of the demand surge, yet it is seen as better protected from the fallout.
Where Bitcoin miners fit in
Converted Bitcoin mines appear on the list of solutions that could help narrow the gap. Bitcoin mining facilities already have what AI developers desperately want: grid connections, power capacity and industrial sites built to handle heavy electrical loads.
The research frames this as an ancillary piece of the solution rather than a core one.
What this means for the AI buildout
According to the research, major tech players sit in a more advantageous position to secure power. Smaller developers face greater risk of delays. They have less leverage with utilities, less capital to fund onsite generation, and fewer options when grid capacity runs short.
