Elon Musk believes the AI race has reached a point where no one can slow it down.
In a recent wide-ranging interview with Business Insider, Musk predicted that AI could surpass the combined intelligence of humanity within roughly five years. He also acknowledged a 10–20% chance of AI going catastrophically wrong, while arguing that its potential benefits still make moving forward worthwhile.
Whatever the risks, the competitive race shows little sign of slowing.
And that means every major player now faces the same brutal incentive: Keep building or risk being left behind. OpenAI, Anthropic, Google, xAI, and their competitors are pouring capital into larger models, more powerful chips, and enormous data center campuses—with each new investment forcing the others to accelerate.
It also means the industry is charging toward the same physical bottleneck at the same time: electricity.
And the latest numbers from America’s largest grid operator show that power demand is already outrunning supply.
PJM Interconnection serves more than 67 million people across 13 states and Washington, D.C. In its latest capacity auction, PJM secured 6.8 gigawatts less power than it would need to meet its one-event-in-10-years reliability standard.
That is roughly equivalent to the output of six or seven large nuclear reactors.
More importantly, the shortfall appeared remarkably quickly.
For the 2026/2027 planning year, PJM procured just 139 megawatts more than its projected reliability requirement. For 2027/2028, it fell 6.6 gigawatts short. The deficit then widened to 6.8 gigawatts for 2028/2029.
In other words, one of America’s most important power suppliers went from narrowly meeting its reliability target to facing a multigigawatt deficit.
PJM says nearly 5.1 gigawatts of the increase in its forecast peak demand for 2027/2028 came from data centers.
Simply put, while Musk’s timeline may sound aggressive, PJM’s numbers show the infrastructure strain is already arriving.
AI is concentrating demand faster than the grid can respond
AI data centers place far greater demands on the electrical system than the facilities built for conventional cloud computing.
Traditional enterprise data center racks have commonly operated around 5–10 kilowatts. Modern AI racks increasingly consume 40 kilowatts or more, while the densest configurations can exceed 100 kilowatts.
That means AI companies can pack far more computing power—and far more electricity demand—into the same physical space.
Supporting that demand requires more than installing additional servers. Data center developers may need larger substations, new transformers, upgraded cooling systems, expanded transmission capacity, and firm agreements for power delivery.
Those projects can take years.
The result is a widening gap between the amount of AI infrastructure companies want to build and the amount the electrical system can support on schedule.
That gap is turning access to power into one of the most valuable assets in the AI economy.
Why Bitcoin infrastructure suddenly matters
This is where an overlooked group of companies enters the story.
Large Bitcoin miners spent years securing many of the same assets AI data centers now desperately need: land, substations, grid interconnections, power contracts, cooling equipment, and sites capable of supporting energy-intensive computing.
Bitcoin mining and AI computing use different hardware, and converting a mining site into an AI data center still requires significant investment.
But these companies often begin the process several years ahead of a developer starting from an empty piece of land.
IREN Limited (IREN) has demonstrated how valuable that head start can become.
IREN began primarily as a Bitcoin miner and has since expanded aggressively into AI cloud services and data center infrastructure.
The company has already built operating infrastructure, secured major customers, and demonstrated its ability to attract capital for large-scale AI projects.
In November 2025, IREN announced a multiyear agreement with Microsoft valued at approximately $9.7 billion to provide AI cloud infrastructure powered by NVIDIA GPUs.
That agreement helped establish IREN as one of the most prominent examples of the Bitcoin-to-AI infrastructure transition.
DigiPower X (DGXX) is pursuing a similar transition from a much earlier stage.
The company says it has approximately 400 megawatts of secured power across data center sites in Alabama, New York, and North Carolina.
That does not mean DGXX already has 400 megawatts of completed, revenue-producing AI capacity.
But it gives the company access to the raw resource increasingly constraining the broader industry.
And DGXX has begun converting that potential into commercial agreements.
In May 2026, DGXX signed a 10-year agreement with Cerebras Systems for a planned 40-megawatt AI data center in Alabama. The project calls for an initial 15 megawatts of IT capacity, followed by another 25 megawatts for a combined 40 megawatts.
The initial contract term is valued at approximately $1.1 billion, subject to DGXX meeting its construction and development obligations. The first phase is targeted to enter service in December 2026, with the full deployment targeted for the end of the first quarter of 2027.
Yet the difference in how the market values the two companies remains enormous.
As of July 27, 2026, IREN carried a market capitalization of approximately $12.5 billion. DGXX was valued at approximately $365 million.
That valuation gap reflects far more than access to power.
IREN has a longer operating history in AI, significantly more completed infrastructure, larger customer contracts, greater access to capital, and considerably more institutional coverage.
DGXX still needs to build out its sites, bring contracted capacity online, fund its expansion, and prove it can execute at scale.
But the comparison shows what can happen when investors begin viewing an energy-intensive Bitcoin operator as an AI infrastructure company.
IREN represents what the market is willing to pay once that transition gains commercial credibility.
DGXX represents the earlier-stage possibility that investors eventually assign greater value to its secured power, grid access, and developing AI customer relationships.
Why the power trade is the AI trade
The conventional AI trade has focused on chipmakers and the hyperscalers spending hundreds of billions of dollars to build new computing capacity.
Those companies remain central to the trend.
But every chip they install requires electricity. Every new AI cluster requires a grid connection. And every data center opening depends on enough power being available at the right location and at the right time.
PJM’s latest numbers show that those megawatts can no longer be taken for granted.
That creates a structural advantage for companies that already control scarce power infrastructure.
Land can be purchased. Servers can be ordered. Capital can be raised.
A permitted site with secured power, a completed interconnection, and a functioning substation is much harder to replicate.
That is the structural advantage the market is only beginning to recognize.
Bottom line
The central takeaway from Musk’s interview is clear: The AI race will continue regardless of which company reaches the next breakthrough first.
That makes the shared bottleneck more important than the identity of the eventual winner.
Right now, that bottleneck is power.
The chip companies and hyperscalers remain the obvious AI investments. The infrastructure supplying electricity beneath them offers a less crowded—and potentially much earlier—way to approach the same trend.
Watch the megawatts.
That’s where the next phase of the AI buildout will either accelerate or stall.
For more analysis on how to position for the AI power buildout, tune in to Wall Street Unplugged.
