•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
•
AI’s rapid expansion is driving an urgent scramble for electricity, with hyperscalers prioritizing how quickly new computing capacity can be powered over cost, according to Fred Thiel. “The big demand right now for energy is from obviously the hpc ai sector and they need energy that is on now and not on in 2029,” he said.
Thiel linked the immediate energy crunch to high-performance computing (HPC) requirements for AI. He described hyperscalers as being locked in a “capacity war,” competing on how many GPUs they can run at scale. “It’s all about a capacity war who has gpu’s running… the more stuff you have running the smarter your model is,” he said.
He added that the timing of bringing capacity online is a key factor shaping infrastructure decisions. “The timing of bringing capacity online is prioritized over cost by hyperscalers,” Thiel said, emphasizing that operational readiness is central to maintaining competitive advantage.
Thiel said energy provisioning for AI is influencing broader infrastructure development, with providers and operators focusing on readiness to support fast-growing demand. The urgency, he argued, reflects the speed at which AI technologies are becoming strategically important.
Thiel also pointed to international opportunities in bitcoin mining, arguing that some regions have less demand for AI-driven hyperscale capacity than the United States. “A lot of our growth for bitcoin mining is actually targeted internationally… internationally there is much less demand for ai in hyper scalar capacity,” he said.
He said international markets can be more economically viable for mining than the US, citing differences in energy market dynamics. He highlighted underutilized energy resources in countries such as Saudi Arabia and France, where renewable energy may be available but not connected to the grid for sale. He also cited nuclear generation: “66 of the energy is nuclear… and the capacity at the nuclear power plants is running below 70%.”
In Thiel’s view, grid connectivity limits the ability to monetize available power, creating conditions where mining can use excess energy that would otherwise remain underutilized.
On bitcoin mining capacity, Thiel said costs for mining capacity are at their lowest point “most probably ever.” He attributed this to oversupply and weak demand. “The cost of a terahash of bitcoin mining capacity is at its lowest point most probably ever… it’s a combination of oversupply and no demand,” he said.
He connected reduced demand for mining rigs to strategic shifts by public miners. “40% of miners in the us are the public miners… they’re not buying any more miners,” Thiel said, adding that public miners are increasingly focusing on HPC and AI. He said this shift affects future investment in mining infrastructure and reflects broader industry pricing dynamics.
Thiel argued that AI infrastructure is moving toward modular approaches to reduce costs and improve deployment speed. “Take the infrastructure down to a million dollars a megawatt and solve that problem through modularizing a data center,” he said. He also suggested that AI inference does not require the largest, tightly interconnected clusters. “In the world of inference you don’t need to have all these huge clusters that are interconnected,” he said.
He emphasized that inference is central to how AI businesses generate revenue. “Revenue generation in the ai world is only done through inference that’s what you pay for,” he said, adding that modular data centers can better align with the operational needs of AI applications.
Thiel said the industry is also adjusting to long timelines for power access. He described how, beyond 2029, some operators may pursue building their own power generation because permitting and development can take years. “As you get beyond ’29 there you can look at i’m going to go build my own power generation because that’s going to take a couple of years to three years just to get permitting etcetera,” he said.
He characterized this as a shift in which data center providers increasingly act as energy developers, reflecting operational challenges in energy provisioning. “It’s all about time to time to energization of the data center,” he said, describing energy development as a critical component of future data center planning.
Thiel also discussed AI’s effect on traditional enterprise software economics, arguing that AI-driven agentic technologies can generate insights that would otherwise be produced by human operators. “All the value accretes at the end… the moat that sap has is being eroded now by ai,” he said, adding that while SAP may not be “ripped out,” AI can replicate the insights humans would have produced.
He said the automation of insights changes competitive dynamics across the tech industry and underscores the need for adaptation as software value erodes under AI-driven efficiency and automation.
Premium gym chains are entering a “golden era” that is ending or already in decline, as rising operating costs collide with shifting consumer preferences toward more flexible, community-based ways to exercise. Long-term memberships are shrinking, margins are pressured by higher rents and facility expenses, and competition from smaller, more personalized…
