Elon Musk spent years telling the world that a small patch of desert could power the entire United States with solar panels, yet today his xAI division is operating 62 unpermitted methane gas turbines across Tennessee and Mississippi to run Grok, an AI chatbot that fewer people are using each month. The turbines emit more than 6 million tons of greenhouse gases annually, according to xAI's own permit applications, while Grok has plummeted from second place to fifth place globally in just a few months.

What Happened to Musk's Solar Energy Vision?

In July 2017, Musk stood before the National Governors Association and made a compelling case for solar power. He explained that a 100-by-100-mile patch of desert in Nevada, Texas, or Utah could power the entire United States, with just one square mile of battery storage needed for 24/7 power. This wasn't a casual remark; it became the foundation of Tesla's entire business thesis.

Tesla's original 2006 Master Plan stated the company's "overarching purpose" was "to help expedite the move from a mine-and-burn hydrocarbon economy towards a solar electric economy." Musk repeated this desert solar pitch multiple times over the years, and independent analyses confirmed the math worked. The point was simple: solar energy was so abundant, cheap, and reliable that the energy transition was just an execution problem, not a technology problem.

Fast forward to 2026, and the picture has shifted dramatically. While xAI has spent $697 million on Tesla Megapacks for its data centers, it hasn't purchased a materially significant amount of solar panels from Tesla, according to SpaceX's recent Securities and Exchange Commission filing. Instead, the company is doubling down on natural gas, planning to buy $2.8 billion more gas turbines.

Why Is Grok Burning So Much Energy If Nobody's Using It?

The energy consumption becomes even more puzzling when you look at Grok's actual market performance. The chatbot entered 2026 as the second most-popular AI chatbot globally, but by April it had dropped to fifth place, behind ChatGPT, Claude, Gemini, and DeepSeek. Monthly active users fell 12.5% in a single month to 12.2 million, while Claude surged 44% to 23 million users. Downloads crashed 60%.

In the enterprise market, adoption is even worse. Only 7% of companies reported using Grok in March 2026, compared to 48% for Claude and 40% for Gemini. A survey of 260,000 Americans found that just 0.174% paid for Grok in the second quarter of 2026. So Musk is burning 6 million tons of greenhouse gases per year to power a chatbot that fewer and fewer people actually use.

With Grok unable to fill the Colossus data center's 220,000 Nvidia graphics processing units (GPUs), Musk leased the entire facility to Anthropic, the company he had publicly called "misanthropic and evil" just three months earlier. By May, Anthropic was paying xAI $1.25 billion per month for the compute, a deal worth over $40 billion through 2029.

How Is Musk Justifying the Gas Turbines Now?

The answer lies in SpaceX's upcoming initial public offering (IPO). SpaceX filed its registration statement with the SEC on May 20, targeting a valuation between $1.75 trillion and $2 trillion, which would be the largest IPO in history with a planned $75 to $80 billion raise. In that filing, Musk introduced a new energy thesis: space-based solar power.

SpaceX claims that space-based solar arrays can generate "more than five-times the energy" of terrestrial ones thanks to 24/7 illumination. The company filed with the Federal Communications Commission for an "orbital data center" constellation that could include up to one million satellites in low Earth orbit. Musk now says that within two to three years, the lowest cost way to generate AI compute will be in space.

The problem is obvious: this is the same man who spent a decade telling everyone the answer was already on the ground. If a 100-by-100-mile patch of desert can power the entire United States, as Musk repeatedly stated, then why are we now talking about launching solar panels on rockets? The sun hasn't moved. The math hasn't changed. Solar panel efficiency has only improved, and costs have only come down.

Steps to Understanding the Financial Incentives Behind the Shift

• IPO Valuation Strategy: SpaceX needs to justify a $2 trillion valuation to investors, and a space-based solar thesis creates future demand for Starship launches, transforming SpaceX from a rocket company into an energy infrastructure company that trades at much higher multiples.
• Terrestrial Solar Downplayed: Terrestrial solar barely gets a mention in the SpaceX S-1 filing, and when it does, it's only to argue how much better space solar would be, despite decades of evidence supporting ground-based solar viability.
• Sunk Cost Recovery: The gas turbines and compute infrastructure are already built, and capital expenses are sunk; electricity is now the only major operating cost, making the Anthropic lease deal essential to recovering investment and generating IPO-ready revenue.
• Regulatory Pressure Mounting: The Environmental Protection Agency closed the regulatory loophole xAI was exploiting in January 2026, but thermal drone footage from February showed the turbines still running, and both the NAACP and Earthjustice have asked courts for emergency action to stop the illegal pollution.

The economics of orbital data centers remain challenging at best. Power costs for Starlink satellites already run multiples higher than terrestrial data centers. Radiation shielding for AI chips in orbit is expensive and unproven at scale. And it remains unclear whether AI training workloads can be distributed effectively across space-based infrastructure.

What's clear is that Musk's pivot from terrestrial solar to space-based solar isn't driven by new physics or engineering breakthroughs. It's driven by the need to justify a historic IPO valuation and to generate revenue from infrastructure that was built to power a chatbot that lost the AI market race. The 2017 version of Elon Musk, who made the case for solar power with such clarity and conviction, would have had a field day with that math.