SpaceX Doesn't Need Solar. Natural Gas Is Powering the AI Buildout, and Most Investors Still Aren't Positioned for It

When Elon Musk unveiled TeraFab in March 2026, the vision was cinematic: a 100-million-square-foot semiconductor mega-factory producing one terawatt of AI compute per year, fueled by solar power from orbit. Musk called it "the most epic chip-building exercise in history so far." He talked about scaling civilization through space-based solar.

Then on August 6, his own energy team walked into a public meeting in Grimes County, Texas, and said something very different. Riley Trettel, who oversees energy and data center development for SpaceX, told residents: "We're bringing our own power. We're going to be building natural gas fired power plants, and we're going to be building very large battery arrays."

No mention of Tesla solar. No mention of the orbital solar vision that Musk pitched at the Seaholm Power Plant in Austin five months earlier. Just natural gas turbines and batteries, on the ground, in Texas, starting "nearly immediately."

The media reaction was predictably focused on the irony. Elon Musk owns the world's largest battery company and a rapidly growing solar business, yet his own chip factory is being powered by fossil fuels. The narrative machine cranked out headlines about hypocrisy, missed opportunities, and the limits of clean energy.

I've been very surprised by how little attention the real story has gotten. The hypocrisy angle is a distraction. What matters is that the person who was supposed to be solar's greatest evangelist has just confirmed what the structural data has been telling us for months: the AI buildout is going to be powered by natural gas. Not because anyone wants it to be. Because physics, timelines, and scale leave no other option for the next five to seven years.

The false narrative: solar can power the AI revolution

The consensus story going into 2026 was that renewable energy, backed by ever-cheaper batteries, would absorb the massive electricity demand from AI data centers. Solar was the star of that show. The EIA's Short-Term Energy Outlook does project solar to supply the largest increase in U.S. power generation over the next two years, growing roughly 20% annually. But solar provides less than 10% of actual generation capacity. More importantly, it delivers roughly one-quarter of the energy per unit of installed capacity compared to natural gas.

That matters for data centers because they don't need energy on weekends when the sun is shining. They need it at 3 a.m. in January, running at full load, seven days a week, for the next decade.

AI racks require up to 150 kilowatts of power, roughly 10 times the density of traditional cloud computing. Lawrence Berkeley National Laboratory projects U.S. data center energy demand will double by 2028. Deloitte forecasts AI data center demand alone will grow from 4 gigawatts in 2024 to 123 gigawatts by 2035. That's enough to electrify nearly 100 million East Coast households.

None of that load can wait for solar farms to be built, permitted, and connected to an already-broken grid with multi-year interconnection queues. Natural gas turbines can be deployed in two to five years. Solar-plus-transmission projects face permitting delays of the same duration, then still can't run 24/7.

Musk knows this. He already proved it with xAI's Colossus data centers in Memphis, which he powered in 122 days using trailer-mounted natural gas turbines from APR Energy, a company he reportedly acquired for over $1 billion. The grid connection route would have taken months, if not years, plus regulatory approvals and community resistance — Memphis residents have already sued over pollution concerns.

TeraFab is the same playbook at a larger scale. SpaceX controls over 13,000 acres in Grimes County. They'll build gas plants on site, pair them with battery storage, and never touch the ERCOT grid. They even committed not to raise electricity bills for local ratepayers. The initial phase is $16.8 billion, with total expansion costs potentially reaching $119 billion. That's the kind of capital allocation that doesn't bend to a solar-powered dream.

Why this is the opposite of a clean-energy setback

Here's where the market gets it wrong. The false narrative isn't just about solar. It's the belief that this natural gas demand for data centers is temporary, a bridge fuel story that will fade once renewables or nuclear scale up. That belief is pricing energy stocks too conservatively.

The United States now leads the world in natural gas-fired power capacity in development, with approximately 252 gigawatts in the pipeline, accounting for nearly one-quarter of the global total. If built, this would expand the existing U.S. gas fleet by nearly 50% at an estimated cost of over $416 billion. More than one-third of this capacity is slated to directly power data centers on site. Texas alone has 80.6 gigawatts in development, a four-fold increase from the previous year. Only China has more gas power in development than Texas.

Industry forecasts project 4 to 8 billion cubic feet per day of added natural gas demand from data centers by 2030, on top of current U.S. consumption of 91.6 Bcf/d. Williams projects 10 Bcf/d of incremental power demand by 2035. That's not a bridge. That's a structural demand shift.

And there's a supply-side constraint making the timing even more acute. GE Vernova has sold out its gas turbine production through 2029. Mitsubishi and Siemens Energy face backlogs into 2028. New orders now face delivery dates of 2030 or later. These three manufacturers control more than 75% of the global gas power market. This is not a market that can instantly flood the zone with new turbines to meet demand. The bottleneck is real.

Where the money is actually flowing

TeraFab is the headline. But the money isn't in building chip factories. It's in the infrastructure that delivers the fuel. As a 4th-generation oil and gas man, I look at free cash flow and dividend commitment first, and the natural gas midstream sector is where the structural data converges.

Three names deserve attention.

EQT Corporation is the largest U.S. natural gas producer, and it's the purest play on volume demand growth. The company generated $3.8 billion in trailing-twelve-month free cash flow, up 77% year over year. That FCF growth rate is exceptional and reflects both rising volumes and better pricing. The balance sheet is in good shape: $12.5 billion in total debt against $28.9 billion in equity, for a debt-to-equity ratio of just 19.6%. The dividend yield is only 1.27%, though, which is the trade-off. EQT trades at 12 times trailing earnings and 5.6 times EV/EBITDA, the cheapest multiple among the names I'm covering. The low payout ratio of 12% gives management enormous flexibility to reinvest or return capital. In my opinion, EQT is the highest-conviction growth play on data center gas demand, but its yield is a concession you make for FCF acceleration.

Kinder Morgan is the largest natural gas transmission network in North America, moving 40% of all U.S. gas. It's a toll road business: the volume goes through, and Kinder Morgan collects. The company's 2026 project backlog exceeds $10 billion, with 93% tied to natural gas, and management has identified a "shadow backlog" of over $10 billion beyond that. Kinder Morgan generated $3.2 billion in free cash flow on a trailing basis, with a 3.8% dividend yield and 14 consecutive years of dividend payments. The debt load is heavier, at $41.2 billion total against $32.9 billion in equity, but the FCF covers the dividend seven times over. Kinder Morgan trades at 20 times trailing earnings and 13.2 times EV/EBITDA. That's not cheap, but it's not unjustified for a toll road with a $10 billion backlog.

Energy Transfer is where I look for income. The company generates $5.2 billion in trailing free cash flow and pays an 8.2% dividend yield, the highest of any major energy infrastructure name in the market. Its dividend payout ratio sits at less than 1% of FCF, which means the dividend is almost trivially covered. Energy Transfer has executed the Desert Southwest Expansion for 2.3 Bcf/d of data center power demand and is currently negotiating over 150 data center and power opportunities. It delivered gas to an Oracle data center in Texas in January 2026. The balance sheet carries $97.4 billion in total debt, which looks daunting, but the company generates $12.1 billion in operating cash flow and has $1.0 billion in cash. Energy Transfer trades at 14 times trailing earnings and 8.1 times EV/EBITDA, with a 19-year streak of consecutive dividends.

I favor Energy Transfer for investors who prioritize current income with a structural tailwind. The 8.2% yield, backed by 100-plus times FCF coverage, is not a number you find in this market without a reason to worry. In Energy Transfer's case, the reason is the debt load. But the debt is manageable against $12.1 billion in annual operating cash flow, and the data center gas demand creates a new revenue stream that should only strengthen cash generation going forward.

For investors who want exposure to gas volume growth with a cleaner balance sheet, EQT is the pick. The 77% FCF growth rate and 19.6% debt-to-equity ratio are structurally superior, even if the 1.27% yield won't keep you up at night with income.

One stock I'm avoiding: Williams Companies

The Williams Companies has positioned itself aggressively in the data center power space, with a $7.3 billion execution backlog across four projects serving data centers and a 6 GW additional backlog. The story is compelling. The valuation is not. Williams trades at 28 times trailing earnings and 35 times forward earnings, with an EV/EBITDA multiple of 20.4. The company generated negative $214 million in free cash flow over the trailing twelve months, running a $6.1 billion capital expenditure program that exceeded its $6.0 billion operating cash flow. Its dividend payout ratio is 88.7% of FCF, which, given the FCF is negative, means the dividend is being funded by balance sheet draws or working capital shifts. The 2.9% yield on a stock trading at 28x earnings that can't currently cover its dividend from free cash flow is not a setup I respect. Williams will be fine if the data center buildout plays out exactly as planned. But the margin for error at this valuation is essentially zero.

What about Tesla?

The irony of Musk choosing gas over solar for TeraFab doesn't hurt Tesla. It arguably helps. Tesla's energy storage division, driven by Megapack deployments, is on track for $18.3 billion in 2026 revenue, with margins outperforming its automotive segment. TeraFab will need "very large battery arrays" to smooth its gas-fired generation, and Tesla is the obvious supplier. The company's Q2 2026 results showed energy storage as a primary growth engine. Solar may not be powering the factory, but Tesla batteries almost certainly will be smoothing the output. That's not hypocrisy. That's cross-pollination: gas for baseload, batteries for stability.

The bottom line

The TeraFab story is not about Elon Musk abandoning solar. It's about the structural reality that the AI buildout demands gigawatts of continuous, dispatchable power, and natural gas is the only fuel that can deliver at the required scale and timeline. Musk's decision just confirms what energy data has been showing since 2025: gas-fired capacity additions are surging, turbine supply is bottlenecked, and data center operators are building behind-the-meter plants to bypass a grid that can't keep up.

For investors, the allocation question isn't whether natural gas matters. It's which companies actually capture the value. Energy Transfer at an 8.2% yield with $5.2 billion in FCF and a massive data center deal pipeline is my top pick for income-oriented exposure. EQT at 12x earnings with 77% FCF growth is the growth play. Kinder Morgan is the steady toll road.

I rate Energy Transfer as a Buy for its combination of yield, FCF coverage, and data center deal momentum. I rate EQT as a Buy for its FCF acceleration and clean balance sheet. I rate Kinder Morgan as a Hold — solid yield and backlog, but the 13x EV/EBITDA valuation already reflects much of the near-term optimism. I rate Williams Companies as a Sell at current levels, given its 28x earnings multiple, negative free cash flow, and a dividend that the operating cash flow can't cover.

The New Age of Energy Abundance isn't over. But for the next five years, the AI revolution is a gas-fired one. The investors who position accordingly will collect the dividend while the narrative folks debate solar.