ARK Invest Chief Futurist Brett Winton stated that SpaceX’s proposed $100 billion Starbase facility in Louisiana is "massively overbuilt" for connectivity services alone, signaling a strategic pivot toward large-scale artificial intelligence infrastructure.

The investment firm highlighted that the site’s capacity is designed to support an AI satellite constellation far exceeding current market models, positioning SpaceX to capture a significantly larger share of the emerging space-based computing market.

Starbase Infrastructure Scale

According to an investor note released on Monday, the Starbase project spans roughly 125,000 acres at Pecan Island in Vermilion Parish. The facility will feature five complexes, each equipped with two Starship towers.

Winton noted that operations will begin with ten launch pads and eventually expand to over a dozen towers. This infrastructure aims to support around 30 flights per day. The site will include on-site propellant production, power generation, deep-water shipping, vehicle processing, employee housing, and likely an airport.

SpaceX President Gwynne Shotwell indicated that current capacity cannot accommodate Starship’s intended launch cadence, necessitating this expansion.

AI Connectivity Economics

ARK Invest anticipates the capacity will serve payloads worth trillions of dollars annually. This includes billions for Starlink and trillions for the Starmind AI constellation.

Winton estimated that a single reusable Starlink-loaded Starship flight could generate about $4 billion in lifetime net cash flow. This stands against $1 billion in combined launch, satellite manufacturing, ground station, and acquisition costs.

Cost Advantage in AI Satellites

Winton expanded on these figures via X, noting that SpaceX is targeting a larger connectivity opportunity than ARK currently models. SpaceX claims a $1.6 trillion opportunity, while ARK estimates the range at $600 billion over the decade.

"Starbase Louisiana is massively overbuilt for just serving connectivity," Winton said. "The reason to build so big: the AI opportunity can compound at much larger scale."

ARK projects that terrestrial data center costs per Gigawatt will move north of $60 billion per GW in 2029 and cross $76 billion in 2032. In contrast, SpaceX’s AI satellites could weigh 2,000 kg each. Citing Starlink’s per-satellite cost of $1,000/kg, Winton noted that launch costs could fall south of $100 per kg if Starship becomes reusable.

This efficiency could translate to SpaceX launching satellites for $38 billion per GW. After the 100th launch of its AI satellites, ARK believes costs will optimize down to $32 billion per GW as manufacturing efficiency improves and weight is stripped from satellites.

What the Numbers Show

The data reveals a stark divergence between terrestrial and orbital infrastructure costs. While terrestrial data center costs are projected to rise above $76 billion per GW by 2032, SpaceX aims to reduce orbital deployment costs to $32 billion per GW through reuse and manufacturing scale. This suggests a potential cost advantage of more than 50% for space-based AI compute compared to ground-based alternatives, assuming the projected launch cadence and reusability targets are met.

SPCX shares fell 0.91% to $143.29 during pre-market trading on Tuesday.

SpaceX CEO Elon Musk mocked critics who claim artificial intelligence data centers cannot operate in orbit due to cooling limits, arguing they lack understanding of basic thermal constraints.

Musk posted on X on August 31, 2026, that self-described experts dismiss orbital AI without knowing fundamental parameters such as heat rejection per square meter of radiator, coolant temperature, or maximum operating temperature of graphics processing units.

Response to Skepticism

The CEO was responding to investor Gavin Baker, who suggested physicists should be cautious about declaring orbital computing impossible after only a few hours of study. Baker noted that Space Exploration Technologies Corp. (NASDAQ: SPCX) engineers have spent significantly longer addressing the problem.

Baker cited a quote from Keanu Reeves: "Even if you say 1+1=5, you’re right. Have fun."

Engineering Challenges

Cooling remains a significant engineering hurdle for spacecraft, which cannot dump heat into surrounding air and must radiate waste heat into space.

Reuters reported in April that experts question whether SpaceX can economically manage heat, radiation exposure, launch costs, and hardware obsolescence.

Orbital Compute Strategy

SpaceX states its AI1 satellite is designed around these thermal constraints. The company’s Starmind page lists up to 250 kilowatts of peak compute and 175 kilowatts average power, alongside solar power and radiators.

SpaceX has asked the Federal Communications Commission to authorize up to 1 million orbital data-center satellites between 500 and 2,000 kilometers above Earth.

Industry Support

Musk has previously called solar-powered orbital data centers a "no-brainer" and predicted space would become the lowest cost place for AI within two or three years. He later stated, "Space is the only way to scale at scale."

NASA Administrator Jared Isaacman backed orbital data centers, arguing they could tap abundant solar energy to help the U.S. compete in AI and space races.

Lawrence Berkeley National Laboratory estimates data centers could consume 11.8% of U.S. electricity by 2030, with scenarios ranging from 9.5% to 15.3%.

What the Numbers Show

The divergence between Musk’s confidence in managing thermal loads via specific radiator and coolant metrics and expert skepticism regarding economic viability highlights a key tension in scaling AI infrastructure beyond Earth.