SpaceX’s AI Division Generates $7.56 Billion Overtaking Starlink

Morning, {{first name | folks}}! AI is expanding into new territory this week, from SpaceX’s booming computing business to language models heading for the Moon. Meanwhile, OpenAI is moving deeper into drug development, Anthropic’s IPO filing reveals the financial complexity behind its growth, and new research challenges how AI fairness is measured.

SpaceX generated $7.56 billion in third-quarter revenue from its “Super Intelligence” division, up from $813 million a year earlier. The division brought in more than Starlink’s $5.25 billion, helping push SpaceX’s total revenue to $14.58 billion, up 161% year over year.

Most of the division’s revenue came from renting data-center computing capacity to companies including Anthropic, Google, and Reflection AI, rather than selling its own AI models. SpaceX spent $16.44 billion on AI infrastructure during the quarter, underscoring the enormous investment behind its push to become a major AI computing provider.

Astrolab plans to send its FLIP rover to the Moon later this year, carrying a computer from Hewlett Packard Enterprise that can run AI models on the lunar surface. The mission will put large language models, image analysis, and laser-based mapping to the test, with NASA also planning experiments onboard.

Running AI on the Moon could let future missions analyze images, search technical documents, and process scientific data without waiting for instructions from Earth. But the hardware must first handle lunar radiation, extreme temperatures, and limited computing resources. The upcoming mission will test how much AI can actually do in those conditions.

OpenAI has acquired Orchestra Bio, a startup that builds AI agents for biotech and pharmaceutical companies. Its team and technology will join OpenAI’s Rosalind platform, which helps scientists tackle life sciences research.

Orchestra’s software helps teams track drug studies, monitor manufacturing, and catch problems before they delay clinical trials. A manufacturing setback or missed milestone can push a drug program back months, and OpenAI is betting that AI can help research teams manage those moving parts before delays pile up.

Anthropic generated $16.3 billion in revenue in the first half of 2026, up from $1.1 billion a year earlier. Yet its draft IPO filing reports an $80.7 billion net loss, largely driven by accounting charges tied to fundraising instruments rather than ordinary operating expenses.

The company also spent nearly $9 billion on computing infrastructure in six months, while flagging potential future commitments of up to $750 billion. The figures reveal two sides of Anthropic’s growth: surging demand for its AI and the enormous financial commitments required to support it.

Researchers at Ai2 have introduced FairStress, a benchmark designed to expose bias that standard AI fairness tests can miss. Across scenarios involving hiring, healthcare, lending, and housing, the researchers found that a model’s initial decision could look fair, yet its response to follow-up prompts could vary depending on the demographic group involved.

The team tested nine methods to reduce these disparities, but most failed to address them consistently, and some lowered accuracy. That makes fairness harder to establish through a single test: models need to be evaluated across different prompts and scenarios to reveal biases that might otherwise go unnoticed.