Three businesses share one valuation but only one produces cash
The SpaceX S-1 reveals three distinct businesses bundled under one valuation, and only one of them—Starlink—makes money.
That fact sits uncomfortably beside a targeted $75 billion raise at a $1.75 trillion valuation, a number that depends less on what SpaceX is than on what it claims it will become. Starlink generates $1.2 billion in quarterly profit, underwriting most of the company’s $18.7 billion in 2025 revenue. The launch division, despite dominance, loses money. The AI segment, assembled from xAI and X, posted a $4.28 billion net loss in Q1 2026, matching its entire loss the previous year. The valuation treats these as one business. The cash flows do not.
The gap is being bridged, for now, by a contract. SpaceX is receiving $1.25 billion per month from Anthropic for cloud compute, a deal tied to the Colossus supercomputer that could add $2.5 billion in quarterly revenue if it scales. That revenue sits inside the same AI unit that is currently absorbing billions in losses. The implication is mechanical: without that inflow, the loss profile is not narrowing. It is widening.
Markets and policy move at different speeds but shape the same behaviour
This is what “AI is inside every number” looks like in practice. AI is inside every number is not a slogan here; it is an accounting condition. The fastest-growing line of revenue is tied to a single counterparty, in a segment that has yet to demonstrate operational stability. The rest of the business—launch, hardware, infrastructure—remains capital-intensive and loss-making. The profitable core is doing something else entirely: subsidising the experiment.
That structure would be easier to carry if the surrounding market were static. It is not. In the UK, where policymakers are attempting to shape the same technological wave, consumption patterns are already shifting faster than regulation. TikTok Shop saw a 50% year-on-year increase in UK sales during Black Friday and Cyber Monday 2025, driven by friction-free in-app purchasing that is changing consumer habits. The same design features that accelerate commerce—continuous feeds, embedded transactions—also sustain attention. Technological features can be especially powerful in shaping young users’ behaviour, with mechanisms such as infinite scrolling, where new content keeps loading automatically, and autoplay systems that start the next piece of content without user action. The page does not end. The session does not close.
The policy response is arriving after the behaviour. In London, one proposed “announceable” is a ban on “harmful” social media for under-16s, timed deliberately; the prime minister wants to announce it before the byelection. The justification rests on developmental risk: if children are on social media at night instead of sleeping, it could have a negative impact on their development, in a context where sleep is essential for development. The intervention targets use. The business models target time.
The same government is, in parallel, embedding AI into the welfare state. The Department for Work and Pensions is trialling an “AI work assistant” that will provide a 24/7 resource for people navigating job searches, with a trial lasting around three months to gather data and feedback. Access requires infrastructure: a GOV.UK login. The pitch is efficiency and reach. The mechanism is data capture at scale.
All of this sits within an economy that is digitally mature but uneven in its outcomes. The United Kingdom ranked 10th out of 134 countries in the Networked Readiness Index 2023, a framework that reveals the correlation between ICT uptake and economic and social impacts. The capability exists. The distribution of gains remains contested. That tension is now political. “Keir has entered his legacy era”, an unnamed minister says, with conversations focused on what can be announced before Makerfield. The line being prepared for public delivery is simpler: “the tech revolution must work for everyone, not just a privileged few”.
The difficulty is that the underlying economics are already selecting their beneficiaries. The platforms that drive engagement monetise duration. The commerce layers monetise conversion. The infrastructure providers monetise compute. Each operates on scale, concentration and, increasingly, cross-subsidy. In SpaceX’s case, a profitable satellite network carries a loss-making AI division whose near-term viability depends on a single contract. In the UK, a consumer base that has seen over 87% of adults shop online in the past year is being drawn deeper into ecosystems designed to collapse discovery, persuasion and purchase into one continuous act.
Political timing is compressing these dynamics. Two weeks from now, it seems more likely than not that Andy Burnham will take his seat as MP for Makerfield, though a byelection win is not certain. The campaign is going his way, and with it the pressure on a leadership that will not stand aside but is already being assessed in terms of what can be fixed, or framed, before the vote. Technology policy, in this context, becomes something else: a sequence of announcements calibrated to a calendar.
That is where the structures begin to overlap. A government racing to define the terms of a “tech revolution” is doing so in an economy where the most aggressive forms of that revolution are already operational, and in a capital market where the largest technology offering in history is asking investors to price a bundle of businesses whose internal economics do not reconcile. The connective tissue is not rhetoric. It is dependency—on user attention, on data flows, on a small number of counterparties willing to fund loss-making growth in exchange for capacity.
SpaceX’s filing makes that dependency visible. Total accumulated deficit: $41.3 billion sits alongside a revenue line that, in its fastest-growing segment, depends on one client paying $1.25 billion a month to keep the system running at scale. Remove or weaken that contract and the valuation is no longer a question of future growth. It is a question of present solvency.
Cover photo Bruno Sanchez-Andrade Nuño CC-BY-2.0
