Grok's Washington stumble punctures part of SpaceX's AI premium

SpaceX’s IPO pitch has found a new weak point. Not launch cadence, broadband churn or even the usual Musk discount. The weak point is xAI. A Reuters report on Thursday found only three Grok or xAI references across more than 400 public federal AI use cases, giving sceptical investors a cleaner way to ask how much of SpaceX’s valuation really deserves an AI premium.

Why that matters is straightforward: the public-market story around SpaceX has widened well beyond rockets and even beyond Starlink. The company still has the sort of assets that can carry a blockbuster float on their own: a dominant launch business, a profitable satellite network and a founder who can still command attention. More recently, though, the pitch has stretched to include the idea that SpaceX can also house, fund and distribute a serious AI business through Grok and related infrastructure. Reuters’ Washington snapshot suggests that case remains thinner in practice than it looks in pitch decks.

Federal procurement teams read the problem differently from equity buyers. In their telling, Grok’s issue is not price. The General Services Administration said agencies could use the model for $0.42 each under a OneGov agreement, yet FedScoop reported that higher-level security authorisation could still take years. In Washington, a cheap pilot matters only after trust, compliance and workflow fit are in place. Grok appears to be trying to skip that order.

“rapidly deploy AI throughout the government for the benefit of the country”

— Elon Musk, via Reuters

For analysts, the pricing detail is almost secondary. More important is the idea that if Grok cannot win meaningful usage inside one of the world’s most process-driven technology buyers, the market may start to split SpaceX’s AI assets into two buckets: product promise on one side, monetisable infrastructure on the other. Less glamorous. Probably lower-multiple too.

The numbers investors have been given already point that way. Bloomberg reported that xAI posted a $6.4 billion operating loss in 2025 on $3.2 billion of revenue, a profile that looks more like a capital sink than a mature software business. Axios argued that the hotter parts of the SpaceX story are still doing most of the financial work. Analysts are no longer asking whether AI belongs in the story at all. They are asking whether it deserves to lift the valuation or sit beside the cash engines as an expensive option.

From the procurement side, the story answers an obvious question: why did a discounted deal not translate into visible adoption? Because Washington does not buy language models the way consumers buy chatbots, and it does not even buy them the way large companies do. Agencies buy audit trails, accreditation, data handling, integration history and reputational safety. On that score, Grok arrived with a low sticker price but little evidence that it had cleared the trust hurdle.

Why cheap access has not moved agencies

In software markets, aggressive pricing can manufacture momentum. In Washington, it often does the opposite: it signals that the harder part of the sale still sits elsewhere. Reuters found enterprise Grok use running at two per 1,000, down from five per 1,000 at its peak. More than a disappointing adoption curve, it suggests agencies are testing Grok at the edges while keeping more sensitive workflows with vendors they already understand.

Vineet Jain of Egnyte called the federal data a “canary in the coal mine”. The phrase lands because federal buyers are rarely first movers, but they are good at exposing which vendors have institutional depth and which ones are still running on executive sponsorship. Musk can attract meetings. Procurement officers still need to be comfortable that a tool will not create a policy problem six months later.

Regulators make that hurdle harder, not easier. Grok has already drawn scrutiny over antisemitic and other abusive outputs, including a Senate demand for answers. Every federal pilot therefore comes with a second question: not only can the model do the job, but can the agency defend choosing it if something goes wrong? A $0.42 entry price barely matters once that becomes the frame.

“far out of proportion to its market share”

— Page Hedley, Guidelight AI Standards, via Wired

Put that sceptical view inside the IPO debate and the discount-rate issue becomes easier to see. A small software product can still inject a large governance premium into a public offering if it is tied to the controlling shareholder, marketed as a growth lever and exposed to political scrutiny. Wired’s reporting on former OpenAI staffers warning SpaceX investors was not about current market share. At issue was Grok’s safety record sitting in the same file as launch economics and broadband subscriber growth. After Reuters’ Washington story, that no longer looks like a fringe concern.

One basic market-structure point follows. Federal adoption is not the whole AI market, and it does not need to be. But it is a useful stress test for enterprise credibility because the customer is large, documentation-heavy and reputationally conservative. If Grok cannot yet clear that bar despite a subsidised path in, investors have to assume the commercial path may prove slower than the rhetoric around Musk’s AI empire suggests.

The better AI business may be compute

None of this kills the AI case around SpaceX. It changes where the strongest economics seem to live. The clearest evidence of AI demand in the filing cycle has not come from Grok subscriptions or public-sector adoption. Instead, it has shown up in infrastructure. Wired reported that Anthropic agreed to pay SpaceX $1.25 billion a month through May 2029 to access its computing capacity. That is a very different kind of AI business, closer to toll-road economics than to software leadership, but it is tangible.

From an analyst’s seat, that deal partly answers a core question: is the AI segment option value or a drag? Both, probably. Grok itself can still consume capital, generate headlines and struggle for adoption, while the physical backbone built around the broader AI push throws off real revenue from outside tenants. SpaceX would not be an AI loser in that scenario. It would be something more prosaic: a company that can monetise the build-out even if its in-house model trails OpenAI, Anthropic or Google.

For the IPO, that cuts both ways. Public investors like visible cash flows. Infrastructure businesses, though, usually do not command the same speculative premium as a category-winning model platform. If the market starts to conclude that the most bankable AI line inside SpaceX is rented compute rather than Grok adoption, the bull case shifts from software dominance to capacity ownership. Valuable, yes. The multiple is different.

The contrast with Starlink also sharpens. Starlink already looks like an operating business with scale, pricing power and a widening addressable market. xAI still looks like an ambition that needs time, capital and evidence. Bloomberg’s loss figures and Reuters’ federal adoption data point in the same direction: the AI story is not broken, but it is not yet self-validating either. Public investors may accept that gap for a while, especially in a hot deal. They are less likely to ignore it if comparable AI names begin to trade on execution instead of charisma.

So the Washington stumble matters most here. Reuters did not reveal a fatal defect in SpaceX’s IPO. Instead, it showed which part of the narrative is still aspirational. SpaceX can probably still sell investors on launches, satellites and connectivity. What looks harder is asking them to pay up now for a chatbot whose cleanest validation has yet to come, and whose strongest current economics may lie in the servers around it rather than in the model itself.

The fairest read for now is that SpaceX’s AI premium should narrow, not disappear. Grok’s weak federal footprint does not mean xAI cannot improve, or that government demand will never follow. But the public-market case has to be phrased more carefully. The company is not offering a pure AI winner. It is offering a hybrid: one proven cash engine, one potentially lucrative infrastructure arm and one flagship model that still has to earn trust in exactly the kind of institution that makes hype hardest to sustain. In an IPO this large, that distinction is not a side note. It is the valuation story.