xAI lost $6.36 billion in 2025 — and it's the only frontier AI company with audited books

Three segments, one IPO: where the money goes

SpaceX's prospectus splits the combined entity into Connectivity (Starlink), Space (launch services), and AI (xAI plus X). The contrast between segments is the story investors will price on June 12 — and the template every other AI IPO will be measured against.

Connectivity is the cash engine. Starlink generated $11.4 billion in 2025 revenue, up 49.8% year over year, with segment adjusted EBITDA of $7.17 billion at a 63% margin, per Finrep's S-1 breakdown. This is the business subsidizing everything else.

Space (rockets, Starship development, government contracts) posted $4.09 billion in revenue and $653 million in EBITDA — profitable at the margin but capital-intensive.

AI is the drag. Revenue grew 22.2% to $3.2 billion in 2025, but operating losses widened from $1.56 billion in 2024 to $6.36 billion in 2025. Capital expenditure in the AI segment totaled $12.73 billion for the year — more than half of SpaceX's consolidated $20.74 billion capex. Q1 2026 AI capex alone hit $7.72 billion, dwarfing Connectivity ($1.33 billion) and Space ($1.05 billion) combined.

The consolidated picture: SpaceX swung from a $791 million net profit in 2024 to a $4.94 billion net loss in 2025, almost entirely because xAI's Colossus data centers came online. Q1 2026 consolidated net loss: $4.28 billion on $4.69 billion in revenue. Starlink prints cash; Grok burns it faster than any software company in history.

Colossus economics: revenue vs. depreciation

xAI's revenue is real but structurally mismatched to its cost base. The S-1 describes Colossus I coming online in 122 days with roughly 100,000 Nvidia H100 GPUs and 130 megawatts of power, followed by Colossus II with 110,000 GB200s in 91 days and a further 220,000 GB300s planned. Grok-5 is training on Colossus II toward multi-trillion-parameter scale. The X platform contributes 550 million monthly active users and roughly 350 million daily posts — distribution that no pure-play AI lab can match.

Yet the accounting tells a harsher story. AI segment revenue of $3.2 billion in 2025 implies a negative 199% operating margin — losses nearly double revenue. That is not a SaaS business with high gross margins waiting to scale; it is an infrastructure build-out where GPU depreciation, power, and facility costs hit the income statement before inference revenue can catch up. The $30 billion Google GPU lease disclosed in the same filing shows xAI is also a landlord — renting roughly 110,000 chips to Alphabet at $920 million per month — but even that contract revenue does not offset the capex trajectory.

Regulatory friction adds cost without showing up as a line item. The S-1 reveals plans to purchase $2.8 billion in gas turbines over three years for AI infrastructure, including mobile units — even as the NAACP sued xAI over unpermitted turbines near Memphis and the EPA found federal violations, per The Decoder. Power is not a footnote in frontier AI economics; it is a contested input that can delay capacity and invite litigation.

What this means for OpenAI and Anthropic

OpenAI and Anthropic have filed confidential draft S-1s — their financials remain sealed while the SEC reviews. Private-market narratives dominate: Anthropic's $47 billion annualized revenue run rate and first operating profit; OpenAI's superapp pivot and reported $6 billion Q1 revenue. None of that is audited. xAI's numbers are.

The comparison investors will make is not flattering to private valuations:

• Loss intensity: xAI's 2025 operating loss was 1.99x revenue. If OpenAI or Anthropic disclose similar ratios at $850–965 billion valuations, the multiple expansion story collapses.
• Capex dependency: xAI spent $12.7 billion on AI capex in 2025 and another $7.7 billion in Q1 2026 alone. Frontier labs without Starlink's cash flow must fund this through equity dilution — exactly what the IPO window provides.
• Cross-subsidy unavailable: OpenAI rents Azure capacity from Microsoft; Anthropic runs on AWS and Google Cloud. Neither owns a profitable satellite broadband business throwing off $7 billion in annual EBITDA. Their path to profitability requires either dramatically higher inference margins or a capex slowdown neither has signaled.

Our Anthropic-OpenAI IPO race analysis noted that whoever lists first may set the valuation template. SpaceX listing first with xAI embedded actually front-runs both — and the template says: enormous revenue growth, even more enormous losses, and a bet that infrastructure scale eventually produces margins no one has demonstrated yet. The mega-IPO skepticism piece we published earlier this month asked whether public investors would fund capex at these multiples. xAI's audited P&L is the first hard evidence in that debate — and it argues the skeptics have a point.

Why crypto and equities care

The AI IPO wave is not a siloed tech story. NYDIG's Greg Cipolaro identified six overlapping headwinds depressing Bitcoin below $60,000 — and “high-profile tech IPOs” sit near the top. Institutions raise cash and reduce risk positions ahead of mega-offerings. SpaceX alone targets a $75 billion primary raise; OpenAI and Anthropic could add another $50–100 billion combined if they price at private marks. That is capital rotating out of liquid alternatives — including crypto ETFs, which shed $1.72 billion in the first week of June.

The xAI loss disclosure sharpens the mechanism. Investors who buy SPCX on June 12 are not buying a rocket company; they are buying a consolidated entity where AI losses consumed Starlink's entire 2025 operating profit and then some. If the IPO prices at $1.75 trillion despite $4.9 billion in consolidated net losses, public markets are explicitly betting that AI capex is an investment, not an expense. That bet either validates the private valuations OpenAI and Anthropic carry — or sets up a brutal repricing when their sealed financials finally surface.

Three scenarios for AI IPO pricing

Scenario A — “Infrastructure premium” holds (35–40% probability): SpaceX prices at or above the $1.75 trillion target despite xAI losses. Investors treat AI capex like telecom build-outs in the 1990s — painful early losses, monopoly rents later. OpenAI and Anthropic IPOs price at private marks or higher. Crypto and growth equities stay under pressure through Q3 as capital absorbs the supply, but AI multiples expand.

Scenario B — Loss disclosure triggers skepticism (40–45% probability): xAI's audited numbers become the anchor. Buy-side analysts model OpenAI and Anthropic at 15–20x revenue with explicit loss assumptions, cutting private valuations by 20–30% before public S-1s drop. SpaceX trades flat or down from IPO price as investors price AI as a drag, not a growth engine. Bitcoin stabilizes as IPO-driven selling exhausts itself post-June 12.

Scenario C — Bifurcated market (15–20% probability): Starlink/Space segments command premium valuations; AI segment trades at a discount within the SPCX conglomerate structure. Pure-play OpenAI and Anthropic IPOs delay into 2027 as underwriters wait for a profitability narrative. The catalyst superweek passes without resolving the AI valuation question — leaving risk assets range-bound until Q3 earnings reveal whether enterprise AI revenue is scaling faster than capex.

What to watch next

• SpaceX IPO pricing (June 11) — does the $1.75 trillion valuation hold after roadshow investors digest xAI segment losses?
• SPCX first-day trading (June 12) — aftermarket performance sets the tone for the entire AI IPO calendar.
• OpenAI public S-1 (expected late July–August) — first unredacted OpenAI financials; compare operating loss ratio to xAI's 199%.
• Anthropic public S-1 (expected August–September) — does the claimed first operating profit appear in audited statements?
• Q2 2026 xAI segment update — if SpaceX files quarterly reports, watch whether AI revenue growth outpaces loss acceleration.

Private markets priced frontier AI on growth curves and strategic positioning. Public markets price on audited cash flows. xAI's $6.36 billion operating loss is the first time those two worlds have collided — and every other AI IPO in the pipeline will be judged against it. The question is not whether frontier AI is valuable. It is whether $850 billion to $2 trillion valuations can survive contact with the income statement.

Sources: Dave Manuel — SpaceX S-1 analysis (May 20, 2026); PitchBook — xAI financials in SpaceX IPO (Jun 2026); Finrep — SpaceX segment breakdown (2026); The Decoder — AI losses and turbine spending (2026); TechTimes — OpenAI September IPO target (Jun 7, 2026).