SpaceX S-1: Audited X and Grok Data, 2026

This post anchors to the SpaceX S-1 filing released today (May 20, 2026) — the first SEC-audited X and Grok metrics since Twitter went private in October 2022. For three and a half years, every X user number was a management claim without auditor sign-off; the S-1 consolidation changed that in a single 400-page document. X reported 550M monthly active users as of March 2026, Grok reported 117M MAU (21.3% of the X ecosystem), and xAI posted a $6.4B operating loss on $3.2B of revenue — a 4× deterioration from FY2024's $1.56B loss.

The financial stakes extend well beyond user counts. xAI's $12.7B capex in FY2025 — nearly four times its segment revenue — accelerated into Q1 2026 with a further $7.7B in a single quarter. The SEC EDGAR filing also disclosed a $1.25B-per-month Anthropic compute arrangement through May 2029, an option to acquire Cursor at up to $60B equity value, and Musk's 85.1% post-IPO voting control via Class B shares. Nasdaq listing is targeted for June 12, 2026 under ticker SPCX at a reported $1.75T pre-IPO valuation seeking ~$75B.

This guide covers every audited number in one reference table, the xAI unit-economics story, the Colossus capex trajectory, the subscription and ad revenue arcs, the path-to-profitability math, a cross-lab burn comparison, and the structural mechanics of both the Anthropic deal and the Cursor option. Source tiers are labeled throughout: audited S-1 line items are distinguished from management disclosures and third-party forecasts. For the platform-scale breakdown — engagement, content, and brand-safety metrics — see the full S-1 stats recap.

The table below assembles every material metric from the SpaceX S-1, labeled by source tier. "Audited" means the figure appears in SEC-filed financial statements signed by auditors. "Management" means a disclosure in the prospectus narrative without auditor certification. "Third-party" means an external estimate not part of the filing itself. Understanding the distinction matters: most downstream coverage treats all three tiers as equivalent.

One scope clarification that caused widespread confusion on filing day: the S-1's $116M "AI-segment advertising revenue" line covers only the xAI-segment ad business — not the full X-platform global ad operation. The eMarketer $2.26B figure is a third-party forecast of the full platform. These are different scopes measuring different things; subtracting one from the other produces a meaningless number.

The xAI segment's FY2025 loss-to-revenue ratio is straightforward arithmetic at the segment level: $6.4B operating loss ÷ $3.2B revenue = approximately $2.00 of loss for every $1.00 of revenue. That ratio deteriorated sharply from FY2024's $1.56B loss on $2.62B revenue — roughly $0.60 of loss per revenue dollar. In a single year, xAI went from burning less than it earned to burning twice as much.

A widely-cited alternative framing — "$26 of loss per $1 earned," attributed to Software Thug's AI lab burn analysis — uses a different denominator. That framing likely excludes advertising and data-licensing revenue and focuses on a narrower product-revenue scope. Both figures are consistent with the same underlying S-1 data; the difference is scope definition. Cite both with their scope labels rather than treating them as contradictions.

The structural story is the capex-to-revenue multiple. xAI spent $12.7B in capital expenditure during FY2025 on $3.2B of segment revenue — a 3.97× ratio. Traditional software companies operate capex at 5-15% of revenue; even capital-intensive cloud infrastructure businesses rarely exceed 40-50%. xAI at 397% is a bet-the-company infrastructure buildout, not a margin story. The question is whether revenue scales faster than the infrastructure depreciation curve.

The Colossus cluster — xAI's Memphis, Tennessee GPU supercomputer — was built in 120 days and disclosed in the S-1 with 220,000+ NVIDIA GPUs (H100, H200, GB200 variants) and 300MW of power capacity for the first phase. The S-1 references a combined Colossus 1 and Colossus 2 capacity of approximately 1 gigawatt, per TechCrunch's filing-day teardown. The S-1 does not separately itemize Colossus 1 vs Colossus 2 capex; the $12.7B FY2025 figure covers the full xAI segment.

The Q1 2026 capex of $7.7B — a single quarter — is the number that reframes the FY2025 annual figure as a floor rather than a ceiling. Mechanically, $7.7B annualizes to approximately $30.8B if Q1's pace is sustained through 2026. Whether that pace sustains depends on the Anthropic compute deal utilization, Grok model roadmap milestones, and the IPO proceeds timing. The Anthropic-Colossus deal covered by The Verge essentially monetizes a portion of this buildout: Anthropic retains ownership of its content, models, and data under the arrangement, while paying $1.25B per month for compute access.

The S-1 discloses 6.3M paid subscribers as of March 2026: 4.4M in the X Premium and Premium+ tiers, and 1.9M across the SuperGrok family (SuperGrok Lite at $10/month, SuperGrok at $30/month or $300/year, and SuperGrok Heavy at $99/month for a six-month introductory period, then $300/month list). The filing does not break the 1.9M Grok count by tier — the $30/month and $300/month plans are materially different revenue contributions per subscriber, but the S-1 treats them as a single aggregate.

The subscription-revenue picture has three data points at different scopes and dates. The S-1 discloses a $365M increase in X + Grok subscription revenue in FY2025 — this is a year-over-year change figure, not a total. In February 2026, The Information reported X+Grok subscription ARR at approximately $1B (an annual run rate based on management disclosure). Q1 2026 added a further $177M sequential increase — annualizing from that quarter alone to roughly $708M, which sits slightly below the February $1B ARR figure in a way consistent with the subscription business still accelerating through Q1. The $1B ARR figure from The Information is the canonical run-rate anchor for this business entering 2026.

The interpretation that matters for the path-to-profitability analysis: subscriptions are the only revenue line inside xAI that is structurally scaling with both user count and product innovation. If SuperGrok Heavy retention holds at $99-300/month for a non-trivial fraction of the 1.9M Grok subscribers, the average revenue per Grok subscriber could be meaningfully higher than the per-user X Premium ARPU. The S-1 gives us neither the tier split nor the retention data to confirm that hypothesis — but the $1B ARR against 1.9M Grok subscribers implies an average of roughly $44/month across all Grok tiers if all ARR is attributed to Grok, which is implausibly high. The $1B ARR clearly spans X Premium and Grok combined.

The most quotable single sentence from a market-context perspective: eMarketer forecasts X's FY2025 global advertising revenue at $2.26B — the first projected annual increase since Musk's October 2022 acquisition. That represents a +16.5% year-over-year recovery. The US-only forecast is $1.31B. Pre-Musk Twitter reported $4.51B in audited advertising revenue in FY2021 (its last full public-company year, per the Twitter, Inc. FY2021 10-K filed with the SEC). $2.26B ÷ $4.51B = 50.1% — three years after the acquisition, X's advertising business is at half of pre-Musk scale.

The recovery narrative is real but contextually undersized. A +16.5% annual increase from a depressed base still leaves the platform at half its prior peak. The advertiser exodus that followed the 2022 acquisition — driven by brand-safety concerns, content moderation changes, and verification policy shifts — appears to have partially reversed, but the structural reset in the advertiser base means the $4.51B 2021 level is not a near-term target. For the advertiser-facing analysis of X's ad revenue trajectory, including brand-safety enforcement data from the S-1 and what the recovery means for platform planning, see the dedicated Day 5 advertiser brief.

The chart above makes the scope disparity visually explicit. The $4.51B and $2.26B figures are both full-platform global ad-revenue numbers, measured at different points in time. The $116M S-1 figure is an xAI-segment sub-line and is not additive to the eMarketer estimate. All four bars use the 2021 Twitter figure as a 100% baseline to show the relative recovery scale — which is the editorially relevant comparison — not to imply the bars are comparable on the same dimension.

The breakeven arithmetic from the FY2025 segment data is straightforward. With $3.2B revenue and $6.4B operating loss, the xAI segment needs either (a) revenue to approximately triple — to roughly $10B — at current cost structure, or (b) costs to fall by approximately 50% at current revenue. Neither path is near-term.

The revenue-tripling path depends on Grok adoption scaling materially beyond the current 117M MAU and 1.9M paid subscriber base, plus the Anthropic deal sustaining — or expanding — its $1.25B/month compute rate. If Anthropic exercises its 90-day termination clause, the compute revenue disappears and the breakeven revenue target rises further. The cost-cutting path is constrained by the infrastructure depreciation schedule: $12.7B of FY2025 capex and $7.7B of Q1 2026 capex will amortize into the income statement over multi-year hardware lifetimes regardless of operating decisions. Cutting R&D or headcount doesn't move the capex depreciation needle.

A more nuanced read is that breakeven is not the near-term investor thesis. SpaceX's filing presents xAI as a growth infrastructure investment — Starlink's $4.4B operating income partially subsidizes the xAI buildout. The $75B IPO raise, if successful, provides additional runway. The question is whether the Grok product and the Colossus compute-as-a-service business can demonstrate a credible inflection in the FY2026-2027 numbers before the capital window narrows. For the xAI-specific deep dive on the $6.4B loss and Anthropic compute deal economics, see the dedicated Day 5 analysis.

Comparing frontier AI lab burn rates requires careful source-tier labeling. xAI is the only lab with SEC-filed FY2025 financials. OpenAI's figures come from internal documents reported by Fortune in November 2025. Anthropic's trajectory comes from its own internal forecasts published via Software Thug's AI lab financial analysis. Neither OpenAI nor Anthropic has filed public financial statements with the SEC; the audit-quality gap is significant for anyone building financial models from these numbers.

With those caveats explicit: xAI's ~200% segment-level burn ratio (operating loss as a percentage of revenue) is the structural outlier. OpenAI reportedly posted approximately $9B in losses on $13B of revenue in FY2025 — a roughly 69% burn ratio — and projects $74B in operating losses by 2028. Anthropic's trajectory is moving in the opposite direction: burn reportedly around 33% of revenue in 2026 and projected to fall to 9% by 2027. xAI and Anthropic represent the two ends of the frontier-lab financial discipline spectrum.

The chart uses xAI FY2025's 200% as the scale-setting bar precisely because it is the only audited figure in the set. OpenAI and Anthropic figures are presented with different disclosure quality and should be weighted accordingly in any financial analysis. The directional story — xAI burning hardest, Anthropic converging toward operational discipline — is consistent across multiple secondary sources even if the exact percentages remain uncertain for the non-SEC filers. The Q2 2026 inference-pricing matrix covers the compute-economics context driving these cost structures.

The Anthropic compute arrangement disclosed in the S-1 is structured as follows: Anthropic pays approximately $1.25B per month for xAI compute capacity (Colossus cluster access) through May 2029. At $1.25B/month over the full term, the headline value reaches approximately $40-45B (secondary sources vary slightly on the total; $45B is the more commonly cited figure). The S-1 notes that Anthropic retains ownership of its content, models, and data under the arrangement.

The structural nuance that most coverage omits: the deal carries a 90-day mutual termination clause. Either party may exit on 90 days' notice. At $1.25B/month, 90 days of notice period equals approximately $3.75B of contractually firm minimum value — not $40-45B. The $45B headline is the value of the contract if neither party exercises termination through May 2029; it is not a locked-in revenue commitment.

This distinction matters significantly for how investors and analysts should model xAI's revenue trajectory. If Anthropic develops sufficient in-house or alternative compute capacity — or if the commercial relationship sours — the termination clause converts a multi-year revenue anchor into a 90-day exposure. The $3.75B floor is meaningful but changes the valuation narrative considerably from the $45B headline. The phrase "revocable purchase order" — used in BitMEX's S-1 teardown — is an accurate characterization of the economic reality.

The SpaceX × Cursor arrangement disclosed in the S-1 (and corroborated by VC Corner's teardown and BitMEX's IPO guide) has two components. First: a $10B services arrangement — $1.5B upfront termination fee plus $8.5B in deferred services fees — that provides Cursor with meaningful downside protection if SpaceX walks away from the commercial relationship. Second: an option for SpaceX to acquire Cursor at up to a $60B implied equity value (Class A SpaceX stock), exercisable approximately 30 days post-IPO.

The financial-engineering reading: the $10B structure is effectively a guaranteed minimum transaction value for Cursor's shareholders regardless of whether the acquisition option is exercised. If SpaceX exercises the $60B option, Cursor shareholders receive SpaceX equity at a $60B implied value. If SpaceX does not exercise, Cursor retains the $10B in committed commercial value ($1.5B termination fee + $8.5B deferred services). The structure aligns incentives: SpaceX has reason to grow Cursor's value toward the $60B ceiling (making the option worth exercising), and Cursor has reason to build Colossus integration into its product (making it valuable to SpaceX).

The puzzle: the $60B valuation for Cursor represents a significant premium to any private-market comparable in the AI-coding category as of early 2026. The option is exercisable in SpaceX Class A stock — meaning Cursor shareholders bear the SpaceX-valuation risk on the equity they receive. If the post-IPO SpaceX stock trades below the implied $1.75T pre-IPO valuation, the $60B Class A option is worth proportionally less in market value. The H1 2026 AI-coding retrospective provides context on Cursor's market position relative to Claude Code and Codex CLI that informs how analysts should think about the $60B ceiling.