X Revenue: Who Is Subsidizing Whom?

The SpaceX S-1, filed May 20, 2026, is the first audited window into X's financial performance since Elon Musk took the platform private in October 2022. The headline finding for advertisers and marketers: X's ad revenue reportedly came in at $1.8B in FY2025 — down roughly $100M year-over-year and at only 39.9% of Twitter's pre-acquisition $4.51B FY2021 ad scale, according to Social Media Today's analysis of the filing and the Twitter 10-K FY2021.

At stake is the entire premise of Musk's $44B acquisition thesis: that Twitter was an undermonetized platform that a new owner could scale to $12B in ad revenue and $10B in subscription revenue by 2028. Three years into that five-year plan, according to Techdirt's Mike Masnick analyzing the filing, “the combined X/xAI advertising business is at somewhere under $3 billion.” What the S-1 reveals is not just how far short the numbers fall — it reveals who is keeping the lights on.

This post builds the cross-subsidy P&L crosswalk that no other coverage has assembled explicitly: the revenue waterfall inside the xAI/X AI segment, the Musk 2022 pitch deck vs. audited 2025 reality comparison, the structural inversion from an ad-led to an AI-compute-led business, and what this means for brands still buying X inventory. For platform-scale stats — MAU, daily posts, supported accounts — see our companion piece X platform statistics 2026: the audited S-1 numbers. This post is the financial layer.

The most important analytic discipline for anyone reading S-1 coverage is the revenue-scope disambiguation. Three separate ad-revenue figures appear in coverage of the filing, and they refer to meaningfully different things. Conflating them produces a distorted picture of X's advertising business.

$116M — the AI-segment narrow ad line. This figure comes from TechCrunch's Rebecca Bellan (May 20, 2026). It appears to represent the advertising revenue sub-line specifically allocated to AI/Grok-related ad inventory within the broader AI segment breakout. This is a narrow-scope line, not the full X-platform ad business.

$1.8B — the X-platform global advertising business. This figure comes from Social Media Today's Andrew Hutchinson (May 21, 2026): “X generated $1.8 billion in ad sales in 2025.” This is the broader X-platform heritage ad business — the Twitter-descended display and promoted-post inventory that brands buy. The $116M appears to be a sub-line within this $1.8B; they are not additive.

$2.26B — the eMarketer third-party forecast. This figure comes from eMarketer's 2025 X ad-revenue forecast, published before the S-1 filing. eMarketer projected $2.26B and 16.5% year-over-year growth — characterizing 2025 as the first projected annual increase since Musk's acquisition. The S-1 disclosure of the actual $1.8B figure (with a $100M YoY decline) contradicts the forecast. This is not unusual: third-party forecasts get superseded by audited filings, and eMarketer worked from available signals before the S-1 was filed. The $2.26B forecast is now superseded for FY2025 purposes.

The practical implication for media buyers: the audited X-platform ad business is reporting at $1.8B, not $2.26B. That is the figure to use in competitive channel analysis. And with a reported $100M YoY decline, the trend line matters as much as the level.

Musk's 2022 pitch deck — shared with billionaire co-investors at the time of the Twitter acquisition and later reported by the New York Times, Washington Post, and analyzed by Techdirt's Mike Masnick (May 22, 2026), projected a five-year transformation of Twitter into a $26.4B revenue business by 2028: $12B from advertising, $10B from subscriptions, and the remainder from data licensing. The 2022 deck was not a public disclosure; it was a fundraising document. Its projections were aspirational rather than underwritten. The S-1 now provides the first audited reality check against those projections.

Masnick's framing deserves direct quotation: “Three years into his five-year plan to reach $12 billion, the combined X/xAI advertising business is at somewhere under $3 billion.” That is the journalist's analytical rollup — not a single S-1 line item — but it is consistent with the $1.8B X-platform ad revenue figure from the filing. The subscriber attainment gap is sharper: Musk projected 69M paid Twitter Blue subscribers by 2025; the S-1 discloses 4.4M X Premium / Premium+ paid subscribers — 6.4% of the projection.

What the pitch-deck comparison does not prove is that the enterprise was misconceived from the start. Musk's 2022 projections may have assumed a faster AI tailwind, a different regulatory environment for advertising, or a more aggressive subscriber acquisition strategy. What the comparison does demonstrate is the scale of the execution gap — and why the revenue mix has shifted so dramatically from the original ad-and-subscription thesis toward AI infrastructure revenue (Anthropic) and external compute customers.

For the Grok product roadmap context that underlies the subscription growth, see our analysis of Grok 4.20's 2M-context multi-agent release — the prior-generation Grok model that the FY2025 AI capex was funding.

The framing question nobody else has built explicitly: in the combined X/xAI segment, which business is funding which? The conventional assumption — that Twitter/X's ad cash flow funds Grok's development — does not survive contact with the S-1 numbers. The math points in the opposite direction.

X's gross contribution to the AI segment — combining the reported $1.8B in X-platform ad revenue (per Social Media Today) with the $365M FY2025 subscription revenue increase — sums to approximately $2.165B. That covers roughly 17% of the xAI segment's $12.7B in FY2025 AI infrastructure capex. The other 83% is funded externally: through Anthropic's compute lease, data licensing customers, external AI buyers, and — most critically — SpaceX's Starlink and Launch operating income, which generated $4.4B in operating profit in FY2025 while the xAI segment generated a $6.4B operating loss. These are derived figures based on reported segment totals; the S-1 does not publish a formal X-vs-Grok P&L split.

The structural inversion, framed cleanly: pre-acquisition Twitter was an ad business with a thin subscription line. The 2025 S-1 reveals a segment in which X advertising is a declining minority revenue stream, subscriptions are growing (but still sub-$1B ARR), and the dominant revenue inflection is the Anthropic compute deal — an AI infrastructure deal that X the platform had no role in creating. The xAI talent-retention crisis (50+ engineer departures since the SpaceX merger) adds a human-capital dimension to this burn: the people building the AI infrastructure that subsidizes X are themselves a retention risk.

The derived per-X-MAU economics underscore the gap. $3.2B segment revenue divided by 550M X monthly active users (per the S-1) implies roughly $5.82 in revenue per X MAU per year. Meta's global ARPU runs approximately $70 per quarterly active user per their 2025 10-K — a fundamentally different monetization order of magnitude. These denominators differ (MAU vs. DAU), so the comparison is illustrative rather than apples-to-apples, but the directional gap is not small.

For the competitive AI model context that drives this capex — where Grok competes against Claude, GPT-5.5, and Gemini — see our AI model head-to-head comparison.

The single most important financial disclosure buried inside the X revenue story is the Anthropic compute deal. Per TechCrunch's May 20 coverage and Axios's confirmation, Anthropic is reportedly paying xAI $1.25B per month for access to xAI's Colossus data facilities, with the contract running through May 2029. Allie Garfinkle at Fortune (May 21, 2026) confirmed this is a cash payment arrangement: “Anthropic is apparently paying SpaceX a whopping $15 billion a year.”

At $1.25B per month, the annualized run-rate is approximately $15B per year and the total contract value through May 2029 is approximately $45B — subject to the clause details. Measured against the xAI segment's $3.2B in FY2025 revenue, the Anthropic deal at full run-rate would represent roughly 469% of that revenue base. In practical terms: if the deal were fully in effect for FY2025, it alone would have been larger than all other segment revenue combined by a factor of approximately 4.7×.

The critical qualifier is the 90-day mutual termination clause. The contractually firm minimum — the amount that cannot be walked away from on 90 days' notice — is approximately $3.75B. The $45B total contract value is the maximum if neither party exercises the termination option. Anthropic has separately disclosed approximately $80B in committed cloud infrastructure spending (per SaaStr) and is reported to be building its own data center capacity targeting late 2026. If Anthropic's own capacity comes online before May 2029, the economic rationale for paying $1.25B/month to xAI weakens materially.

The Anthropic deal also has a ramp structure: discounted rates during the May-June 2026 initial period. The FY2025 contribution from this deal was likely much smaller than the annualized run-rate suggests, since the deal was disclosed in the S-1 as new. The forward revenue impact — going from FY2025's partial contribution toward the full $15B/year run-rate — is the IPO narrative's load-bearing beam. If the deal terminates, that beam collapses.

The S-1 discloses total paid subscribers but does not break them down by tier. The filing's verbatim language (quoted in Social Media Today, May 21) is: “X and xAI currently have approximately 6.3 million active paid subscribers, comprising about 4.4 million X Premium and Premium+ paid subscribers and about 1.9 million SuperGrok, SuperGrok Heavy and SuperGrok Lite paid subscribers.” That gives us two buckets; the tier-level breakdown within each bucket is not disclosed.

The pricing stack spans a 33× price range across six active tiers — from X Premium Basic at $3/month to SuperGrok Heavy at $99/month introductory (then $300/month list). What follows is the verified pricing as of the filing date, drawn from X's pricing pages and cross-referenced with earlier research. The $99/month for SuperGrok Heavy is the real 6-month introductory promo price; $300/month is the list price. Do not confuse this with X Premium+ ($40/month), which is a separate tier.

The $177M sequential subscription revenue increase in Q1 2026 allows a rough implied ARPU derivation — though these are derived estimates, not platform-disclosed figures. If the Grok-side (1.9M subs) and X-side (4.4M subs) split the $177M sequential lift in proportion to their subscriber counts (roughly 30%/70%), Grok contributes approximately $53M per quarter to the lift, implying an annualized Grok-side ARPU of approximately $112 per Grok subscriber per year. That sits between the SuperGrok Lite $10/month and SuperGrok $30/month tiers — consistent with a mix weighted toward the $30/month standard SuperGrok tier. The implied X-side blended ARPU is approximately $28 per X subscriber per year — consistent with heavy weighting toward the $3/month Basic tier.

For the broader context of subscription-based revenue models in 2026, including how AI-powered SaaS businesses are structuring subscription tiers, see our 2026 subscription commerce and recurring revenue guide.

The xAI / X AI segment as reported in the S-1 breaks into four revenue streams, per CJ Gustafson's Mostly Metrics breakdown (May 21, 2026): X advertising, X+Grok subscriptions, data licensing, and selling AI compute to outside customers. The bars below show the itemized sub-lines disclosed by TechCrunch from the S-1's AI segment detail. Note that the TechCrunch scope ($116M AI-segment ad line, $365M subscription increase, $465M AI solutions, $88M data licensing) represents a narrower scope than Social Media Today's $1.8B X-platform ad figure. The two are not additive — they reflect different carve-outs of the same underlying business.

The segment's total FY2025 revenue was $3.2B (per TechCrunch, +22.2% from $2.62B in FY2024, as quoted by Techdirt citing the S-1 directly). Against a $6.4B operating loss and $12.7B in FY2025 capex, the segment's operating margin is approximately -200% and its capex-to-revenue ratio is approximately 4:1. Even the $1.2B adjusted EBITDA loss (per Bram Berkowitz at Motley Fool, May 21) — which strips out significant non-cash charges — reflects a business burning cash at a rate that X's revenue cannot approach covering.

The Q1 2026 picture is more acute. The segment generated $818M in revenue against a $2.47B operating loss — a -302% operating margin in a single quarter (per Yahoo Finance reporting on the S-1, May 21, 2026). AI segment capex in Q1 2026 alone was $7.7B — more than the entire FY2025 segment revenue of $3.2B in a single quarter of infrastructure spend.

For the Q2 2026 frontier AI model landscape that this capex is meant to support — including the Grok 5 roadmap — see our Q2 2026 frontier-model release tracker. The burn is an investment thesis: the infrastructure being built now is meant to power the AI models that will generate the revenue in 2027 and beyond.

Pre-acquisition Twitter's revenue mix was approximately 89% advertising and 11% data licensing (per Twitter's 10-K FY2021, filed with the SEC: $4.51B ad revenue and $0.57B data licensing out of $5.08B total revenue). There were no material subscription or AI infrastructure revenues because Twitter had no paid subscription product (Twitter Blue had just launched in June 2021 as a limited beta) and no AI compute infrastructure to sell.

The 2025 S-1 reveals a fundamentally different revenue mix in the AI segment. Using the TechCrunch itemized sub-lines as the available proxy: AI solutions and infrastructure ($465M) plus data licensing increase ($88M) together represent approximately 54% of the $1.034B itemized total. Subscription revenue increase ($365M) is 35%. The AI-segment narrow ad line ($116M) is 11% — a near-mirror inversion of the 89% advertising dominance in 2021.

This structural inversion is not merely a composition shift — it reflects a fundamentally different business model. Twitter's ad business scaled with user engagement and brand-safety perception; it declined when advertisers lost confidence in content moderation and brand adjacency. The replacement revenue streams — AI compute leasing and subscriptions — scale with infrastructure capacity and product value delivered, which are separable from the platform content controversy that damaged the ad business. The risk profile is also different: the ad business is slow to recover (brand-safety reputation rebuilds over years); the Anthropic deal is a contractual-but-terminable revenue line with a 90-day cliff.

For the Grok 4.3 product context — the current flagship model inside the AI segment — see our Grok 4.3 platform-asymmetry analysis. For the broader Q2 2026 Grok roadmap that Musk was articulating before the S-1 filing, see our Grok roadmap preview.

The S-1's ad-revenue disclosures are the most direct financial signal available to brand marketers evaluating X as a paid media channel. Three interpretive points stand out.

First, the $100M YoY decline is an audited disclosure, not a media narrative. Brand-safety concerns about X have circulated since 2022, but they have been debatable because X's financials were private. The S-1 makes them audited public record: the platform's advertising revenue declined approximately $100M year-over-year in 2025. The S-1 partially attributes Q1 2026's continued decline to an ad-platform overhaul — which could be a temporary disruption signal, or it could be masking structural erosion. Marketers now have an audited data point rather than anecdotal sentiment to anchor budget allocation discussions.

Second, the ad-platform overhaul mentioned in the S-1 is analytically significant for performance advertisers. If X restructured its advertising infrastructure in a way that “impacted ad sales for a short period of time during the rebuild” (per Social Media Today's S-1 summary), that rebuild could be positive for performance-campaign efficiency once complete. Or it could reflect a strategic pivot away from display advertising toward AI-mediated ad formats. The S-1 does not specify the nature of the overhaul. Marketers running active X campaigns should monitor CPM and CTR trends against the pre-overhaul baseline.

Third, X's revenue mix is shifting toward subscriptions and AI compute, not advertising. That means the platform's financial incentives are increasingly aligned with subscriber growth (Premium, Premium+, SuperGrok) rather than advertiser satisfaction. That incentive shift can manifest in platform-policy decisions — content moderation standards, ad adjacency policies, bidding-system design — that favor subscriber experience over advertiser brand safety. Brands allocating significant X budget should account for this structural incentive shift in their risk models.

For teams thinking about where agentic AI changes the SEO and paid media picture in 2026 — including how Grok-indexed content on X interacts with AI search surfaces — our agentic SEO services team is tracking the X/Grok content-discovery intersection as part of the broader AI-search landscape. Our paid media services team works with clients on cross-channel allocation decisions that account for platform-level financial and incentive risks of the kind the S-1 reveals.