SpaceX's Trillion-Dollar Gamble: Massive Losses, Musk's Control, and the Anthropic Wildcard
For every dollar SpaceX earned in the first quarter of 2026, the company lost 91 cents. That is the heart of the S-1 prospectus the company filed with the US Securities and Exchange Commission on 20 May — a $4.28 billion net loss on $4.69 billion of revenue. The gap covers what it costs to run a launch business, a satellite broadband empire, and the world's largest AI training cluster off one balance sheet. The same filing values the merged company at $1.75 trillion to $2 trillion ahead of a Nasdaq debut on 12 June under the ticker SPCX. The roadshow opens 4 June. Pricing follows on 11 June. The raise runs to as much as $75 billion — more than double the Saudi Aramco record of $29.4 billion that has held the largest-IPO-ever title since 2019.
The Filing In One Page
Bank of America, Goldman Sachs, Morgan Stanley, JPMorgan Chase and Citigroup are lead underwriters. About 30 per cent of the offering goes to retail allocation through Schwab, Fidelity, Robinhood, SoFi and E*Trade — an unusual tilt for a deal this large. A 5-for-1 stock split lands ahead of listing to pull the per-share price down to roughly $105 from the pre-split fair-market value of $526.59. Nasdaq's revised inclusion rules will let SPCX enter the Nasdaq 100 inside 15 days of listing, which means passive index funds will buy the stock irrespective of what the governance disclosures say. Polymarket traders are pricing a 51 per cent chance the first trading day closes between $1.5 trillion and $2 trillion. The retail allocation is the headline; the index buying is the engine.
The Numbers That Set The Stage
The S-1 shows a company that swung from a $791 million profit in 2024 to a $4.94 billion loss in 2025 — the same year revenue rose 33 per cent to $18.67 billion. The accumulated deficit on the balance sheet sits at $41.3 billion. Capital spending nearly doubled in twelve months, from $11.2 billion to $20.7 billion. The Q1 2026 net loss of $4.28 billion accounts for 86.6 per cent of the entire full-year 2025 loss compressed into 90 days. The Q1 2025 comparison loss was $528 million. Read the trajectory before reading anything else.
Period Revenue Net result Notes
Strip out the xAI segment and the legacy SpaceX returns to a roughly $13–14 billion annual run-rate at meaningful operating profit. Add xAI back and the same company prints $4 billion quarterly losses. The merger turned a profitable company into a loss-maker by design.
Three Businesses, One Ticker
The strangest thing about the filing is that it sells three structurally different businesses at one price. Three games run at the same table. Connectivity, anchored by Starlink, brought in $11.4 billion in 2025 — 61 per cent of all revenue, up 48 per cent from $7.7 billion in 2024, with $7.2 billion of adjusted EBITDA at a 63 per cent margin that beats every land-based telco on Earth. Space, anchored by Falcon 9 and the still-developing Starship, did $4.1 billion in revenue and an operating loss of $657 million as Starship development eats the launch business's profits. The Artificial Intelligence segment, absorbed in February 2026 when SpaceX paid $250 billion in stock for xAI, contributed $3.2 billion of pro-forma revenue against a $6.4 billion operating loss — close to three dollars lost for every dollar earned.
Sold apart, the three businesses would carry three different price-to-sales multiples and three different investor bases. Sold together, the merged company gets the AI-infrastructure multiple while running on the satellite-broadband cash flow. The S-1 floats a $28.5 trillion total addressable market — language the company calls "the largest actionable TAM in human history" — and roughly 93 per cent of that number sits inside the AI segment's enterprise applications line. Starlink earns the money. xAI carries the valuation. The bet asks public investors to price all three at once.
Starlink Is The Engine
Starlink ended 2025 with 9 million subscribers across 164 countries, then climbed to 10.3 million by the end of March 2026 — up from 5 million a year earlier. The satellite factory now turns out more than 4,000 units a year, around 340 a month, a manufacturing cadence that places every other satellite operator on Earth in a different industrial weight class. The Direct-to-Cell constellation already runs 650 satellites in orbit, active users crossed 10 million by early 2026, and the projection for end-of-year is 25 million.
The pressure point inside Starlink is the average revenue per user, which dropped 18 per cent over two years — from $99 a month in 2023 to $81 in 2025 — as the service rolled into lower-income geographies and bundled with promo tiers like the US Mobile entry plans at $47, $77 and $117 a month. The offset is enterprise. Maritime services head towards $1.94 billion in 2026 revenue. Commercial aviation installations are up 68 per cent year-on-year. Starshield, the defence-grade arm, runs on a $537 million Pentagon contract and projects $3.2 billion of 2026 revenue split between $761 million of software-as-a-service fees and $2.5 billion of constellation-as-a-service revenue. Quilty Space sees Starlink reaching $20 billion of revenue in 2026 with $14 billion of adjusted EBITDA.
Starlink sub-segment FY 2025 revenue FY 2026 projected Key driver
Falcon Pays, Starship Spends
Falcon 9 and Falcon Heavy ran roughly 165 orbital launches in 2025 — about 85 per cent of US orbital output and close to twice China's annual cadence. The Center for Security and Emerging Technology has flagged the resulting near-monopoly as a structural risk for US national security procurement, since federal launch contracts have one place to land at scale. The cash Falcon throws off, though, gets pulled straight back into Starship. The S-1 lists $3.8 billion of segment-specific capital spending in 2025 and more than $3 billion of R&D dedicated to Starship development, then another $930 million of Starship R&D in Q1 2026 alone. Starship has to hit full rapid reusability before the bigger V3 Starlink satellites can fly, and before any Mars cadence becomes real. Until then, the Space segment runs on Falcon and bleeds on Starship — the Ferrari that wins every constructor's championship while pouring the prize money into an experimental engine that has yet to leave the dyno.
The xAI Money Pit
The AI segment is where the math falls apart. Pro-forma 2025 numbers: $3.2 billion of revenue against a $6.4 billion operating loss, driven by $12.7 billion of capital spending on GPU procurement and data centre construction. Pre-merger, xAI burned $7.8 billion in the first three quarters of 2025 and carried over $17.5 billion of high-cost debt — which SpaceX has now refinanced through a $20 billion bridge facility secured against the merged balance sheet. Servers and networking gear sit at $23.39 billion on the balance sheet. Data centre infrastructure: $2.97 billion. Construction-in-progress: $14.05 billion. Almost all of it AI-related. Outstanding contractual commitments add up to $25.45 billion, 95 per cent of which fall in fiscal 2026 and 2027 — cloud capacity, power purchase agreements, GPU orders the company has already signed for. The launch and satellite businesses are now backing the credit on a stack of speculative AI hardware. The casino floor is paying for the chip stack.
The Anthropic Anomaly
The single most editorially interesting line in the S-1 sits inside the related-party and contract disclosures. In May 2026, xAI signed Cloud Services Agreements with Anthropic — the company that builds Claude, the leading frontier competitor to xAI's own Grok — under which Anthropic will pay $1.25 billion a month through May 2029 for 300 megawatts of compute at Colossus 1 in Memphis, Tennessee, backed by more than 220,000 Nvidia GPUs spanning H100, H200 and GB200 chips. The first two months run at a discounted rate while xAI completes the ramp. Total contract value: roughly $15 billion a year, $40–45 billion over the full term. A single contract larger than Starlink's entire 2024 revenue.
Read that paragraph again. xAI has built the largest coherent AI training cluster on Earth and rented its full output to a frontier-model competitor. Anthropic uses Colossus 1 to train Claude while xAI moves Grok 5 onto the newer Colossus 2 facility next door. Two of the four leading frontier-model companies in the world now share a campus and a power substation. The SpaceX language in the filing describes the deal as monetising "unused compute capacity in our infrastructure" and adds that the arrangement "provides multiple pathways to generate returns on invested capital." The subtext, which the filing leaves out, is that xAI overbuilt compute relative to its own model training needs and required a counterparty to soak up the excess before going public. Anthropic — supply-starved, racing to scale Claude — turned out to be the buyer that converted a stranded asset into the largest enterprise line in the entire prospectus.
The strategic read goes deeper. Compute supply is the single largest constraint on frontier-model labs in 2026. Securing three years of guaranteed capacity at scale, even at a premium, removes the bottleneck that has shaped every Anthropic deployment decision since GPT-4 era. For xAI, the deal converts excess silicon into pricing data — the deal will be quoted by every analyst pricing a frontier-compute contract for the next two years. The market just got its first published price for a fully-loaded 300-megawatt cluster: $1.25 billion a month. That single number sets the floor for every comparable lease going forward.
The 90-Day Kill Switch
The clause that matters in the Anthropic contract is one sentence long: either party can terminate the agreement on 90 days' notice. At a contract worth $15 billion a year, the termination provision is the emergency stop wired into the entire xAI revenue thesis. If Anthropic decides to migrate workloads onto its own dedicated infrastructure — and Anthropic's compute strategy until May 2026 was anchored on Amazon's Trainium silicon and Google's TPU footprints, not Nvidia chips — the contract evaporates inside a quarter. The xAI segment loses its single largest revenue line. The merged company loses the only credible AI-segment revenue datapoint that supports the trillion-dollar story. The full-fat valuation assumes the Anthropic line holds for three years. The 90-day clause is the trap door the casino keeps under the table.
Musk Owns The Casino
Elon Musk holds 12.3 per cent of Class A shares and 93.6 per cent of Class B shares. Class A gets one vote per share. Class B gets ten. The arithmetic produces 85.1 per cent voting power, which drops after IPO but stays above 50 per cent — the threshold that grants "controlled company" status on Nasdaq and exempts SpaceX from the rule requiring a majority-independent board. Class B holders alone elect 51 per cent of the directors. Class A and B vote jointly for the rest, which means Musk's combined voting power controls every board seat in existence. The charter also locks in his three titles — Chief Executive Officer, Chief Technical Officer, Chairman of the Board — for as long as he wants them. Removing him from any of those positions requires a majority Class B vote, which translates to Musk voting himself out.
On 13 May 2026, a week before the S-1 dropped, NYC Comptroller Mark Levine, NYS Comptroller Thomas DiNapoli and CalPERS CEO Marcie Frost sent a joint letter to SpaceX objecting to the structure. The company has stayed silent in response. Public Class A shareholders — however many billions of dollars they collectively put in — will buy economic exposure and near-zero governance voice. The Nasdaq rapid-inclusion rule means passive funds will buy regardless. Active managers who want SPCX exposure to track the market will buy regardless. The structure is built to convert public capital into private control with as little friction as the law allows. The dealer sets the rules at his own table.
The Mars Compensation Package
The most extraordinary line in the prospectus is inside the executive compensation disclosure. Musk's grant is structured as 15 tranches of 66,666,665 shares each — roughly 1 billion shares in total — vesting on two simultaneous conditions. The first is a market-cap milestone, starting at $500 billion and stepping up in $500 billion increments to $7.5 trillion. The second is that SpaceX put a permanent Mars colony of at least one million people on the planet, with Musk still running the firm when each milestone clears. If both conditions clear across the full structure, Musk picks up around 1.3 billion shares — a package worth, at the upper milestone, in the trillions.
The compensation tells you what kind of company SpaceX wants its shareholders to underwrite. The market-cap milestones reward financial growth. The Mars condition asks public-market investors to fund a level of engineering ambition unique in public-company history — a Saturn V launch attached to a stock vesting calendar. The package may sit unvested for the entire duration of the listing. The fact that it has been written into a public S-1 with audited financial reviewers signing off is itself a statement about what kind of equity instrument SPCX is. The bonus structure is a science-fiction novel in legal prose.
The Grok Liability
The S-1 discloses a litigation reserve in excess of $500 million, a chunk of it tied to xAI's Grok chatbot. Regulators in Ireland, Canada, Brazil, the United Kingdom and several US states are investigating allegations against Grok's "Spicy" mode, which has been associated with the generation of non-consensual explicit imagery and harmful content. The investigations span data protection, AI safety and content liability regimes — and they sit awkwardly beside SpaceX's existing relationships with NASA, the Department of Defense and the Five Eyes intelligence apparatus. A material adverse finding in any of those investigations would leave the Starlink and Falcon businesses untouched, but it would sit inside the merged company's risk disclosures for the entire duration of the public listing. Underwriters are watching.
What India Gets — And When
The India angle runs through three threads. The first is Starlink's commercial entry. IN-SPACe granted Starlink final regulatory approval on 8 July 2025 to operate its Gen 1 LEO constellation over Indian territory using Ka and Ku bands, with the licence valid until July 2030. The Department of Telecommunications cleared the company shortly after. What remains pending is spectrum allocation — TRAI proposed a five-year administrative allocation extendable by two years in May 2025, while Starlink lobbied for a 20-year term and Reliance Jio and Bharti Airtel backed three-to-five-year windows. Final security clearance and trial spectrum have yet to arrive. Until both arrive, Starlink stays in regulatory holding for full commercial service in India.
The second thread is distribution. In March 2025, Bharti Airtel signed a partnership to sell Starlink equipment and broadband through its retail and enterprise channels. Hours later, Reliance Jio announced a near-identical deal — Starlink hardware sold through Jio's nationwide retail footprint, backed by Jio's 515 million subscribers. The Mukesh Ambani-Sunil Mittal duopoly that fought Starlink's regulatory entry harder than any other lobby on Earth has now bound itself to selling Starlink terminals across the Indian market. Reliance and Bharti opened the innings at opposite ends, swung at every regulatory delivery for three years, and then walked off the field as Starlink's distribution partners. Reliance is now reported to be evaluating its own LEO constellation, which would put it in direct competition with the constellation it currently distributes hardware for. The captain who walks out to bat against his own opening partner.
The third thread is the absent Indian listed comparator. India has launched 18 PSLV missions in the last 18 months and built sovereign LEO ambitions through ISRO, the IN-SPACe regulator and the SpaceTech PLI scheme. All of that activity has produced sub-$1 billion private valuations. Pixxel and Skyroot trade only in private markets. Bharti's stake in Eutelsat OneWeb sits inside a London-listed parent. The SPCX listing will, for the first time, give Indian institutional investors a direct way to own space-economy exposure — and the structural absence of a domestic equivalent puts pressure on the Indian listings pipeline to produce one before the second wave of LEO operators IPOs. The Starshield architecture, walled off from non-US allies by ITAR rules, also re-opens the long-standing question of whether India's sovereign satellite communications stack ends up sourcing from a US-controlled constellation or a home-built alternative. The second-wave Indian LEO operators are still in the pavilion. The first ball of that innings is still to be bowled.
The Orbital Data Centre Moonshot
The most fantastical line in the entire S-1 is the disclosure that SpaceX plans to deploy solar-powered AI data centres in orbit beginning in 2028, using Starship to launch up to one million purpose-built compute satellites. The rationale, per the filing, is that orbital deployment routes around terrestrial power-grid limits, water-cooling bottlenecks and land acquisition. Early conversations with major hyperscalers are under way, with Google named in source documentation as one early counterparty. The filing's own risk disclosure flags the orbital data centre concept as engineering science-fiction: thermal dissipation in vacuum remains an open problem, radiation hardening of advanced GPUs is a separate open problem, and the Starship launch cadence required to deploy the constellation remains a pending demonstration.
What the orbital data centre line actually does inside the S-1 is convert speculative engineering into a valuation lever. If even 5 per cent of the projected one million satellites flies, the AI segment's TAM line gets re-rated. If the project stays grounded, the line costs the company nothing — it sits inside forward-looking statements protected by safe-harbour language. The asymmetry is what valuation engineering looks like in a frontier-tech S-1. Beam your cloud workloads up to orbit. The Federation will see you now.
What Has To Be True For $1.75 Trillion
The valuation math is a problem. At the lower $1.75 trillion target, SPCX prices at 93.7 times trailing sales. At the upper $2 trillion, the multiple stretches to 107.1 times. Compressed against forward 2026 revenue projections of $27–30 billion, the multiple compresses to 58.3–64.8 times forward sales. For context, Nvidia — the dominant beneficiary of the same AI capex cycle SpaceX is now buying into — trades at a fraction of that sales multiple even at the height of its 2025 rally. For SPCX to align with conventional tech multiples by 2030, the consolidated revenue line has to clear $150 billion. That requires Starlink to dominate B2B and defence connectivity across global markets, Starship to operate at a near-daily launch cadence, and xAI to convert hardware overhang into a sustainable compute cross-sell at margins the segment is still working towards.
The bull case from independent technology investors — including Jeff Brown of Brownstone Research, quoted in syndicated coverage — argues that SpaceX's vertically integrated launch, satellite manufacturing and AI infrastructure stack will compound into the most valuable corporation in the global economy over the next decade. The bear case, voiced from conservative institutional desks, prices in the $14 billion of cash the AI segment burned in 2025, the 90-day Anthropic exit clause, and the absence of public-shareholder governance rights. Both cases hold together. The IPO will determine which one capital believes.
The House Always Wins
Whether SpaceX can build that future is the easier question — the company has spent a decade making the impossible look routine. The harder question is whether the public market wants to underwrite a private-company governance structure, a single-customer AI revenue line, and a Mars colonisation milestone attached to the largest executive compensation package ever filed. Starlink — the championship-winning chassis — is being used to subsidise Starship and to underwrite the credit risk on $20 billion of xAI hardware. The Anthropic contract is what makes the underwriting arithmetic close in any given quarter. The 85.1 per cent voting structure is what gives Musk the freedom to keep running the bet regardless of what the public shareholders he is about to acquire would prefer. The Mars colony milestone inside the compensation package is what tells you, in plain language, what success looks like to the operator who controls every board seat.
At a casino this size, the house picks the rules, the croupier, the table, and the chair. The dealer is Elon Musk. Public shareholders are the players. The answer arrives on 12 June.
The Filing In One Page
Bank of America, Goldman Sachs, Morgan Stanley, JPMorgan Chase and Citigroup are lead underwriters. About 30 per cent of the offering goes to retail allocation through Schwab, Fidelity, Robinhood, SoFi and E*Trade — an unusual tilt for a deal this large. A 5-for-1 stock split lands ahead of listing to pull the per-share price down to roughly $105 from the pre-split fair-market value of $526.59. Nasdaq's revised inclusion rules will let SPCX enter the Nasdaq 100 inside 15 days of listing, which means passive index funds will buy the stock irrespective of what the governance disclosures say. Polymarket traders are pricing a 51 per cent chance the first trading day closes between $1.5 trillion and $2 trillion. The retail allocation is the headline; the index buying is the engine.
The Numbers That Set The Stage
The S-1 shows a company that swung from a $791 million profit in 2024 to a $4.94 billion loss in 2025 — the same year revenue rose 33 per cent to $18.67 billion. The accumulated deficit on the balance sheet sits at $41.3 billion. Capital spending nearly doubled in twelve months, from $11.2 billion to $20.7 billion. The Q1 2026 net loss of $4.28 billion accounts for 86.6 per cent of the entire full-year 2025 loss compressed into 90 days. The Q1 2025 comparison loss was $528 million. Read the trajectory before reading anything else.
Period Revenue Net result Notes
| FY 2024 | ~$13.1 billion (pre-merger) | $791 million profit | Standalone aerospace business |
| FY 2025 | $18.67 billion | $4.94 billion loss | First full year of xAI drag |
| Q1 2025 | comparable base | $528 million loss | Pre-merger run-rate |
| Q1 2026 | $4.69 billion | $4.28 billion loss | Full xAI capex now visible |
Three Businesses, One Ticker
The strangest thing about the filing is that it sells three structurally different businesses at one price. Three games run at the same table. Connectivity, anchored by Starlink, brought in $11.4 billion in 2025 — 61 per cent of all revenue, up 48 per cent from $7.7 billion in 2024, with $7.2 billion of adjusted EBITDA at a 63 per cent margin that beats every land-based telco on Earth. Space, anchored by Falcon 9 and the still-developing Starship, did $4.1 billion in revenue and an operating loss of $657 million as Starship development eats the launch business's profits. The Artificial Intelligence segment, absorbed in February 2026 when SpaceX paid $250 billion in stock for xAI, contributed $3.2 billion of pro-forma revenue against a $6.4 billion operating loss — close to three dollars lost for every dollar earned.
Sold apart, the three businesses would carry three different price-to-sales multiples and three different investor bases. Sold together, the merged company gets the AI-infrastructure multiple while running on the satellite-broadband cash flow. The S-1 floats a $28.5 trillion total addressable market — language the company calls "the largest actionable TAM in human history" — and roughly 93 per cent of that number sits inside the AI segment's enterprise applications line. Starlink earns the money. xAI carries the valuation. The bet asks public investors to price all three at once.
Starlink Is The Engine
Starlink ended 2025 with 9 million subscribers across 164 countries, then climbed to 10.3 million by the end of March 2026 — up from 5 million a year earlier. The satellite factory now turns out more than 4,000 units a year, around 340 a month, a manufacturing cadence that places every other satellite operator on Earth in a different industrial weight class. The Direct-to-Cell constellation already runs 650 satellites in orbit, active users crossed 10 million by early 2026, and the projection for end-of-year is 25 million.
The pressure point inside Starlink is the average revenue per user, which dropped 18 per cent over two years — from $99 a month in 2023 to $81 in 2025 — as the service rolled into lower-income geographies and bundled with promo tiers like the US Mobile entry plans at $47, $77 and $117 a month. The offset is enterprise. Maritime services head towards $1.94 billion in 2026 revenue. Commercial aviation installations are up 68 per cent year-on-year. Starshield, the defence-grade arm, runs on a $537 million Pentagon contract and projects $3.2 billion of 2026 revenue split between $761 million of software-as-a-service fees and $2.5 billion of constellation-as-a-service revenue. Quilty Space sees Starlink reaching $20 billion of revenue in 2026 with $14 billion of adjusted EBITDA.
Starlink sub-segment FY 2025 revenue FY 2026 projected Key driver
| Consumer broadband | $8.14 billion | $11.30 billion | 16.8M residential subscribers target |
| Enterprise | $1.38 billion | $1.68 billion | Regional B2B and land mobility |
| Maritime | $1.25 billion | $1.94 billion | ~75,000 net vessel installs |
| Commercial aviation | $640 million | $1.08 billion | Peak install year for business jets and fleets |
| Starshield (defence) | $2.10 billion | $3.20 billion | $761M SaaS + $2.5B constellation-as-service |
Falcon 9 and Falcon Heavy ran roughly 165 orbital launches in 2025 — about 85 per cent of US orbital output and close to twice China's annual cadence. The Center for Security and Emerging Technology has flagged the resulting near-monopoly as a structural risk for US national security procurement, since federal launch contracts have one place to land at scale. The cash Falcon throws off, though, gets pulled straight back into Starship. The S-1 lists $3.8 billion of segment-specific capital spending in 2025 and more than $3 billion of R&D dedicated to Starship development, then another $930 million of Starship R&D in Q1 2026 alone. Starship has to hit full rapid reusability before the bigger V3 Starlink satellites can fly, and before any Mars cadence becomes real. Until then, the Space segment runs on Falcon and bleeds on Starship — the Ferrari that wins every constructor's championship while pouring the prize money into an experimental engine that has yet to leave the dyno.
The xAI Money Pit
The AI segment is where the math falls apart. Pro-forma 2025 numbers: $3.2 billion of revenue against a $6.4 billion operating loss, driven by $12.7 billion of capital spending on GPU procurement and data centre construction. Pre-merger, xAI burned $7.8 billion in the first three quarters of 2025 and carried over $17.5 billion of high-cost debt — which SpaceX has now refinanced through a $20 billion bridge facility secured against the merged balance sheet. Servers and networking gear sit at $23.39 billion on the balance sheet. Data centre infrastructure: $2.97 billion. Construction-in-progress: $14.05 billion. Almost all of it AI-related. Outstanding contractual commitments add up to $25.45 billion, 95 per cent of which fall in fiscal 2026 and 2027 — cloud capacity, power purchase agreements, GPU orders the company has already signed for. The launch and satellite businesses are now backing the credit on a stack of speculative AI hardware. The casino floor is paying for the chip stack.
The Anthropic Anomaly
The single most editorially interesting line in the S-1 sits inside the related-party and contract disclosures. In May 2026, xAI signed Cloud Services Agreements with Anthropic — the company that builds Claude, the leading frontier competitor to xAI's own Grok — under which Anthropic will pay $1.25 billion a month through May 2029 for 300 megawatts of compute at Colossus 1 in Memphis, Tennessee, backed by more than 220,000 Nvidia GPUs spanning H100, H200 and GB200 chips. The first two months run at a discounted rate while xAI completes the ramp. Total contract value: roughly $15 billion a year, $40–45 billion over the full term. A single contract larger than Starlink's entire 2024 revenue.
Read that paragraph again. xAI has built the largest coherent AI training cluster on Earth and rented its full output to a frontier-model competitor. Anthropic uses Colossus 1 to train Claude while xAI moves Grok 5 onto the newer Colossus 2 facility next door. Two of the four leading frontier-model companies in the world now share a campus and a power substation. The SpaceX language in the filing describes the deal as monetising "unused compute capacity in our infrastructure" and adds that the arrangement "provides multiple pathways to generate returns on invested capital." The subtext, which the filing leaves out, is that xAI overbuilt compute relative to its own model training needs and required a counterparty to soak up the excess before going public. Anthropic — supply-starved, racing to scale Claude — turned out to be the buyer that converted a stranded asset into the largest enterprise line in the entire prospectus.
The strategic read goes deeper. Compute supply is the single largest constraint on frontier-model labs in 2026. Securing three years of guaranteed capacity at scale, even at a premium, removes the bottleneck that has shaped every Anthropic deployment decision since GPT-4 era. For xAI, the deal converts excess silicon into pricing data — the deal will be quoted by every analyst pricing a frontier-compute contract for the next two years. The market just got its first published price for a fully-loaded 300-megawatt cluster: $1.25 billion a month. That single number sets the floor for every comparable lease going forward.
The 90-Day Kill Switch
The clause that matters in the Anthropic contract is one sentence long: either party can terminate the agreement on 90 days' notice. At a contract worth $15 billion a year, the termination provision is the emergency stop wired into the entire xAI revenue thesis. If Anthropic decides to migrate workloads onto its own dedicated infrastructure — and Anthropic's compute strategy until May 2026 was anchored on Amazon's Trainium silicon and Google's TPU footprints, not Nvidia chips — the contract evaporates inside a quarter. The xAI segment loses its single largest revenue line. The merged company loses the only credible AI-segment revenue datapoint that supports the trillion-dollar story. The full-fat valuation assumes the Anthropic line holds for three years. The 90-day clause is the trap door the casino keeps under the table.
Musk Owns The Casino
Elon Musk holds 12.3 per cent of Class A shares and 93.6 per cent of Class B shares. Class A gets one vote per share. Class B gets ten. The arithmetic produces 85.1 per cent voting power, which drops after IPO but stays above 50 per cent — the threshold that grants "controlled company" status on Nasdaq and exempts SpaceX from the rule requiring a majority-independent board. Class B holders alone elect 51 per cent of the directors. Class A and B vote jointly for the rest, which means Musk's combined voting power controls every board seat in existence. The charter also locks in his three titles — Chief Executive Officer, Chief Technical Officer, Chairman of the Board — for as long as he wants them. Removing him from any of those positions requires a majority Class B vote, which translates to Musk voting himself out.
On 13 May 2026, a week before the S-1 dropped, NYC Comptroller Mark Levine, NYS Comptroller Thomas DiNapoli and CalPERS CEO Marcie Frost sent a joint letter to SpaceX objecting to the structure. The company has stayed silent in response. Public Class A shareholders — however many billions of dollars they collectively put in — will buy economic exposure and near-zero governance voice. The Nasdaq rapid-inclusion rule means passive funds will buy regardless. Active managers who want SPCX exposure to track the market will buy regardless. The structure is built to convert public capital into private control with as little friction as the law allows. The dealer sets the rules at his own table.
The Mars Compensation Package
The most extraordinary line in the prospectus is inside the executive compensation disclosure. Musk's grant is structured as 15 tranches of 66,666,665 shares each — roughly 1 billion shares in total — vesting on two simultaneous conditions. The first is a market-cap milestone, starting at $500 billion and stepping up in $500 billion increments to $7.5 trillion. The second is that SpaceX put a permanent Mars colony of at least one million people on the planet, with Musk still running the firm when each milestone clears. If both conditions clear across the full structure, Musk picks up around 1.3 billion shares — a package worth, at the upper milestone, in the trillions.
The compensation tells you what kind of company SpaceX wants its shareholders to underwrite. The market-cap milestones reward financial growth. The Mars condition asks public-market investors to fund a level of engineering ambition unique in public-company history — a Saturn V launch attached to a stock vesting calendar. The package may sit unvested for the entire duration of the listing. The fact that it has been written into a public S-1 with audited financial reviewers signing off is itself a statement about what kind of equity instrument SPCX is. The bonus structure is a science-fiction novel in legal prose.
The Grok Liability
The S-1 discloses a litigation reserve in excess of $500 million, a chunk of it tied to xAI's Grok chatbot. Regulators in Ireland, Canada, Brazil, the United Kingdom and several US states are investigating allegations against Grok's "Spicy" mode, which has been associated with the generation of non-consensual explicit imagery and harmful content. The investigations span data protection, AI safety and content liability regimes — and they sit awkwardly beside SpaceX's existing relationships with NASA, the Department of Defense and the Five Eyes intelligence apparatus. A material adverse finding in any of those investigations would leave the Starlink and Falcon businesses untouched, but it would sit inside the merged company's risk disclosures for the entire duration of the public listing. Underwriters are watching.
What India Gets — And When
The India angle runs through three threads. The first is Starlink's commercial entry. IN-SPACe granted Starlink final regulatory approval on 8 July 2025 to operate its Gen 1 LEO constellation over Indian territory using Ka and Ku bands, with the licence valid until July 2030. The Department of Telecommunications cleared the company shortly after. What remains pending is spectrum allocation — TRAI proposed a five-year administrative allocation extendable by two years in May 2025, while Starlink lobbied for a 20-year term and Reliance Jio and Bharti Airtel backed three-to-five-year windows. Final security clearance and trial spectrum have yet to arrive. Until both arrive, Starlink stays in regulatory holding for full commercial service in India.
The second thread is distribution. In March 2025, Bharti Airtel signed a partnership to sell Starlink equipment and broadband through its retail and enterprise channels. Hours later, Reliance Jio announced a near-identical deal — Starlink hardware sold through Jio's nationwide retail footprint, backed by Jio's 515 million subscribers. The Mukesh Ambani-Sunil Mittal duopoly that fought Starlink's regulatory entry harder than any other lobby on Earth has now bound itself to selling Starlink terminals across the Indian market. Reliance and Bharti opened the innings at opposite ends, swung at every regulatory delivery for three years, and then walked off the field as Starlink's distribution partners. Reliance is now reported to be evaluating its own LEO constellation, which would put it in direct competition with the constellation it currently distributes hardware for. The captain who walks out to bat against his own opening partner.
The third thread is the absent Indian listed comparator. India has launched 18 PSLV missions in the last 18 months and built sovereign LEO ambitions through ISRO, the IN-SPACe regulator and the SpaceTech PLI scheme. All of that activity has produced sub-$1 billion private valuations. Pixxel and Skyroot trade only in private markets. Bharti's stake in Eutelsat OneWeb sits inside a London-listed parent. The SPCX listing will, for the first time, give Indian institutional investors a direct way to own space-economy exposure — and the structural absence of a domestic equivalent puts pressure on the Indian listings pipeline to produce one before the second wave of LEO operators IPOs. The Starshield architecture, walled off from non-US allies by ITAR rules, also re-opens the long-standing question of whether India's sovereign satellite communications stack ends up sourcing from a US-controlled constellation or a home-built alternative. The second-wave Indian LEO operators are still in the pavilion. The first ball of that innings is still to be bowled.
The Orbital Data Centre Moonshot
The most fantastical line in the entire S-1 is the disclosure that SpaceX plans to deploy solar-powered AI data centres in orbit beginning in 2028, using Starship to launch up to one million purpose-built compute satellites. The rationale, per the filing, is that orbital deployment routes around terrestrial power-grid limits, water-cooling bottlenecks and land acquisition. Early conversations with major hyperscalers are under way, with Google named in source documentation as one early counterparty. The filing's own risk disclosure flags the orbital data centre concept as engineering science-fiction: thermal dissipation in vacuum remains an open problem, radiation hardening of advanced GPUs is a separate open problem, and the Starship launch cadence required to deploy the constellation remains a pending demonstration.
What the orbital data centre line actually does inside the S-1 is convert speculative engineering into a valuation lever. If even 5 per cent of the projected one million satellites flies, the AI segment's TAM line gets re-rated. If the project stays grounded, the line costs the company nothing — it sits inside forward-looking statements protected by safe-harbour language. The asymmetry is what valuation engineering looks like in a frontier-tech S-1. Beam your cloud workloads up to orbit. The Federation will see you now.
What Has To Be True For $1.75 Trillion
The valuation math is a problem. At the lower $1.75 trillion target, SPCX prices at 93.7 times trailing sales. At the upper $2 trillion, the multiple stretches to 107.1 times. Compressed against forward 2026 revenue projections of $27–30 billion, the multiple compresses to 58.3–64.8 times forward sales. For context, Nvidia — the dominant beneficiary of the same AI capex cycle SpaceX is now buying into — trades at a fraction of that sales multiple even at the height of its 2025 rally. For SPCX to align with conventional tech multiples by 2030, the consolidated revenue line has to clear $150 billion. That requires Starlink to dominate B2B and defence connectivity across global markets, Starship to operate at a near-daily launch cadence, and xAI to convert hardware overhang into a sustainable compute cross-sell at margins the segment is still working towards.
The bull case from independent technology investors — including Jeff Brown of Brownstone Research, quoted in syndicated coverage — argues that SpaceX's vertically integrated launch, satellite manufacturing and AI infrastructure stack will compound into the most valuable corporation in the global economy over the next decade. The bear case, voiced from conservative institutional desks, prices in the $14 billion of cash the AI segment burned in 2025, the 90-day Anthropic exit clause, and the absence of public-shareholder governance rights. Both cases hold together. The IPO will determine which one capital believes.
The House Always Wins
Whether SpaceX can build that future is the easier question — the company has spent a decade making the impossible look routine. The harder question is whether the public market wants to underwrite a private-company governance structure, a single-customer AI revenue line, and a Mars colonisation milestone attached to the largest executive compensation package ever filed. Starlink — the championship-winning chassis — is being used to subsidise Starship and to underwrite the credit risk on $20 billion of xAI hardware. The Anthropic contract is what makes the underwriting arithmetic close in any given quarter. The 85.1 per cent voting structure is what gives Musk the freedom to keep running the bet regardless of what the public shareholders he is about to acquire would prefer. The Mars colony milestone inside the compensation package is what tells you, in plain language, what success looks like to the operator who controls every board seat.
At a casino this size, the house picks the rules, the croupier, the table, and the chair. The dealer is Elon Musk. Public shareholders are the players. The answer arrives on 12 June.
Next Story