Vol. 1 · No.3 · 16 June 2026 THE AI INDUSTRY'S FORTNIGHTLY readcompute.com
Vol. 1
№ 3
June
£3.99
Compute!
Issue №3 · June 2026 Footnotes throughout. Disclaimers nowhere. UK £3.99 · US $5.99
The Lede

SpaceX: $135 a share. $1.77 trillion.Fixed price — no range. Musk keeps 82%+ of the vote. The biggest IPO ever, three times Alibaba. Filed: SPCX.

→ p. 04
The Money

Anthropic: $965bn post → $1tn float.S-1 confidential, 1 June. Bankers undisclosed. The quiet filer files quietly.

→ p. 05
Issue №3 typographic cover — S-1, S-1, S-1. The IPO Special.
FREE!
Pull-out
Poster
Inside!
THE TRILLION-DOLLAR S-1 STAMPEDE · 1 MAY – 8 JUNE 2026

S-1, S-1,
S-1.

Three frontier laboratories filed S-1s with the SEC in forty days. Combined private-market valuation, at print: $2.84 trillion — roughly two-thirds of the GDP of India. The bankers, in all three cases: mostly the same five firms. Yahoo!'s IPO, in 1996, was also led by Morgan Stanley.
The Joke

OpenAI, on timing: "We have not decided."Mr Altman conjugated the verb "decide" six times on CNBC. None in the future tense.

→ p. 05
The Service

The Web. Vol. 3 — in dollars.Twelve cap tables. Seven cross-investors. The same five bankers. Pull out, pin up, follow the cheque.

→ pp. 14–15
On the record. On the cap table. On the nose.
FRONTIER DISPATCH · MIDTOWN MANHATTAN ·

The Same Five Firms.

The SpaceX roadshow opened on Wednesday. The lead deck contained eleven references to Starlink, four to Starship, and none to dual-class voting. The price on the screen was $135 a share — fixed, no range — implying $1.77 trillion. Goldman Sachs leads the ticket. Morgan Stanley, which led Yahoo!'s IPO in 1996, is the second name on it.1

The SpaceX roadshow opened in midtown on Wednesday, . The lead presentation lasted fifty-seven minutes. The CFO, Bret Johnsen, took questions. The questions, per a person in the room, were "polite." The price on the screen was $135 a share — fixed, no range; demand sensitivity is, evidently, for other people. At $135, the valuation is $1.77 trillion, and the implied price-to-sales multiple, on $18.67bn of 2025 revenue, is 95×.2

This is what the IPO class of June '26 looks like from the bankers' side: three deals, five firms, no daylight. Goldman Sachs leads the SpaceX ticket; Morgan Stanley, Bank of America, Citigroup and JPMorgan Chase complete it, per the amended filing of 3 June. Anthropic and OpenAI have not named their syndicates. We are told, off the record, by a person in proximity to a person in proximity to a deal, that "the firms are talking." We have been told this every fortnight since November.3

When the same five firms underwrite every deal at every valuation, the pricing discipline that distinguishes a $400bn float from a $1.4tn float is structurally absent. This is not a moral claim. It is an arithmetic one. Five firms splitting the underwriting on $2.84 trillion of paper, even at an all-in 1%, is $28 billion in fees. The marginal value of a no, on this math, is negative.4 Continued on page 5.

"We are scarce in chips." — Sarah Friar, CFO of OpenAI · The Information · 47th recorded use of the phrase since 20249
  1. The slide-count is per a person in the room at the SpaceX roadshow opening, 4 June. The deck has not been publicly released; Compute! has not seen it. The reported count of dual-class references — zero — is, by definition, easier to verify.
  2. Morgan Stanley led Yahoo!'s 1996 IPO and is, per the amended filing of 3 June, the second name on SpaceX's ticket, behind Goldman Sachs. The intervening Yahoo! P/S compression — from a peak above 80× in early 2000 to roughly 1× at acquisition by Verizon in 2017 — is not, on a quick read of the prospectus, referenced in SpaceX's filing.
  3. The source asked not to be identified. The source has been not-identified by us, fortnightly, since November.
  4. Underwriting fees on a $75bn raise at 1.5–3% imply $1.1bn–$2.25bn for the SpaceX deal alone. The all-in fee pool across the three filings, on conservative assumptions, exceeds $5 billion. This is, on the bankers' published league tables, a footnote.

Pseuds Corner

A standing column · Issue 3
"We have not decided." — Sam Altman, on the timing of OpenAI's IPO · CNBC ·
"I am not focused on the timing." — Sam Altman, on the same subject, twenty seconds later · CNBC ·
"We are scarce in chips." — Sarah Friar, CFO of OpenAI · The Information · the 47th recorded use of the phrase since 2024
"Up to 5 gigawatts of compute capacity through 2030." — Anthropic–AWS joint statement · · roughly the peak generation of the Hoover Dam
"As soon as 2027." — Ali Ghodsi, CEO of Databricks, on the timing of an IPO · public remarks · the fourth recorded "as soon as next year" since 2022

The Frontier Circular

A bulletin of trivial movements · fortnight ending 8 June
  1. Mr. Altman appeared on CNBC and informed the host that OpenAI had filed an S-1. The host, who had read the press release forty minutes earlier, professed surprise.
  2. Mr. Amodei did not appear on CNBC. The Anthropic communications team was, as ever, busy.
  3. Mr. Musk attended the SpaceX roadshow opening in person. He brought, per a person who was in the room, a Diet Coke and a small folder. The folder was, by all accounts, the prospectus. He did not open it.
  4. Ms. Friar declined to specify the inference gross-margin trajectory for calendar 2027, citing "scarce chips." She has cited "scarce chips" forty-seven times since 2024. The chips remain scarce.
  5. Mr. Ghodsi told a private audience he would IPO "as soon as 2027." He has been telling private audiences he would IPO "as soon as next year" since 2022. We are presently at four next years.
  6. Mr. Wang, of Meta Superintelligence Labs, formerly of Scale AI, was at filing both. His X biography reads "Building." It does not specify what, or where.
  7. The corkboard has, for the first time in three issues, been re-felted in green, used to indicate valuations above $500bn. Red string remains for cap-table relationships; aqua for cloud commitments; yellow for NVIDIA, which is in everything.
Compute! · No.3 · p. 04

Of the three filings, only SpaceX's has a price — and as of Wednesday's amended filing, an unusually exact one: $135 a share, 555.6 million shares, a $75 billion raise at $1.77 trillion, on $18.67bn of 2025 revenue and a $4.94bn net loss. Mr Musk — henceforth, in these pages, the trillion-dollar tinkerer — retains more than 82% of the voting control after the offering.5 The alternative ticker MUSK was, per a person familiar, "discussed" and "rejected"; the filed ticker is SPCX. The reasons for rejection were not, at filing, supplied.

The amended filing also notes, in passing, that the xAI unit bought $269 million of Tesla megapacks in April, and that Tesla holds 18.99 million SpaceX shares — worth $2.56 billion at the fixed price. "We have historically collaborated with Tesla through commercial, licensing, and support agreements," the filing says. Historically.

Anthropic filed two weeks earlier with the tonal restraint of a librarian filing a tax return. The company — henceforth the quiet filer — raised $65 billion in May at a post-money valuation of $965 billion. It has not named its bankers. It has not named its timing. It has, however, committed to $100 billion of AWS spending over ten years — a figure that exceeds IBM's total cloud revenue over the same period.6 The compute partner is not, in this arrangement, an arm's-length supplier. It is, structurally, a co-issuer of the equity, without sitting on the cap table.

OpenAI's filing, announced openly on Monday, contained no timing. "We have not decided," said Sam Altman — henceforth, in these pages, the present-perfect filer — on CNBC, at 09:14 ET. The transcript has been reviewed. The verb "decide" appears six times. On no occasion is it conjugated in the future tense. On three occasions it is conjugated in the present perfect, the tense one uses to describe something that has not yet been done but may, at some indeterminate point, be done.7

The stakes, for the reader. At $1.77tn, SPCX debuts on 12 June as the seventh-largest company in the United States — above Tesla, at roughly $1.6tn, which is, awkwardly, also Mr Musk's. That is, by mechanical index inclusion, a position in every American 401(k) and every Canadian RRSP that holds the index. Anthropic and OpenAI, if they follow at their rumoured marks, will follow by the same mechanism. The public — meaning the reader's pension — is not being asked, in any of these cases, whether it wants to be a forced buyer of equity at 95× sales in a company whose founder retains more than 82% of the vote. It will be a forced buyer regardless.8

SpaceX, at $135 a share: 95× sales.
Yahoo!, at its 2000 peak: ~80×. — Fig. 2 · price-to-sales on stated revenue · Yahoo! is no longer a publicly listed equity
  1. Per the amended filing of 3 June: 555.6 million shares at a fixed $135 (a $75bn raise), an underwriters' option for 83.33 million more ($11.2bn), "over 82%" voting control retained by Mr Musk, and up to 5% of the stock reserved for a direct share program for "certain employees and persons." The persons are not named.
  2. IBM's segmented cloud revenue (Red Hat + IBM Cloud) ran approximately $25bn annually in 2024–25. The ten-year extrapolation is Compute!'s; the comparison is mechanical.
  3. The CNBC transcript has been reviewed by the Compute! desk. The verb "decide" appears six times. The future tense appears in approximately zero of them.
  4. S&P 500 index inclusion rules require a U.S. domicile, positive earnings over the most recent four quarters, and a public float of at least 50%. SPCX, per the prospectus, meets one of three. The rules have, on past evidence, proved flexible.
  5. A text-search of Ms Friar's public remarks since Q1 2024 returns 47 occurrences of "scarce chips" or near-variants. The chips have, by any independent measure, been less scarce in 2026 than in 2024.
Compute! · No.3 · p. 05

The Class of '26, Ranked

By stated valuation · tiebreak by S-1 status · n=8
Eight AI-adjacent companies ranked by stated valuation as of June 2026, with S-1 status and tiebreak notes.
#SubjectValuationTiebreak
1SpaceX (+xAI)the trillion-dollar tinkerer · S-1 public, 20 May '26$1.77tn at $135, fixed$4.94bn net loss; founder keeps 82%+ of the vote
2Anthropicthe quiet filer · S-1 confidential, 1 Jun '26$965bn post; ~$1tn targetrevenue undisclosed
3OpenAIthe present-perfect filer · S-1 confidential, 8 Jun '26"north of $100bn"inference gross margin ≈ 33%
4DatabricksCap'n Q3-of-Whenever · still private; talking$165–175bn (Jun '26 talk)four "next years" since 2022
5Andurilthe deferred patriot · still private; Series H Feb '26$61bn postcustomers all wear the same uniform
6Scale AIthe partial exit · 49% sold to Meta, Jun '25~$29bn impliedfounder now wears two name tags
7Groqthe inference upstart · still private$6bn post2025 revenue forecast cut from $2bn+ to $500m
8Hugging Facethe dormant unicorn · last raise Aug '23$4.5bnnothing has happened in thirty-four months
  1. Already Out: Cerebras · the wafer-scale punter · Nasdaq: CBRS · IPO reportedly priced at $56.4bn, raising $5.55bn — per Forge Global; not yet confirmed by Reuters at filing. The corkboard has been holding a yellow string for the position for three weeks. We should, eventually, like to take it down.

Separated at Birth

A standing column · cited numerically · Issue 3 · № SAB.3
portrait
pending
Ms. Friar, on chip allocation
portrait
pending
Mr. Johnsen, on chip allocation

Has anyone noticed that the two chief financial officers of the two most valuable private companies in the United States — both addressing investors inside the same calendar fortnight, both pre-IPO, both naming the same vendor as the binding constraint on their respective business plans — have, between them, told investors the same thing on six separate recorded occasions? Compute! raises the matter only because the constraint is, mechanically, the same constraint: NVIDIA's quarterly chip allocation. Both companies are, in this sense, in the same queue. Mr Huang is not, on present evidence, in any rush.10

  1. NVIDIA's H100 / B200 allocation has been the binding constraint for both companies' 2026 capacity plans. Mr Huang's gross margin remains, at filing, above 70%.

What No One Is Pricing In

A sober column · new this issue · three items

1 · The compute margin is moving the wrong way. OpenAI's inference gross margin reportedly halved in 2025, from roughly two-thirds to roughly one-third, as model size outgrew chip-cost amortisation. Anthropic and the xAI division of SpaceX face the same physics. If a $1bn quarterly compute bill grows 50% in 2027 while revenue grows 30%, the operating-leverage story embedded in the multiples does not work. The S-1s do not project compute trajectories beyond two years. The bankers' models, we understand, do. [Unverified at IPO level — visible at unit-economics level.]

2 · The cloud invoice is the cap table. Anthropic has committed $100bn of AWS spend over ten years. OpenAI's Microsoft contract, per filings during last year's Musk litigation, contains minimum-commit clauses. SpaceX, post-xAI merger, has folded NVIDIA exposure onto the SpaceX balance sheet. The compute partners are not arm's-length suppliers. They are co-issuers of the equity, without sitting on the cap table. If a single partner renegotiates, the listed entity's economics shift faster than any S-1 risk factor admits. [Unverified — visible in public commitments.]

3 · The dual-class control is non-diversifiable. SpaceX's amended S-1 leaves Mr Musk more than 82% of the vote. If Mr Musk is hit by a bus, departs to Mars, or simply tires of the company, the residual reprices. Public shareholders have no governance lever — and, by mechanical index inclusion at $1.77tn, no choice about owning the equity. The risk is, in the most literal sense, forced. [Unverified — structurally obvious.]

Compute! · No.3 · p. 12
The Web
Vol. 3 · The Cap-Table edition. Every cheque below is on the record. Pull out, pin up, follow the dollars.
Every cheque on this poster is on the record. Sourced from SEC filings, Reuters, Bloomberg, The Information, CNBC, and the subjects' own public statements · Jun 2025 – Jun 2026.
the three S-1s cap-table history cloud commitments NVIDIA, in everything
Compute! · No.3 · pp. 14–15
Your AI IPO
Horoscope.
Compute! · Issue №3 · The Stars, Underwritten
Six filers. Variable lengths. Earth signs file confidentially. Fire signs roadshow. SpaceX, having merged with xAI in February, is now both — which is, astrologically, a problem.

SpaceX (+xAI).

Fire AND Earth sign · The Trillion-Dollar Tinkerer's Vehicle
Roadshow 4 June. Debut 12 June, on the Nasdaq, at $135 a share — a fixed price, no range. $1.77tn on $18.67bn of 2025 revenue and a $4.94bn net loss. The amended S-1 leaves the founder more than 82% of the vote. Seventh-largest company in America on day one — above Tesla, which is, awkwardly, also his. Awarded, by acclamation, the Elon Premium trophy (№3.03) — same trophy as last issue, and the issue before.
Hired: every banker on Wall StreetLost: the distinction between "rocket company" and "AI company"Avoid: governance roadshows

OpenAI.

Fire sign · The Revolving Door, With A Form
Filed confidentially 8 June. Told CNBC, on the timing: "we have not decided." Holds the inaugural Schrödinger's S-1 trophy (№3.01) for the most committed-uncommitted filing of the fortnight.
Hired: the bankers, presumablyLost: ~¼ of inference gross margin, 2025Avoid: questions about timing

Anthropic.

Earth sign · The Quiet Filer
Filed an S-1 confidentially on 1 June. Has committed $100bn to AWS over ten years, which is, in the spirit of the season, more than IBM books in cloud over the same period.

Databricks.

Earth sign · Cap'n Q3-of-Whenever
Reported on 9 June to be in talks for a new private round at $165–175bn, up from $134bn four months ago. Run-rate revenue ~$5.4bn (+65% YoY). Multi-cloud. Profitable in theory. The CEO has stated that an IPO is coming "as soon as 2027." Has stated the same of every year since 2022.
Hired: enterprise customersLost: nothingAvoid: committing

Anduril.

Fire sign · The Deferred Patriot
Raised $5bn at $61bn post in February, four months after $2.5bn at $30.5bn post. Revenue ~$2.2bn. Customers all wear the same uniform. Will, Compute! understands, file — though not, on present evidence, before the mid-terms. (№3.05.)

Cerebras.

Earth sign · The Wafer-Scale Punter
Per Forge: already done. Per Reuters: unconfirmed. Per Compute!: both.
Compute! Coming Next Fortnight No. 4 · 30 June 2026
NEXT ISSUE · COMING ATTRACTIONS

12 June,
09:30.

By the time №4 lands, SPCX will have rung the opening bell or rescheduled it. Compute! will be in the gallery, with a stopwatch, a calculator, and a polite question about the dual-class. Plus: which of the three S-1s priced first, who blinked, and a Web pull-out, Vol. 4, the morning-after edition.
01
Verdict Watch

The Open
of SPCX.

The opening trade, annotated. The first downward revision, predicted. A sober tally of how much the founder retained, how much the public bought, and how much the bankers earned for showing up.

02
Long Read

The Bankers'
Yearbook.

The five firms underwriting every frontier offering. Their fees. Their conflicts. Their alumni — and the analysts who, three years ago, set price targets we have not heard about since.

03
Pull-out Poster · pp. 14–15

The Web,
volume four.

Same corkboard, different string. Every block trade, every lock-up, every secondary cross — with dates, with destinations, with receipts.

Every quote is real. Every footnote checked. Compute! · Vol.1 · No.3 · £3.99 / $5.99

Accessibility

Compute! targets WCAG 2.2 Level AA conformance. The corkboard pull-out poster includes a written long-description of every relationship line for screen-reader users; every image carries descriptive alt text; every quote names its speaker, date and source.

Full statement →

Last reviewed: . Conformance target: WCAG 2.2 AA.