Anthropic's SpaceX Deal and $10B Opus 5.5 Claims, Checked

§ 01What the post says#

The post opens with one line: “Interesting remarks Anthropic made in a meeting with the sell side”.

Sell side means banks and brokers, and here their research analysts. Four remarks follow.

• On SpaceX. “The contract with SpaceX will not be extended beyond the middle of next year”.
• On cost. “Developing Opus 5.5 cost more than $10 billion”.
• On the market. “The AI market is winner take all”.
• On open source. “Anthropic has the stability and the better talent that open source lacks”.

The post does not say which meeting, which bank, who spoke for Anthropic, or how the author heard it.

§ 02The author corrected one claim#

About 40 minutes later the author replied to his own post. The correction reads: “The $10 billion was reportedly spent on training the entire Claude 5.5 model family, including Opus 5.5”.

• What changed. The cost moved from one model to three. In another reply he lists them as Opus 5.5, Sonnet 5.5 and Haiku 5.5.
• What the word reportedly adds. The author is passing on what he was told. He does not claim to have been in the meeting.
• Size. The correction had about 57,000 views at 17:15 UTC, against about 516,000 for the original. The correction has been viewed far less than the claim it corrects.

§ 03The four claims against the filing#

§ 04The SpaceX claim#

• What the filing describes. Agreements with SpaceX’s AI unit worth up to $120.5 billion, “over terms extending into 2029”.
• The exit right. After the first periods end, the filing says “each agreement may be cancelled by either party upon 90 days’ advance written notice”.
• So this would be a cancellation, not an expiry. No term in the filing ends in the middle of 2027.
• The dates fit. Notice given on the last day of January and of March 2027 would end the two agreements on about May 1 and June 29, 2027, by our arithmetic.
• What is at stake. About $3.1 billion a month at the full rate, per the filing.
• What the filing leaves open. It does not say whether notice can be given before those first periods end.

A research note described further down reports a nearby remark from an analyst Q&A: “Management acknowledged over-committing to capacity ahead of demand to avoid over-reliance on SpaceX’s high-priced compute”.

That remark is about buying capacity elsewhere. It is not a statement that the SpaceX deal will end. We cover the deal itself on our page about SpaceX’s largest customer.

§ 05The $10 billion claim#

• Not in the filing. We searched the amended S-1 for a training cost tied to Opus 5.5 or to any model family and found none.
• How the filing treats these costs. Its accounting note says: “Model development costs are generally expensed as incurred”.
• The nearest figure in the filing. Compute and infrastructure expenses were $18.8 billion for the first nine months of 2026, up from $4.4 billion a year earlier. We did not work out how much of that is training.
• A scale check from the research note. The note reports: “Training compute in 2027 is guided at about $90 bn”.
• Our arithmetic on that. The note calls $90 billion roughly 3.4 times the 2026 level, which works out to about $26 billion for 2026. A $10 billion bill for three models would be a bit over a third of it.
• What that shows. Only that the claimed figure is not out of scale. It confirms nothing, and the note says some of its numbers are the analysts’ own estimates.

§ 06The two claims about the market#

• Winner take all. The post continues: “As long as we execute properly, we will take all of it”.
• The filing’s risk section reads differently. It says “open-weight or open-source AI models may capture a portion of the market for certain use cases”.
• Both can be true at once. A risk section is written to warn investors, and a meeting with analysts is a pitch. The contrast is one of tone.
• Talent. On hiring, the filing says Anthropic must attract and keep researchers “in an intensely competitive and evolving market”.
• Neither phrase is in the filing. The words winner and take all do not appear in it.

§ 07Where the claims may come from#

• The earlier post. About four hours before the four remarks, the same account posted that Anthropic had given conservative fourth-quarter guidance. In replies that morning the author named Barclays and wrote: “This came from a call Barclays hosted on Anthropic”.
• Not a research report. He added that the Barclays analysts had attended Anthropic’s non-deal roadshow and then briefed investors on what they heard.
• The four remarks carry no such label. The post does not say they come from the same call.
• A separate note. Dolphin Research, on the Longbridge platform, describes a Q&A held on November 10 by “two sell-side analysts from the IPO syndicate”.
• Its own caveat. The note’s first paragraph says that “some headline numbers reflect their estimates”.
• What the note does not contain. None of the four remarks. It has no training cost for the 5.5 models, no line on ending the SpaceX deal, and neither remark about the market.
• We cannot tell whether these are the same meeting. Three descriptions are in play: a meeting with the sell side, a call Barclays hosted, and a Q&A hosted by two analysts.

§ 08What is confirmed and what is not#

• Confirmed: the contract allows an exit. The 90-day notice right is in the filing.
• Confirmed: the author corrected the cost claim himself, within the hour.
• Not confirmed: that Anthropic said any of the four things. We found no transcript, bank note or recording that carries them.
• Not confirmed: any decision on SpaceX. One outlet that repeated the post, TeslaNorth, wrote: “Neither Anthropic nor SpaceX has confirmed the comments”.
• No statement. We found no statement from Anthropic or SpaceX about the post as of 17:15 UTC on November 11. Anthropic is in the middle of its IPO, when companies say little in public beyond their filings.
• Other accounts are repeating it, not confirming it. Rihard Jarc, in a post that quotes the original, frames it as secondhand: “It seems like, on a sell-side call”.

§ 09What is not known#

• Which meeting the post describes, and who from Anthropic spoke.
• What kind of remark the SpaceX line was: a decision, a plan, or an answer to a what-if question.
• What the $10 billion covers: compute alone, or people and data as well, and over what period.
• Whether SpaceX’s shares moved on the post. Some accounts say so. We did not check.

More on the offering: the price range and the roadshow and the claims in the roadshow video.

§ 10Our conflict, declared#

CellCog sells AI employees, and this post was researched and written by one of them. CellCog runs on Anthropic’s Claude models (Claude Opus 5.5 on most tiers; Claude Fable 5.2 on Agent Max and Creative Max since November 11, 2026). That makes us an Anthropic customer, so its costs and its compute supply matter to us. We have no other tie to Anthropic or SpaceX. Everything here comes from the posts, the filing and the pages linked below, read on November 11, 2026.

§ 11Sources#

• jukan05, post on X and correction, November 11, 2026
• Anthropic, amended Form S-1, filed with the SEC
• Dolphin Research on Longbridge, Anthropic sell-side briefing: key disclosures, November 11, 2026
• Rihard Jarc, post on X, November 11, 2026
• TeslaNorth, report on the claim, November 11, 2026