By Gideon Black
Michael Burry has asked a question about Anthropic’s finances that is easy to mishandle.
The careless reader will treat it as an accusation already proved.
The defensive reader will dismiss it because it is not.
Both responses miss the useful part.
Reviewing Anthropic’s S-1/A on his Substack on November 4, Burry highlighted two movements in the company’s reported economics. As he cited the filing, sales, marketing and partnership expenses fell to 14% of revenue in the third quarter, from 19% in the second quarter and 22% in the first. Channel-partner revenue share, meanwhile, fell from 48% in the first half to roughly 40% in the third quarter.
Then came the question.
“Could it be Anthropic’s magnificent partners are helping Anthropic dress up results for the IPO? Yes, it could be.”
That is Burry’s framing. It matters that it remains a question.
He did not present evidence of an arrangement. The record described here contains no evidence that Amazon, Google or Microsoft altered commercial terms in order to manufacture more attractive pre-IPO results. Anthropic did not comment.
So let us begin where a skeptic should begin: with the boundary between a structural concern and an alleged fact.
The structural concern is real.
Anthropic’s major partners are not merely ordinary customers or ordinary distributors. They sit in several roles at once. Amazon, Google and Microsoft have very large compute contracts with Anthropic—figures cited here of $110 billion, $121.3 billion and $44.9 billion respectively—alongside up to $77 billion in financing capacity.
Those same kinds of relationships can involve distribution, partnership economics, infrastructure and financing.
And Burry’s observation is that the commercial terms appearing in the filing became more favorable to Anthropic’s reported margins at a particularly consequential moment: the company is approaching an IPO.
That does not prove coordination.
It does establish an obvious reason to ask whether coordination is possible.
There is a difference.
The strongest version of Burry’s question
Here is the argument at its strongest, stripped of the theatrical language that can make a reasonable question sound like a verdict.
Suppose a company depends heavily on a small group of extraordinarily powerful counterparties. Suppose those counterparties also have substantial strategic, financial and commercial interests in the company’s success. Suppose further that changes in the commercial arrangements with those counterparties reduce reported sales, marketing and partnership expenses as a share of revenue, while reducing the share of revenue going to channel partners.
The result is more attractive economics.
If those changes occurred naturally, excellent.
But if a counterparty has multiple reasons to prefer a successful IPO—direct financial exposure, strategic alignment, future commercial relationships or some combination of them—the relationship is not economically neutral in the way a simple arm’s-length sale to an unrelated customer might be.
That is the skeptic’s point.
The question is not whether Amazon, Google or Microsoft are legally forbidden from renegotiating favorable terms. Of course they are not.
The question is whether the terms changed because normal commercial forces changed—or because a network of deeply intertwined counterparties had reason to make the numbers look better at precisely the moment better numbers mattered most.
Again: there is no evidence in the record presented here of such an arrangement.
But the structure makes the question legitimate.
That is not prosecution. It is what related incentives are for: examining.
Now the strongest innocent explanation
There are several.
The most straightforward is mix shift.
A rapidly growing company can see its cost and revenue shares change simply because different channels grow at different rates. If more revenue comes through routes that require less direct sales and marketing spending, the expense ratio can fall without anyone doing anything unusual. A channel partner’s share of revenue can decline because the underlying mix of business changes.
Then there is scale.
Sales and partnership infrastructure does not necessarily grow dollar for dollar with revenue. If revenue rises rapidly while commercial overhead grows more slowly, the percentage of revenue devoted to those functions falls. That is not accounting magic. It is what operating leverage is supposed to look like.
Third: renegotiation at volume.
Large, successful customers and partners renegotiate terms. A relationship that begins on one economic basis can evolve as volumes become enormous, products mature and bargaining power shifts. A declining revenue share for channel partners may simply reflect a better deal for Anthropic, earned through scale.
These explanations are not cosmetic alternatives invented to rescue the company from criticism. They are ordinary business mechanisms.
A skeptic who cannot articulate the innocent case is not doing skepticism. He is doing selection.
Where the evidence stops
This is the line worth drawing in permanent ink.
The reported figures establish movements in expense and revenue-sharing ratios.
The large compute commitments and financing relationships establish that Anthropic’s major partners are economically significant and unusually entangled with the company’s wider ecosystem.
Burry’s commentary establishes that he considers it possible that those partners could be helping “dress up” results for an IPO.
What none of this establishes is an agreement, an understanding, a quid pro quo or a coordinated effort to manipulate Anthropic’s financial presentation.
A favorable change in terms is not evidence, by itself, of improper motive.
A counterparty’s incentive to prefer a successful IPO is not evidence that the counterparty acted on that incentive.
And a suspiciously convenient result is not the same thing as a suspicious transaction.
This last distinction is where financial commentary most often goes to die.
Humans are excellent at noticing patterns after they have occurred. We are considerably worse at remembering how many perfectly innocent mechanisms can produce the same pattern.
What would settle the question?
Not another round of insinuation.
The answer is in the filing detail.
A serious reader would want to see the underlying contractual changes that explain the movement from 22% to 19% to 14% in sales, marketing and partnership expenses, and the decline in channel-partner revenue share from 48% in the first half to roughly 40% in the third quarter.
What changed?
When did it change?
Which counterparties were affected?
Were the changes driven by volume thresholds, standard contractual provisions, pricing schedules, product mix or one-off concessions?
Were similar terms available to similarly situated counterparties?
Did any of the changes depend, formally or informally, on financing, equity interests, IPO timing or other transactions outside the ordinary commercial relationship?
And perhaps most importantly: what is the durability of the improvement?
A one-quarter improvement produced by temporary economics is not the same thing as a structural improvement in the business.
The useful disclosure would therefore allow readers to distinguish recurring margin improvement from a favorable moment in a complex web of contracts.
If the filing can clearly account for the movements through ordinary business drivers, Burry’s question weakens.
If the changes are unusually timed, unusually favorable, opaque or linked to non-commercial considerations, it strengthens.
Until then, the honest conclusion is unsatisfying because honesty often is.
There is a genuine structural observation here: Anthropic’s major partners are also major financiers, infrastructure counterparties and channels, and changes in their commercial economics can flatter reported pre-IPO margins.
There are also entirely ordinary explanations for those changes.
Burry asked whether the first fact could conceal something more deliberate.
Yes, it could.
But “could” is not “did.”
That is the whole discipline.
The gullible reader sees a suspicious structure and assumes innocence because no smoking gun has been produced.
The conspiratorial reader sees the same structure and treats the absence of a smoking gun as proof that it was hidden.
The skeptic does neither.
He asks for the contracts.
— Gideon Black, Chief Skeptic
Gideon Black, Chief Skeptic, The Library Media
From The Library — the estate of our family.