ED ZITRON: Take-or-pay contracts make up a large part of the $1 trillion, so the funded and memorandum distinction does not rescue the original claim

The Gist

Ed Zitron says many of the deals Baker called unserious are take-or-pay contracts, which Baker himself says are serious. So Baker's argument contradicts itself unless he shows which deals are just memos. This is a steelman: each side is reconstructed in its strongest fair form, not endorsed, and no winner is declared.

Conclusion

Because a large share of the commitments Baker dismissed are take-or-pay contracts that his own standard treats as serious, the dismissal does not hold for that share, and the headline aggregate should be broken down before anyone relies on it.

Premises

1. Baker's original post called the roughly $1 trillion of commitments ridiculous and not even vaguely serious.
2. Reuters reported that Anthropic's commitments are largely take-or-pay or non-cancellable, with about $111 billion to Google, $110 billion to Amazon and $31 billion to Microsoft payable regardless of usage.
3. Baker's later posts say contracts, including take-or-pay, are taken seriously.
4. If much of the $1 trillion is take-or-pay, the later standard contradicts the original dismissal; Baker did not provide a memorandum example when asked.
5. The $1 trillion headline itself is an aggregate without a stated method.

Assumptions

• Baker's term was "unfunded", and a take-or-pay contract can be unfunded; the disagreement partly turns on whether "unfunded" or "contract" governs.
• OpenAI's commitments include more memoranda than Anthropic's, so the share differs by lab.
• "Made up number" is read as criticism of an unsourced aggregate, which applies equally to the $1 trillion figure in both men's usage.

Analysis

Argument type: Inductive.

Potential Fallacies

• Hasty generalization / scope shift (P2 to P4 and the title/conclusion) — Evidence about one lab (about $252B of Anthropic take-or-pay obligations) is used to support a claim about 'a large share' of a multi-lab $1T aggregate. Assumption A2 says the mix differs by lab and that OpenAI has more memoranda, so the evidence does not transfer without a figure for the rest of the total.
• Unestablished antecedent (conditional treated as finding) (P4 to Conclusion) — P4 is conditional ('If much of the $1 trillion is take-or-pay'), but the conclusion asserts that antecedent as established. No premise quantifies the take-or-pay share of the total, so the modus ponens is incomplete.
• Equivocation on 'serious' (P1, P3, P4) — 'Not even vaguely serious' may refer to the plausibility or funding of the aggregate, while 'contracts are taken seriously' refers to legal bindingness. Assumption A1 acknowledges these can come apart, so the claimed contradiction depends on treating them as one standard. The argument concedes the ambiguity yet still presents the inconsistency as the central finding.
• Argument from silence (P4 (final clause)) — Baker's not supplying a memorandum example in a social-media thread is treated as meaningful evidence about the composition of the $1T. Non-response has many innocent explanations, and memoranda and letters of intent are known to exist in this sector, so the inference is weak.
• Self-undermining symmetry / partial tu quoque (P5, A3) — P5 and A3 say the $1T has no stated method, which supports disaggregation. But the argument also relies on shares of that same figure, so uncertainty about the denominator weakens its own numerator. Noting that both sides use the figure also does not establish whether it is right.
• Is-ought gap (Conclusion (second clause)) — The call that the aggregate 'should be broken down before anyone relies on it' rests on an unstated bridging norm: unsourced, heterogeneous aggregates should be decomposed before use. The norm is reasonable, but it is not stated.

Counterarguments

These challenges are not ranked.

• Conclusion (High impact) — The documented take-or-pay total is about $252B, a quarter of $1T, and nothing is shown for the other three quarters. On the argument's own premises, Baker's dismissal could still stand for most of the aggregate, and 'a large share' is overstated.
• Premise 4 (High impact) — Baker can hold a consistent position: contracts are serious, but a blended aggregate mixing binding contracts, memoranda, options and possibly double-counted items is not. Criticizing the aggregate is not criticizing every component.
• Premise 3 (High impact) — A take-or-pay obligation is only as good as the obligor's ability to pay. If the lab lacks revenue or financing, the contract is an unsecured claim on a possibly strained counterparty, so 'unfunded' remains a legitimate criticism and does not conflict with taking contracts seriously.
• Premise 2 (Medium impact) — Reuters' characterization is secondary and may flatten real terms such as multi-year back-loaded payments, minimum-volume floors, step-downs, force majeure and renegotiation, which would make 'payable regardless of usage' overstate effective liability. The $252B may also overlap with, or sit outside, the $1T.
• Premise 5 (Medium impact) — If the $1T has no stated method, any decomposition of it is equally unsourced, which undercuts the shares the argument wants to draw from it. The criticism applies to the author's own use of the figure as well.
• Premise 4 (Low impact) — Failing to answer a request for an example on social media is a weak signal. Public non-binding deals such as letters of intent exist in this sector, so the memorandum category is real regardless of Baker's reply.

Suggested Improvements

• Quantifying the take-or-pay share — Add a premise or table that classifies the $1T by lab, counterparty and instrument (take-or-pay, cancellable, memorandum or letter of intent), with sources, and states the take-or-pay fraction of the total. 'Large share' is the load-bearing claim and is currently unquantified. About a quarter is documented, and a figure for OpenAI and others would either confirm or refute the thesis.
• Defining 'serious' — State separately whether Baker's dismissal was about legal bindingness, funding capacity, or the aggregate's construction. Quote the original and later posts in full, and argue that the later standard covers the same property as the original. This would close the equivocation gap that A1 leaves open and show that the two statements actually conflict.
• Addressing funding — Add evidence on the funding side (revenue, capital raised, cash coverage against scheduled payments) for the take-or-pay obligations. Because 'unfunded' may govern, showing that binding obligations are also fundable, or not, speaks directly to Baker's actual term.
• Narrowing the claim — Reframe the title and conclusion as 'the blanket dismissal is overbroad for at least the documented take-or-pay portion' and not 'does not rescue the original claim'. This matches what the evidence supports and keeps the strong decomposition recommendation without overreaching.
• Consistency of the aggregate critique — Apply the demand for a method to the author's own use of the $1T figure, and state a materiality threshold for when disaggregation is required. This removes the self-undermining symmetry from P5 and A3 and keeps the normative principle workable.
• Contract terms — Verify the Reuters figures against filings or other reporting and note duration, ramp, termination and renegotiation terms. Nominal total value and effective near-term liability differ, and this affects how 'serious' each obligation really is.

Scenario Tests

• A full breakdown shows that take-or-pay contracts make up well over half of the $1T, spread across several labs. (Supports) — The 'large share' claim is vindicated and Baker's blanket dismissal is clearly overbroad, though the funding question remains.
• A breakdown shows take-or-pay is a minority (e.g. under 30%), with the rest non-binding memoranda, letters of intent and vendor-financing announcements. (Challenges) — Baker's dismissal is largely vindicated for the aggregate, leaving only a narrow partial point about the binding portion.
• Baker's original term 'unfunded' is read as the governing criterion, and the labs lack financing for their obligations. (Challenges) — The take-or-pay evidence becomes largely irrelevant to the dismissal, since a binding contract can still be unfunded and unserious in that sense.
• Reuters' figures are confirmed by filings, and the labs' revenue and financing credibly cover the scheduled payments. (Supports) — Binding obligations are shown to be both contractual and financeable, which strengthens the claim that the dismissal fails for that share.
• The $1T is found to double-count the same compute across providers and chip vendors. (Neutral) — Both the original number and any decomposition of it are discredited, which supports the call for a stated method but not either party's substantive claim.

Coherence & Relevance

The argument is coherent in structure, as an internal critique that applies Baker's later standard to his earlier dismissal and ends with a modest call to disaggregate. It is not fully sufficient, because the quantitative bridge from one lab's documented take-or-pay obligations to 'a large share' of the whole is missing, and the alleged contradiction depends on an unresolved ambiguity between 'unfunded' and 'contract' that the stated assumptions concede. The decomposition recommendation is well supported, while the stronger claim that the dismissal fails for a large share is only partly established.

• Baker's original post called the roughly $1 trillion of commitments ridiculous and not even vaguely serious. (Strong) — It establishes the target of the rebuttal but gives no evidence about the composition of the $1T, and it does not say whether the dismissal covered binding contracts.
• Reuters reported that Anthropic's commitments are largely take-or-pay or non-cancellable, with about $111 billion to Google, $110 billion to Amazon and $31 billion to Microsoft payable regardless of usage. (Moderate) — It covers one lab and about a quarter of the aggregate. No bridge links Anthropic's mix to the whole $1T, and A2 says the mix differs by lab.
• Baker's later posts say contracts, including take-or-pay, are taken seriously. (Strong) — It supplies the standard that makes the inconsistency charge possible, but it needs a premise showing that 'serious' here means the same thing as in the original post, given the unfunded-versus-contract split.
• If much of the $1 trillion is take-or-pay, the later standard contradicts the original dismissal; Baker did not provide a memorandum example when asked. (Moderate) — The antecedent is never established, so the conclusion does not follow from the stated premises. The silence clause is weakly probative.
• The $1 trillion headline itself is an aggregate without a stated method. (Moderate) — It supports the decomposition recommendation well but is a double-edged point: it weakens confidence in any share computed from the aggregate.

View this argument on LogicFirst.ai