“The first and most obvious answer is money, which I estimate, back of the envelope as (roughly) a trillion dollars, much of it on Nvidia chips and massive salaries.”
Our reading, the sentence that was checked:
Supposing for the sake of argument that LLMs have run their course and new ideas are needed, Marcus estimates, back of the envelope, the money cost of the field taking so long to recognize this at roughly a trillion dollars, much of it on Nvidia chips and massive salaries.
Assessment: Unresolved Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, A trillion dollars is a terrible thing to waste, paragraph 15.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-a-trillion-dollars-is-a-terrible-1b7d29-a1
We recorded 14 claims from this post. See every claim from this post.
The wording above is the version this site observed on 2026-10-28. The source may have been edited after 2025-11-27. This page does not claim the wording was captured then.
What the claim commits to
- Kind
- Observation
- Deadline, as the speaker wrote it
- None stated. This site does not add one.
- Deadline state when assessed
- No horizon
- Deadline state on 2026-10-29
- No horizon (the release’s reference date)
Conditions
- supposing for the sake of argument that Sutskever and the rest of us are correct (qualifier, stated by the speaker)
- back of the envelope, roughly (qualifier, stated by the speaker)
From the speaker
- what did it cost the field and society
- suppose for the sake of argument that Sutskever and the rest of us are correct
- back of the envelope
- (roughly) a trillion dollars
- much of it on Nvidia chips and massive salaries
Our interpretation
- The plain reading is cumulative worldwide spending on the LLM scaling effort incurred by publication, November 27, 2025. Marcus supplies no starting date, so the start remains unspecified.
- The hypothetical concerns whether that spending was wasted. The amount spent can be examined without deciding whether LLMs can produce AGI.
- Roughly permits estimation uncertainty around $1 trillion. Marcus gives no numerical tolerance, and none is imposed.
- Much means a substantial share spent collectively on Nvidia chips and compensation; it does not necessarily mean a majority.
- Count spending once. Exclude financing raised and unspent future commitments. Broader cloud or AI spending requires an allocation to the LLM scaling effort.
- The accounting basis is estimate because Marcus explicitly presents his own back-of-the-envelope calculation.
Counts as supported if: A credible reconstruction of cumulative spending on the worldwide LLM scaling effort through publication places it near $1 trillion and identifies Nvidia chips and compensation as a substantial share, with uncertainty and overlapping payments addressed.
Counts as contradicted if: A sufficiently complete reconstruction places cumulative spending clearly outside a reasonable approximation to $1 trillion, or establishes that Nvidia chips and compensation were not a substantial share. A partial subtotal alone cannot contradict the total.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Unresolved
The record was checked, and the evidence available by the cutoff does not settle it.
Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.
- Evidence as of
- Deadline state when assessed
- No horizon
- Assessment recorded
- Policy version
- 2026-10-28.7
Rationale
The records establish large infrastructure expenditures, Nvidia sales and compensation offers. They do not establish roughly $1 trillion of cumulative LLM scaling spending by November 27, 2025, or the share spent on chips and salaries. Company capital expenditures cover broader activities, Nvidia revenue overlaps with customer purchases, and compensation offers are not payroll already incurred. These gaps prevent either confirmation or contradiction. The hypothetical claim that the spending was wasted is not settled by spending records.
Limitations
- The spending boundary and starting date remain ambiguous.
- All included evidence is contextual; none measures the full claimed total.
- Some candidate sources could not be read and were excluded.
- Failure to establish the total is not evidence that it was smaller.
Evidence (4 items)
Context only: Microsoft,
Capital expenditures were $34.9 billion
Why it matters here: This establishes substantial quarterly infrastructure spending. It covers cloud and AI offerings across Microsoft, rather than cumulative worldwide LLM scaling costs.
- Publisher
- Microsoft
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- Microsoft, Quarterly capital expenditures including finance leases, USD billions, company reported, 2025-07-01 to 2025-09-30
- Describes events of
- Public from
Context only: NVIDIA,
$51.2 billion
Why it matters here: Nvidia reported substantial Data Center sales before publication. Supplier revenue for one quarter does not identify cumulative LLM chip expenditure or its share of total spending, and cannot simply be added to customers' infrastructure purchases.
- Publisher
- NVIDIA
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Measures
- NVIDIA, Quarterly Data Center revenue, USD billions, company reported, ? to 2025-10-26
- Describes events of
- Public from
Context only: The Register,
at least $10,000,000 a year
Why it matters here: Original reporting provides context for very large compensation offers. It neither establishes payment nor measures aggregate compensation, and no primary record was available for this offer.
- Publisher
- The Register
- Kind of source
- News report
- Measures
- Meta, Annual compensation package offered to one AI researcher, USD dollars per year, estimate, ? to ?
- Describes events of
- Public from
Context only: Goldman Sachs,
Not all capex by hyperscalers is necessarily related to AI.
Why it matters here: This methodological caution explains why broad infrastructure totals cannot automatically be attributed to LLM scaling. The article's later-period forecasts are not treated as outcome evidence for spending through November 2025.
- Publisher
- Goldman Sachs
- Kind of source
- Secondary summary
- Describes events of
- Public from
What would settle it
A transparent reconstruction of worldwide LLM scaling expenditure through November 27, 2025, with a stated starting date, allocations for shared infrastructure, compensation incurred, uncertainty bounds and removal of overlapping supplier and customer payments.
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, A trillion dollars is a terrible thing to waste
The first and most obvious answer is money, which I estimate, back of the envelope as (roughly) a trillion dollars, much of it on Nvidia chips and massive salaries.
- Where
- paragraph 15
- Published
- 2025-11-27
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-a-trillion-dollars-is-a-terrible-1b7d29-a1: Current. Outcome: Unresolved. Evidence as of 2026-10-28. Public date not recorded.
Recorded events
Assessment publishedBy this site
First assessment published after both reviews and the final-text check passed.
About assessment marcus-a-trillion-dollars-is-a-terrible-1b7d29-a1.