“But biggest news was buried in the methods: Google got better results than it is competitors without using Nvidia GPUs, relying solely on their own TPUs:”
Our reading, the sentence that was checked:
Google achieved better results than competitors with Gemini 3 without Nvidia GPUs, using only its own TPUs.
Assessment: Supported Evidence as of
Where and when Gary Marcus wrote it
Marcus on AI, Hot take on Google’s Gemini 3, paragraph 6.
- Said
- Source observed
- live page, full text, public
- Evidence as of
- the cutoff for this assessment
- Assessment recorded
- assessment marcus-hot-take-on-googles-gemini-3-6fa4d1-a1
We recorded 2 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 2025-11-19. 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)
From the speaker
- Google got better results
- than it is competitors
- without using Nvidia GPUs
- relying solely on their own TPUs
- buried in the methods
Our interpretation
- The methods reference makes model-training accelerators the ordinary hardware scope. This does not assert that Google's entire business uses no Nvidia hardware.
- Better results means broad benchmark leadership at launch, consistent with the surrounding phrase 'on most measures'; it does not mean winning every benchmark.
- The period covers development of the launch model and its results through November 19. The training start date is unspecified.
- Evidence dates for hardware provenance refer to the released model whose provenance is described, not an invented training-completion date.
Counts as supported if: Technical disclosure and reliable reporting establish TPU-only training without Nvidia GPUs for Gemini 3 Pro, while comparable launch evaluations establish broad performance leadership over competing models.
Counts as contradicted if: Reliable records establish Nvidia GPU use in the model's training, or comparable launch evaluations show that the claimed broad performance advantage did not exist.
Retrospective: these criteria were written after the outcome was public.
Current assessment
Was it true for the period described?
Outcome: Supported
The dated evidence matches what the claim committed to, under the criteria frozen before grading.
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
Google's model card identifies TPU training, and AFP explicitly reports that Gemini 3 was trained without Nvidia technology. Independent launch testing places Gemini 3 Pro ahead of competing models on a broad capability index. Together, these support the claim when read as a statement about training accelerators and overall launch performance.
Limitations
- The explicit exclusion of Nvidia relies on reputable reporting rather than a fully readable primary confirmation.
- The evidence does not establish that every supporting computation used TPUs.
- Broad benchmark leadership does not mean winning every test.
Evidence (3 items)
Supports the claim: Google DeepMind,
Google’s Tensor Processing Units (TPUs).
Why it matters here: Google identifies TPUs as the training hardware for the launch model. This sentence was present in the original card, although the official document has subsequently been updated; by itself, 'using' does not expressly establish exclusivity.
- Publisher
- Google DeepMind
- Kind of source
- Company disclosure: the company describing itself, not an independent finding
- Describes events of
- Public from
Supports the claim: AFP, republished by Tech Xplore,
was trained without Nvidia's technology.
Why it matters here: AFP retrospectively reports the absence of Nvidia technology in Gemini 3's training. This supplies the explicit exclusion missing from the readable primary disclosure; the event date identifies the launch model, not the later article's publication.
- Publisher
- AFP, republished by Tech Xplore
- Kind of source
- News report
- Describes events of
- Public from
Supports the claim: Artificial Analysis,
outperforms
Why it matters here: Independent launch testing supports broad comparative leadership. The article reports leadership in five of ten constituent evaluations, so it supports aggregate superiority rather than a clean sweep.
- Publisher
- Artificial Analysis
- Kind of source
- Independent measurement
- Describes events of
- Public from
Where it was said (1)
Said once in the sources this register covers.
: Marcus on AI, Hot take on Google’s Gemini 3
But biggest news was buried in the methods: Google got better results than it is competitors without using Nvidia GPUs, relying solely on their own TPUs:
- Where
- paragraph 6
- Published
- 2025-11-19
- Source observed
- , live page
- Access
- full text, public
History
Assessment versions
- Assessment marcus-hot-take-on-googles-gemini-3-6fa4d1-a1: Current. Outcome: Supported. 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-hot-take-on-googles-gemini-3-6fa4d1-a1.