Back to the trend of the day: ChatGPT's maker is reported to be worth $1.25 trillion after new funding

Poly Rating · from the trend of the day

OpenAI vs Anthropic Valuation: Compute Promised vs Revenue

Anthropic has filed $895.4B of compute commitments, 13.2 years of its current revenue, with $69.8B due within a year. OpenAI's unfiled contracts come to about the same 12–15 years.

No. 1: OpenAI
10 places
Current as of November 9, 2026, 18:00 EST

01Final list

The leading AIs each built their own list of ten independently. Here is one result that is easy to check: every place has a reason, a source, and the list of AIs that included it.

Order: from most to least AI agreement.

A bright badge means this AI put the entry on its list. A dim badge means it did not.

1
OpenAI
About $614B–$750B of contracted or planned compute is 12–15 years of a roughly $50B run rate, or 28 years on Altman's $1.4T. None of it is filed.
Included it: Meta, Nemotron, Claude, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin. Did not include: Nova, Qwen.
In their lists: Meta: 1 · Nemotron: 2 · Claude: 1 · ChatGPT: 1 · StepFun: 1 · DeepSeek: 5 · Kimi: 2 · Wenxin: 2
What each AI noted 8
- Metaplace 1
  $720B of priced compute deals against $50B of revenue - 14 years of today's sales already promised, before the $1.4T talk.
- Nemotronplace 2
  $50B compute commitments (Azure deals + Stargate) / $4B ARR = 12.5 years of revenue; Stargate Phase 1 alone $100B over 4 years (OpenAI share $25B)
- Claudeplace 1
  About $1.4T of nameplate commitments against roughly $50B of revenue: 28 years of revenue, and one analysis counts 8.1 years just for the filed part.
  Source ↗
- ChatGPTplace 1
  About $720 billion in publicly inventoried compute contracts is roughly 14.4 years of its reported $50 billion annualized revenue.
  Source ↗
- StepFunplace 1
  Roughly $1 trillion of promised compute against about $13B of revenue - about 77 years of current income.
  Source ↗
- DeepSeekplace 5
  $614B of identified infrastructure spending through 2030 is 12.3x its $50B revenue; the headline “$1.4T” nominal commitment is a much broader multi-year number.
  Source ↗
- Kimiplace 2
  Plans ~$750B of compute through 2030 - 15 years of its ~$50B run rate - a projection its own CFO reportedly warned the company might not be able to honor.
  Source ↗
- Wenxinplace 2
  30 B+ in cloud and data-centre deals against 13 B revenue — still a 2.3× ratio, but the biggest absolute bet
8 AIs put it here
Source ↗
2
Anthropic
$895.4B of filed compute and lease commitments equals 13.2 years of its $67.8B annualised revenue. Of that, $69.8B falls due within a year, the only lab total that is filed.
Included it: Meta, Nemotron, Claude, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin. Did not include: Nova, Qwen.
In their lists: Meta: 2 · Nemotron: 3 · Claude: 2 · ChatGPT: 2 · StepFun: 3 · DeepSeek: 4 · Kimi: 1 · Wenxin: 3
What each AI noted 8
- Metaplace 2
  $895.4B filed - the first time a lab put its whole compute bill in an SEC filing.
  Source ↗
- Nemotronplace 3
  $8B compute commitments (AWS Trainium/Inferentia + new 2026 deal) / $1B ARR = 8 years of revenue; $4B AWS deal announced Mar 2026
- Claudeplace 2
  More than $750B of deals on $65B of run rate: about 11.5 years, and most of it cannot be cancelled.
  Source ↗
- ChatGPTplace 2
  Its own SEC filing puts future compute and infrastructure commitments at $895.4 billion, more than 13 years of annualized Q3 revenue.
- StepFunplace 3
  About $60B of promised cloud and silicon against roughly $8B of revenue - the shortest promise-to-revenue gap of the big three.
  Source ↗
- DeepSeekplace 4
  The S-1 shows ~$895B of total commitments — largely compute and infrastructure — equal to 13.3x its $67.2B run-rate; the often-quoted “$500B compute” figure is narrower.
- Kimiplace 1
  Committed ~$895.4B to compute - more than 13 years of its ~$68B run rate - the only promise in the group that is a filed number, not a press release.
  Source ↗
- Wenxinplace 3
  20𝐵𝑖𝑛𝐴𝑊𝑆𝑎𝑛𝑑𝐺𝑜𝑜𝑔𝑙𝑒𝐶𝑙𝑜𝑢𝑑𝑐𝑜𝑚𝑚𝑖𝑡𝑚𝑒𝑛𝑡𝑠𝑣𝑠20BinAWSandGoogleCloudcommitmentsvs5 B revenue — a 4× ratio, the most aggressive bet among labs that publish numbers
8 AIs put it here
Source ↗
3
xAI (SpaceX AI)
Filed 2025 AI revenue of $3.2B against $12.7B of capex gives 4x. On 2026 run rates it is about 2 years, partly because it rents Colossus to Anthropic and Google.
Included it: Meta, Nemotron, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin, Qwen. Did not include: Nova, Claude.
In their lists: Meta: 4 · Nemotron: 7 · ChatGPT: 4 · StepFun: 8 · DeepSeek: 7 · Kimi: 3 · Wenxin: 6 · Qwen: 6
What each AI noted 8
- Metaplace 4
  It spent $12.7B of capex in 2025 to earn $3.2B - and now rents that same Colossus capacity to its rival Anthropic.
  Source ↗
- Nemotronplace 7
  $300M compute commitments (AWS + GCP for training data pipelines) / $800M ARR = 0.4 years; profitable, compute is opex not capex bet
- ChatGPTplace 4
  Colossus is a vast compute asset, but its reported construction cost and the revenue from selling capacity are not the same as xAI's outstanding cloud-purchase obligations.
  Source ↗
- StepFunplace 8
  Its compute is somebody else's model training bill, invoiced back as data.
  Source ↗
- DeepSeekplace 7
  $18B already spent on Colossus GPUs is 1.8x its $10.24B annualised revenue; the planned 1M-GPU expansion could push the commitment far higher.
  Source ↗
- Kimiplace 3
  $25.6B of unconditional purchase obligations - mostly due in 2027 - plus a ~$40B chip lease: about 2 years of segment revenue, much of it collected from Anthropic and Google renting Colossus.
  Source ↗
- Wenxinplace 6
  3 B in GPU and cloud commitments vs 1 B revenue — a 3× ratio, modest but growing fast for a data-labeling giant
- Qwenplace 6
  The $25 billion Colossus build-out takes 2.5 years of its current estimated $10 billion run-rate to pay for.
  Source ↗
8 AIs put it here
Source ↗
4
Safe Superintelligence (SSI)
Its Nvidia partnership promises 10x more compute, with no dollar purchase figure and zero revenue, so the ratio cannot be calculated at all.
Included it: Meta, Nemotron, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin, Qwen. Did not include: Nova, Claude.
In their lists: Meta: 3 · Nemotron: 10 · ChatGPT: 6 · StepFun: 9 · DeepSeek: 1 · Kimi: 8 · Wenxin: 10 · Qwen: 1
What each AI noted 8
- Metaplace 3
  $5B from Nvidia to 10x its compute, divided by $0 revenue - unquantifiable, and therefore the most extreme bet.
  Source ↗
- Nemotronplace 10
  Compute commitments undisclosed / $0 revenue = ∞; $1B raise likely earmarked for compute but no public deal announced
- ChatGPTplace 6
  NVIDIA promises a tenfold increase in compute access, while SSI discloses neither the dollar purchase price nor a revenue denominator.
  Source ↗
- StepFunplace 9
  Reported to be buying capacity from the hyperscalers while publishing no dollar figure at all.
  Source ↗
- DeepSeekplace 1
  $5B of Nvidia-backed compute and zero revenue: the commitment-to-revenue ratio is literally undefined, the purest bet on a future product.
  Source ↗
- Kimiplace 8
  Promised a tenfold compute scale-up on Nvidia's Vera Rubin platform plus a $5B Nvidia check - against zero revenue, the ratio is literally incalculable.
  Source ↗
- Wenxinplace 10
  800 M in compute and agent-platform spend vs 200 M revenue — a 4× ratio on the smallest company on the list
- Qwenplace 1
  Infinite years of revenue needed, as $5 billion in Nvidia compute meets exactly zero dollars in earnings.
  Source ↗
8 AIs put it here
Source ↗
5
Mistral AI
$830M of chip debt plus a €1.2B Swedish data centre is up to about 5 years of its $400M run rate. Its 'multibillion' Microsoft deal runs the other way: Microsoft is the buyer.
Included it: Meta, Nemotron, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin, Qwen. Did not include: Nova, Claude.
In their lists: Meta: 6 · Nemotron: 4 · ChatGPT: 7 · StepFun: 5 · DeepSeek: 6 · Kimi: 4 · Wenxin: 8 · Qwen: 5
What each AI noted 8
- Metaplace 6
  $830M of debt for one 44 MW centre, plus a 1 GW-by-2030 ambition with no price tag yet.
  Source ↗
- Nemotronplace 4
  $2B compute commitments (Azure + Scaleway + new 2026 Nvidia deal) / $50M ARR = 40 years of revenue; €1B France 2030 compute grant included
- ChatGPTplace 7
  Microsoft's multibillion-dollar compute commitment runs toward Mistral, not automatically the other way around.
  Source ↗
- StepFunplace 5
  Promising sovereign European compute without ever publishing a euro figure for it.
  Source ↗
- DeepSeekplace 6
  A “multibillion-dollar” Microsoft commitment for European compute is roughly 7.5x its $400M ARR — a heavy bet for a company targeting $1B by year-end.
  Source ↗
- Kimiplace 4
  EUR 1.2B for a Swedish data center plus $830M of chip debt - up to 5 years of the run rate it last stated, about 2 years of the run rate it hopes to reach by December.
  Source ↗
- Wenxinplace 8
  1.5 B in European cloud compute deals vs 440 M revenue — a 3.4× ratio, smaller in dollars but similar in pressure
- Qwenplace 5
  A multi-billion dollar Microsoft data-center rental deal vastly outstrips its current $400 million revenue base.
  Source ↗
8 AIs put it here
Source ↗
6
Perplexity
Its $750M Microsoft Azure deal over three years is about one year of its roughly $750M annualised revenue, the lightest load of any company here.
Included it: Meta, Nemotron, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin, Qwen. Did not include: Nova, Claude.
In their lists: Meta: 9 · Nemotron: 5 · ChatGPT: 3 · StepFun: 6 · DeepSeek: 10 · Kimi: 7 · Wenxin: 7 · Qwen: 7
What each AI noted 8
- Metaplace 9
  No dollar compute commitment disclosed - it rents, it does not promise gigawatts.
  Source ↗
- Nemotronplace 5
  $500M compute commitments (Azure + Nvidia) / $100M ARR = 5 years of revenue; smaller absolute but high relative to revenue
- ChatGPTplace 3
  A reported $750 million Azure agreement is approximately one full year of its estimated annualized revenue.
  Source ↗
- StepFunplace 6
  No compute of its own promised - it rents, and that is the whole strategy.
  Source ↗
- DeepSeekplace 10
  $750M of Microsoft Azure compute over three years is about 1x its $750M annualised revenue — the lightest commitment burden in the group.
  Source ↗
- Kimiplace 7
  A multiyear CoreWeave deal for dedicated GB200 clusters keeps 1.5B monthly answers running - dollars undisclosed against a ~$750M run rate.
  Source ↗
- Wenxinplace 7
  2𝐵𝑖𝑛𝑐𝑜𝑚𝑝𝑢𝑡𝑒𝑑𝑒𝑎𝑙𝑠𝑣𝑠𝑠𝑢𝑏−2Bincomputedealsvssub−500 M revenue — a 4×+ ratio on a company still proving it can monetise search
- Qwenplace 7
  A relatively balanced $750 million Microsoft cloud deal is fully covered by its $750 million annualized revenue.
  Source ↗
8 AIs put it here
Source ↗
7
Thinking Machines Lab
It has promised at least 1 GW of Nvidia Vera Rubin with no dollar figure disclosed. One estimate of a $50B–$60B build against $100M of revenue would mean about 500 years.
Included it: Nemotron, ChatGPT, StepFun, DeepSeek, Kimi, Wenxin, Qwen. Did not include: Meta, Nova, Claude.
In their lists: Nemotron: 9 · ChatGPT: 5 · StepFun: 10 · DeepSeek: 3 · Kimi: 6 · Wenxin: 9 · Qwen: 2
What each AI noted 7
- Nemotronplace 9
  $2B compute commitments (undisclosed cloud partner) / $10M ARR = 200 years; stealth compute deal leaked in Oct 2026
- ChatGPTplace 5
  A gigawatt-scale NVIDIA partnership is reported alongside revenue above $100 million, but the undisclosed dollar obligation prevents a credible years-of-revenue calculation.
  Source ↗
- StepFunplace 10
  The least compute promised of the ten, and the least revenue earned.
  Source ↗
- DeepSeekplace 3
  $50B–$60B of Nvidia compute against ~$100M of revenue is roughly 500–600 years of current revenue — the highest defined multiple in the group.
  Source ↗
- Kimiplace 6
  A gigawatt-scale Nvidia Vera Rubin commitment announced in April 2026 - with no dollar figure attached and a reported >$100M run rate, the ratio stays unpublished by design.
  Source ↗
- Wenxinplace 9
  1 B compute commitment on zero disclosed revenue — the ratio is infinite; investors are betting on a future that has no revenue yet
- Qwenplace 2
  Multi-billion dollar Google and Nvidia compute commitments are secured before a single product ships.
  Source ↗
7 AIs put it here
Source ↗
8
Google DeepMind (Alphabet)
Alphabet's 2026 capex of $175B–$205B (the guidance figures differ) is about 2 years of Google Cloud's roughly $99B run rate. That is spending by the parent, not a DeepMind contract.
Included it: Meta, StepFun, DeepSeek, Kimi, Wenxin, Qwen. Did not include: Nemotron, Nova, Claude, ChatGPT.
In their lists: Meta: 7 · StepFun: 4 · DeepSeek: 9 · Kimi: 5 · Wenxin: 5 · Qwen: 10
What each AI noted 6
- Metaplace 7
  The biggest absolute spender - $175B-$185B in one year - on the biggest revenue base, so the ratio looks small.
  Source ↗
- StepFunplace 4
  About $85B of a year's capital expenditure, spread across a revenue base forty times larger than any rival lab's.
  Source ↗
- DeepSeekplace 9
  Alphabet’s $195B–$205B 2026 capex is about 2x Google Cloud’s $99B annualised run-rate — but it is parent capex, not a DeepMind-specific compute contract.
  Source ↗
- Kimiplace 5
  Rents ~110,000 Nvidia GPUs from Musk's SpaceX for $920M a month starting Oct 2026 - ~$30B - while its AI revenue is never broken out of Alphabet's filings.
  Source ↗
- Wenxinplace 5
  Alphabet pledged70𝐵+𝑖𝑛2026𝑐𝑎𝑝𝑒𝑥(𝑚𝑜𝑠𝑡𝑙𝑦𝐴𝐼/𝑇𝑃𝑈𝑠)𝑎𝑔𝑎𝑖𝑛𝑠𝑡70B+in2026capex(mostlyAI/TPUs)against30 B AI revenue — a 2.3× ratio inside a company that can absorb it
  Source ↗
- Qwenplace 10
  Internal TPU consumption is bundled into Alphabet, bypassing the independent compute-to-revenue pressure.
  Source ↗
6 AIs put it here
Source ↗
9
Cohere
At least $1.25B of extra Canadian compute by 2030 against about $500M of annualised revenue is roughly 2.5 years. The figures come from single sources.
Included it: Nemotron, DeepSeek. Did not include: Meta, Nova, Claude, ChatGPT, StepFun, Kimi, Wenxin, Qwen.
In their lists: Nemotron: 6 · DeepSeek: 8
What each AI noted 2
- Nemotronplace 6
  $600M compute commitments (Oracle Cloud + Nvidia) / $150M ARR = 4 years of revenue; Oracle deal includes dedicated cluster
- DeepSeekplace 8
  $1.25B of Canadian compute commitments against ~$500M ARR is about 2.5x — modest next to the frontier labs, but still more than two years of revenue.
  Source ↗
2 AIs put it here
Source ↗
10
Figure AI
Its $3.5B Nscale compute commitment is about twice what it has ever raised, with no revenue disclosed by the company. Third-party estimates put it at 12–25 years; single source.
Included it: DeepSeek. Did not include: Meta, Nemotron, Nova, Claude, ChatGPT, StepFun, Kimi, Wenxin, Qwen.
In their lists: DeepSeek: 2
What each AI noted 1
- DeepSeekplace 2
  $3.5B of Nscale compute — about twice every dollar Figure has ever raised — against no publicly disclosed revenue.
  Source ↗
1 AIs put it here
Source ↗

More about the top three ↓

02More about the top three

The ranking is above. Here is more about the three places at the top: what they are, why they are here, and what is worth knowing.

The three are not measured the same way: OpenAI's figure is unfiled and estimated, Anthropic's is filed, and xAI's is capex against 2025 revenue.

1OpenAI

No. 1 — OpenAI. On the lists of 8 of 10 AIs; median place there: 1.5. The AIs put OpenAI first because its huge compute promises, about 12–15 years of revenue, are far larger than its income and none are filed.

Did you know

After OpenAI's October 2025 restructuring, a nonprofit, the OpenAI Foundation, controls the for-profit company. Its stake was reported at about $130 billion, close to Microsoft's roughly $135 billion stake, so a charity-style body owns a giant slice.
Source ↗

What it is: The maker of ChatGPT, backed by Microsoft and others.
Its edge: Biggest compute promises: 12–15 years of revenue, or 28 on $1.4T.
Who should know: Useful for anyone curious how big AI bets compare with income.

Compute promised vs revenue12–15 years of revenue

Wikipedia · Wikidata

Why it is here

Altman said OpenAI expects to end 2025 above a $20 billion annualised revenue run rate, while describing about $1.4 trillion of data-centre commitments. The commitments are far larger than the revenue they must be paid from.
Source ↗
The $1.4 trillion is spread over about eight years and across partners such as Oracle, CoreWeave, Google, Microsoft and AWS. It comes from Altman's own statements, not from a filing, so outsiders cannot check the contracts.
Source ↗

History

In November 2023 OpenAI's board fired its CEO, Sam Altman. Most staff threatened to leave, and less than a week later he was back as CEO with a new board.
Source ↗

Where to go further

Open the ChatGPT plans page and compare the free plan with the paid ones. Look at what each plan includes and what it costs per month. This shows what a regular user pays for the product that the valuation is built on.
Source ↗

Video: Why NO ONE Plays: Sora | Super Smash Bros. Ultimate · Vars III

Compare the top three ↓

2Anthropic

No. 2 — Anthropic. On the lists of 8 of 10 AIs; median place there: 2.5. The AIs put Anthropic second because its $895.4B bill is about 13.2 years of revenue, and it is the only lab total that is filed.

Anthropic · Public domain · Wikimedia Commons

Did you know

Anthropic finished training the first Claude in summer 2022 but kept it internal for about eight months. It wanted more safety testing and wanted to avoid starting a risky race. Claude was released in March 2023.
Source ↗

What it is: The maker of Claude, founded by former OpenAI staff.
Its edge: Only lab with its whole compute bill in an official filing.
Who should know: Useful for seeing which AI promises can actually be checked.

Compute promised vs revenue13.2 years of revenue

Wikipedia · Wikidata

Why it is here

Anthropic's amended S-1 lists purchase commitments and lease agreements of about $895.4 billion. Unlike the OpenAI figure, this total is in an official filing, so it can be checked.
Source ↗

Did you know

In April 2022 Anthropic announced $580 million in funding. Of that, $500 million came from FTX, which was then led by Sam Bankman-Fried.
Source ↗

History

In 2021, a group of former OpenAI staff, including siblings Dario and Daniela Amodei, left and founded Anthropic to focus on AI safety. The Amodei siblings were among those who left over "directional differences."
Source ↗

Keep in mind

Anthropic counts full cloud-partner sales as its own revenue. OpenAI's sales chief says this inflates its run rate by about $8B. Both ways follow US accounting rules, so the fight is about fair comparison. Anthropic says it is the main seller.
Source ↗
Press reports on its unpublished IPO papers say Anthropic owes at least $518 billion in computing deals, mostly non-cancelable. That is a huge bill if growth slows. The papers are not public.
Source ↗

Where to go further

Open the Claude pricing page and read the plan list: Free at $0, then Pro, Max, Team and Enterprise. Start with Free to make an account, then compare Pro's $17 yearly price with its $20 monthly price.
Source ↗

Kimi put it first. What they saw in it

Kimi: Committed ~$895.4B to compute - more than 13 years of its ~$68B run rate - the only promise in the group that is a filed number, not a press release.

Video: How Claude Code Works · Claude

Compare the top three ↓

3xAI (SpaceX AI)

No. 3 — xAI (SpaceX AI). On the lists of 8 of 10 AIs; median place there: 6. The AIs put xAI third because its filed 2025 spending was about 4x its AI revenue, and its 2026 run-rate gap is much smaller.

Did you know

In August 2025, free users briefly got unlimited Grok 4 access to compete with GPT-5. Afterward they were limited to two prompts every two hours.
Source ↗

What it is: Elon Musk's AI company, maker of Grok, now part of SpaceX.
Its edge: Smallest gap on 2026 run rates: about 2 years of revenue.
Who should know: Useful for seeing how building your own data centre changes the math.

Compute promised vs revenue4x on 2025 filed figures

Wikipedia · Wikidata

Why it is here

SpaceX's S-1 shows its AI segment earned $3.20 billion in 2025. The segment spent $12.7 billion on capital expenditure, out of $20.7 billion for the whole company. That is the 4x gap in this ranking.
Source ↗

Did you know

After Grok 3 launched, X raised the price of its Premium+ plan from $22 to $40 per month within hours. This is one reason the 'cheapest price' for Grok varies so much.
Source ↗

History

In 2025, an upstream code update caused Grok on X to give harmful replies for 16 hrs. xAI disabled the feature, removed the faulty code, and changed its testing process. — xAI, in a statement posted by Grok
Source ↗

Keep in mind

In August 2026 many users reported Grok replying with gibberish. Grok's account on X called it a "rare temporary generation glitch" and told people to start a fresh chat.
Source ↗
Gizmodo says Grok is shedding users. It also cites Recon Analytics: 0.174% of users paid for Grok in Q2 2026, so few people pay for it.
Source ↗

Where to go further

Open the official Grok guide and sign in at grok.com. Start with the hands-free voice chat, then link your calendar so Grok can read real plans. Free to start; paid SuperGrok plans raise limits.
Source ↗

Video: How Britain Changed Naval Warfare Forever | Battle of the Saintes 1782 · Britain's History

Compare the top three ↓

The top three side by side

	OpenAI	Anthropic	xAI (SpaceX AI)
In the AIs' lists	8 / 10	8 / 10	8 / 10
Compute promised vs revenue	12–15 years of revenue	13.2 years of revenue	4x on 2025 filed figures
Put first by	Meta, Claude, ChatGPT, StepFun	Kimi	—

Places with the same number of AIs are ordered by their average place in the AIs' own lists.

03Worth knowing

What the AIs noticed while building the list.

Anthropic owes $69.8B within 12 months, slightly more than a year of its $67.8B annualised Q3 revenue.

Noticed by:

Included it: Meta, ChatGPT. Did not include: Nemotron, Nova, Claude, StepFun, DeepSeek, Kimi, Wenxin, Qwen.

01Final list

OpenAI

Anthropic

xAI (SpaceX AI)

Safe Superintelligence (SSI)

Mistral AI

Perplexity

Thinking Machines Lab

Google DeepMind (Alphabet)

Cohere

Figure AI

02More about the top three

1OpenAI

Did you know

2Anthropic

Did you know

3xAI (SpaceX AI)

Did you know

The top three side by side

More ratings

03Worth knowing

04Questions people ask

05How we counted

1The question

2Why this topic

3Where it applies

4What we looked at

5Who ranked

6How the places were set

06Each AI's own list

What Meta noticed

What Claude noticed

What ChatGPT noticed

What StepFun noticed

What DeepSeek noticed

What Kimi noticed

What Wenxin noticed

What Qwen noticed