Unique Research · Global AI Company Valuation Ranking · October 27, 2026
Twelve Chinese companies make the list — DeepSeek ranks third globally; Moonshot (Kimi) ranks fifth.
Unique Research · Global AI Company Valuation Ranking
$2.91 trillion in total. Two companies take 62.5%. Making this ranking already means you are not a small company — yet No. 1’s valuation is 386× No. 100’s.
Every figure below comes from our newly re-verified Global Unlisted AI-Native Company Valuation Database:
$2.91 trillion: simple sum of Top 100 valuations.
62.5%: Anthropic and OpenAI’s share; the other 98 companies split 37.5%.
34 companies have crossed $10 billion in valuation. The tip is extreme — and a second tier has already formed.
First, the caliber. This ranking covers global unlisted AI-native companies. Valuations use the most recent verifiable financing, secondary equity trades, or media-reported pricing, converted uniformly to USD billions; “over / at least” is shown at the lower bound. Pricing dates and deal calibers are not uniform across companies — this is not a live market-cap ranking, and it is not our value judgment. Cutoff date: October 27, 2026.
One hundred companies sum to $2,906.45 billion. Anthropic is $965.0 billion; OpenAI is $852.0 billion. Together they are $1,817.0 billion — 1.67× the combined total of the other 98.
Concentration stays extreme further down: Top 10 take 75.8%; Top 20 take 83.1%; ranks 51–100 sum to $192.59 billion — only 6.6%.
The ranking median is $5.55 billion; No. 100, Armadin, still has $2.5 billion. Making this list already means you are not small. But No. 1’s valuation is 386× No. 100’s. Sitting on the same ranking does not mean playing the same capital scale.
No. 3 DeepSeek, at about $85.71 billion under ranking caliber, already exceeds the combined AI-company valuations of many countries — yet is still only about one-eleventh of Anthropic. No. 4 Anduril is $61.0 billion; No. 5 Moonshot (月之暗面) about $50.0 billion; No. 6 Cognition $48.0 billion.
Global unlisted AI-native companies · Valuation TOP 5 (USD billions)
1 · Anthropic · 965.0
2 · OpenAI · 852.0
3 · DeepSeek · 85.71
4 · Anduril · 61.0
5 · Moonshot · 50.0
Geography is the same head-effect story. The United States: 74 companies totaling $2,627.63 billion — 90.4% of the ranking. Even stripping out Anthropic and OpenAI, the remaining U.S. companies still take 74.4% of residual valuation. America’s edge is not only the two model giants — it also covers Agents, chips, data infrastructure, and physical AI.
Europe: 10 companies totaling about ~$95.9 billion. France’s Mistral AI, Germany’s Helsing, Sweden’s Lovable and Legora, and the UK’s Wayve land respectively in foundation models, defense autonomy, coding, and specialized applications. Europe has no dual model giants at the same scale — but it has grown a set of clearly positioned specialty companies.
This ranking looks like a hundred-company free-for-all. In practice it is more like two aircraft carriers leading a fleet that is already differentiating.
On October 26, Anthropic publicly filed its S-1, planning to list on Nasdaq under the ticker ANT. Share count and price range are not yet set, so this ranking still uses the $965.0 billion post-money valuation from its May financing this year — and does not treat media-predicted IPO valuations as formal terms.
The filing arrives at the right moment. A valuation ranking records the prices investors were willing to pay; an S-1 is the first systematic public view of a frontier model company’s revenue, customers, cash flow, and long-term compute bill.
Start with growth. In H1 2026, Anthropic’s revenue was $16.286 billion, up about 1,300% year over year — already 3.55× its full-year 2025 revenue of $4.585 billion.
Over the same period, operating loss was $0.693 billion, sharply narrower than the prior-year period’s $2.14 billion; operating cash flow reached $3.929 billion, after an outflow of $1.513 billion a year earlier.
Then the headline loss. H1 net loss was about $80.7 billion, of which about $72.4 billion came mainly from fair-value changes on convertible notes and similar financial instruments.
As the company’s valuation rose, that accounting book loss widened — it should not be read as $72.4 billion of cash burned in day-to-day operations.
Q2’s operating shift is clearer: revenue $11.6 billion, operating profit about $1.0 billion, adjusted operating margin 19%.
As of September 30, customers contributing more than $100,000 in the trailing twelve months numbered over 11,900; over $1 million, more than 1,900; over $10 million, ~250. Net revenue retention was about 900%.
The ideal AI-application business is one customers cannot leave — and whose bills keep growing. Anthropic has written that story into the financials.
One more set of easy-to-miss figures: among high-value customers, more than half already use both the Claude platform and at least one first-party application; at end-2025 that share was about 30%.
Model vendors are no longer only selling API — they are entering customer workflows directly through products such as Claude Code and Cowork.
That is a double-edged sword for application companies on this ranking. Stronger models raise their product ceiling; model vendors are also moving one layer down, competing for users, entry points, and budgets.
Finally, the compute bill that cannot be skipped. Anthropic’s long-term compute and infrastructure commitments with about eight partners exceed $600 billion — many of them non-cancellable, or payable regardless of actual usage.
It is like an airline buying out large blocks of seats years ahead. When traffic surges, scale looks beautiful; when traffic undershoots, empty seats still get billed.
So Anthropic’s risk cannot be read only from the $80.7 billion net loss, nor only from cash flow turning positive. What has to be watched is whether future revenue can keep covering the compute contracts already signed.
We cross-matched the valuation Top 100 with the earlier revenue Top 100 and found 44 overlapping companies.
Estimated combined revenue of the revenue-ranking Top 100 is $148.711 billion; Anthropic and OpenAI together are $117.0 billion — 78.7%. Valuation concentrates at the head; revenue concentrates at the head too — and revenue concentration is higher.
Hit the brakes first: the revenue ranking mixes actual annual revenue, ARR, annualized run rates, CARR, and third-party estimates, with inconsistent disclosure dates. The “valuation / revenue” ratios below are only a ruler for observing how much price has been earned by revenue — not a standard price-to-sales multiple.
Anthropic’s operating analysis should also prefer S-1 actual revenue over the earlier $67.0 billion annualized estimate.
Even with those limits, price layers across companies are still clear.
DeepSeek and Moonshot, on the two rankings’ figures, land at about 85.7× and 50×. They carry expectations for next-generation platforms, entry points, and ecosystems — current revenue can explain only part of the valuation.
Coding and Agents show clearer revenue realization. Cognition at $48.0 billion valuation against a ranking revenue estimate of $1.0 billion is 48×; Lovable about 22.2×.
Infrastructure and data tools sit closer to the ground: Fireworks AI about 17.5×, Modal about 15.8×, Mercor about 10×.
But a low multiple is not the same as cheap. Mercor’s run rate still does not deduct contractor and similar costs; a high multiple is not automatically a bubble. What matters next is growth rate, gross margin, retention, customer concentration, and how compute is procured.
One counter-intuitive conclusion: high AI valuations do not mean every company is far from revenue — and having revenue does not mean valuation is already safe.
Temporarily set Anthropic and OpenAI aside and the remaining 98 companies sum to $1,089.45 billion. The real second tier concentrates mainly in four tracks: AI coding & Agents, robotics & autonomy, chips & compute, and data & developer tools.
Those four categories together are about $592.0 billion — 54% of residual valuation.
Coding & Agents is the most expensive application line. Counting newly added Replit, 16 companies total $160.0 billion. Cognition at $48.0 billion ranks sixth overall; Harvey, Sierra, and Lovable have all crossed $10 billion.
Product competition has already shifted from “help the user generate a piece of content” to “complete a job for the user.” What has to be proven next is accuracy, delivered outcomes, and sustained customer payment — not user counts alone.
Robotics & autonomy is the most numerous track, and the one most dependent on delivery. 21 companies total $217.036 billion — the same headcount as foundation models. Anduril and Figure together are $100.0 billion — 46% of the track.
Defense autonomy, humanoid robots, and robotics foundation models cannot be measured with the same ruler — but they all answer the same question: when does a tech demo become scaled deployment and ongoing collections.
Chips, compute, data & developer tools sit closest to model demand. The two groups together are about $215.0 billion. VAST Data, Scale AI, Etched, Fireworks AI, Modal, and Baseten have all entered the $10 billion club.
Rising model call volume and longer Agent task chains will keep pushing compute and data demand higher; utilization, customer concentration, and long-term procurement contracts equally determine the profit quality of these “shovel sellers.”
Foundation models themselves are not evenly prosperous. 21 companies total about $2.15 trillion — 73.9% of the full ranking. Strip out Anthropic and OpenAI and the other 19 have only $331.865 billion.
It is not that the entire foundation-model track sits at one valuation height — the dual giants pull the track average into another order of magnitude.
China has 12 companies on the valuation ranking, totaling $165.31 billion. DeepSeek and Moonshot together are $135.71 billion — 82.1% of Chinese companies’ combined valuation.
The middle layer is the interesting part.
Across the full ranking, 29 companies have valuations of at least $10 billion but under $50 billion — China has zero. Moonshot sits exactly at $50.0 billion, outside that band. Below DeepSeek and Moonshot, the other 10 Chinese companies are all at or under $4.0 billion. The valuation structure looks like a dumbbell: heavy at both ends, broken in the middle.
Of the remainder, six are robotics firms — Galaxy General (银河通用), Qianxun Intelligence (千寻智能), Xinghai Tu (星海图), Zhiyuan Robot (智元机器人), Zhi Squared (智平方), and Independent Variable Robot (自变量机器人) — totaling about $17.29 billion.
Manus and Genspark represent Chinese teams’ exploration of Agents and globalized applications; Baichuan Intelligence and ModelBest (面壁智能) remain in the foundation-model camp.
Several of these are listed at the lower bound of “over RMB 20 billion,” converting to the same $2.857 billion valuation. That does not mean their true values are identical — it more often reflects primary-market financing clustering at that tier. They also sit close to the $2.5 billion ranking floor; one new round can reshuffle ranks.
The revenue ranking offers another angle. Eight Chinese companies have estimated combined revenue of $3.32 billion — 2.2% of that ranking. Besides DeepSeek and Moonshot sit Kling AI, Yanyu Technology (演语科技), PLAUD, Kuaizi Tech (筷子科技), Haiyi AI (海艺AI), and Ruiqi Software — concentrated in video/image, content production, Agents, and AI hardware.
The two rankings’ inclusion scopes differ; you cannot simply divide one by the other for efficiency. But the structure is clear enough: Chinese AI does not lack single-point stars or early products. The next question that matters more is whether more companies can grow into the $10–50 billion valuation band with verifiable revenue — though some of that middle layer may also be absorbed by large platforms.
One more note: already-listed names such as Zhipu and MiniMax fall outside this ranking’s scope. Part of the thin middle layer also comes from the “unlisted” caliber itself.
The valuation anchor is shifting from “financing price” to “financial multiples.” In the past the first questions were who invested and at what post-money. After Anthropic’s public S-1, the questions become how revenue is recognized, how operating cash flow behaves, whether customers are expanding, and how heavy long-term contracts are. Public markets will not give the industry one uniform answer — they will unpack pricing logic company by company.
The second-tier battlefield is applications and infrastructure. Strip out the top two and coding Agents, robotics, chips & compute, and data tools take 54% of residual valuation. The next wave of $10B companies will most likely still emerge from places closest to work delivery and model demand.
For Chinese companies, the critical variable is the middle layer. Two frontier model firms already command high prices; a cluster of robotics companies sits near $3 billion. Who can climb from that tier above $10 billion is more worth watching than how many more goalkeeper-scale names appear in the short term.
Growth is yielding to growth quality. Revenue growth still matters, but markets will keep asking about retention, gross margin, customer concentration, how much revenue each dollar of compute returns, and whether non-cancellable contracts can flex with demand. Fast growth with a runaway cost structure, and fast growth with improving operating efficiency, will not forever enjoy the same valuation.
This ranking will keep “sliming down.” PitchBook and NVCA data show that in the first three quarters of 2026, AI accounted for 82.7% of U.S. venture deal dollars — yet only 18 companies went public in Q3. A large stock of book valuations is still waiting for an exit.
As Anthropic and other frontier names list, the unlisted ranking will keep seeing companies “graduate.” We need to track both still-private and newly public companies to see the full migration of AI value.
Finally, the boundaries have to be stated clearly.
Scope first. This edition applies a strict AI-native definition: the company’s founding starting point and core delivery must be AI models, applications, or AI-specific technology.
Already-listed entities, and entities that are no longer independent after acquisition, are out. AI businesses inside traditional software companies are not carved out for a separate valuation either. Manus and Genspark are attributed to China as Chinese-team go-global projects — that is not the same as registered domicile or HQ location.
Then valuation caliber. The ranking mixes pre-money, post-money, secondary equity trades, and media-reported pricing; “over” is shown at the lower bound; RMB is converted at 1 USD = 7 CNY. Equal listed values do not mean equal true value; media-reported valuations are not company-confirmed figures.
Price freshness has to be kept in view too. 29 companies’ valuation events fall after September 1, totaling $384.9 billion — 13.2% of the ranking; another 20 still sit on pre-2026 marks.
Anthropic’s $965.0 billion comes from May financing; OpenAI’s $852.0 billion from March financing. A valuation ranking can compare price scales — it cannot pretend every price was struck today.
The revenue ranking is the same: actual revenue, ARR, annualized run rates, and third-party estimates cannot be treated as same-period, same-caliber revenue. All multiples in this essay are for structural observation only — they are not standard price-to-sales ratios or investment advice.
One hundred companies. $2.91 trillion. This ranking does not prove what every AI company is “worth.” It shows capital is already pricing a handful of platform companies and everyone else under two different systems.
If Anthropic lists successfully, it will graduate from the unlisted ranking. Headline total valuation would then fall from $2,906.45 billion to $1,941.45 billion; OpenAI would rise to No. 1 at $852.0 billion — 43.9% of remaining total; DeepSeek would rise to No. 2, but the gap between No. 1 and No. 2 would still be about 9.9×.
One company leaves, the ranking shrinks by a third — and head concentration does not disappear.
Cutoff date is October 27. We will keep updating. The complete ranking for this edition is maintained in Unique Research’s valuation database and will continue to be refreshed; the original Chinese release points to the full table.
Source: Unique Research WeChat essay, October 27, 2026. Mid 2247523152 · idx 1 · sn 268c0ea2c5f822ded7775924bfb63cf6.