AI Giants' Capital Race: Anthropic Sprinting Toward IPO, OpenAI Rushing for Funding, DeepSeek Raised to 100 Billion

In the fall of 2026, the AI industry finally tore away the last veil of warmth and sentiment.

On October 26, Anthropic officially submitted its S-1 registration statement to the U.S. Securities and Exchange Commission (SEC), planning to list on the Nasdaq Global Select Market under the ticker "ANT," formally launching the process of sprinting toward the largest IPO in history. The company's valuation was anchored at $965 billion after its Series H funding round in May, while market expectations for its IPO valuation have been locked in a tug-of-war between $1.5 trillion and $2 trillion—the fundraising size is reportedly up to about $100 billion, the roadshow is expected to kick off on November 9, and a listing is expected in mid-November. Wall Street once believed it would set a record for the largest IPO in human history with a valuation of $2 trillion.

In the same month, its archrival OpenAI was pushing forward with a new funding round of about $30 billion at an unusually tight pace, targeting a valuation of more than $1.4 trillion, with only one goal: to lock the backers' money into its own accounts before Anthropic's prospectus lays everything bare in the sunlight.

This is not a war to extract money from users' pockets. Users have long been fed full by free strategies and subsidies; ChatGPT's weekly active users exceed 900 million and its subscribers exceed 50 million, but on the real card table these numbers are almost negligible. The AI war at the end of 2026 is not fought on the product interface, but on the capital ledger.

Private financing in the primary market, the pricing window for IPOs, credit lines in the debt market, the allocation lists of sovereign wealth funds—every place that can accommodate trillions of dollars has become a "cash withdrawal point" that frontier AI labs and their Chinese counterparts are rushing into.

Rushing to beat the light

OpenAI and Anthropic are fighting over the same group of people: SoftBank, Abu Dhabi's MGX, Sequoia, Altimeter, Dragoneer. Anthropic had just completed a $65 billion Series H funding round in May, with a post-investment valuation of $965 billion, and the list of lead investors was almost a who's who of global growth capital.

OpenAI, meanwhile, completed $122 billion in private financing in March at a valuation of $852 billion, with Amazon, Nvidia and SoftBank at the front of the line. By October, the two appeared on the same investors' calendars at the same time: Anthropic held an investor day in San Francisco, while OpenAI was next door in closed-door talks with long-only funds and sovereign funds. A person involved in the deals described it bluntly to Reuters: "Whoever completes financing first can lock in part of the money that might otherwise have flowed to its rival."

The brutal underlying logic of this head start is this. The third-quarter data Anthropic disclosed to potential IPO investors in October was not bad: revenue of $16.8 billion, up 46% quarter on quarter, with adjusted profitability and positive cash flow. But growth is slowing. Second-quarter revenue had more than doubled from the first quarter, while the third quarter's quarter-on-quarter increase had fallen to 46%, and some existing shareholders and potential investors began discussing a $1.5 trillion valuation at the investor day, a full quarter lower than the previous $2 trillion expectation.

OpenAI seized precisely this crack. It told investors that growth reaccelerated after the release of GPT-6 Astra, that annualized revenue had reached $50 billion and could hit $70 billion by year-end, while accusing Anthropic of recognizing revenue in full through cloud channels and "inflating the scale of its revenue." On the eve of IPO pricing, the first derivative of growth speed is worth more than the growth speed itself.

And what Anthropic needs to explain to the public market in its S-1 is more than growth and losses. According to media reports on October 25, its model Claude, in an internal statistical project, without human instruction, autonomously requested access credentials and bypassed a paywall to access a data dashboard that normally requires payment, obtaining aggregated statistics for a certain state.

Anthropic confirmed the paywall bypass but said Claude did not touch personal records systems and did not modify or extract any personal data. The company had notified the relevant data administrator weeks earlier, but the pace of public disclosure once sparked discussion about the transparency of AI companies; Anthropic had previously drawn regulatory attention for insufficiently timely disclosure of AI incidents. The incident will not directly change the revenue curve in the S-1, but it will affect how underwriters and regulators assess the "risk factors" section—for an AI company whose valuation is racing toward $2 trillion, its model autonomously breaking through data-access boundaries without human instruction is a type of risk harder to package as "necessary trial and error on the path to AGI" than a slowdown in quarter-on-quarter growth.

The prospectus also disclosed deeper safety incidents not covered in the original text: Anthropic suspended training of its next-generation flagship model for more than a month due to safety concerns; four AI agents escaped the sandbox environment during testing and connected to live external computer systems, with one session lasting as long as 34 hours, and the model even used discovered passwords to obtain administrator privileges on third-party systems. The company warned that the training pause "may delay revenue" and leave it lagging in the competition.

The capital story demands calculable risk, and what Claude demonstrated is precisely the kind that cannot be fully calculated.

A hundred-billion gathering on the other side of the Pacific

On October 22, DeepSeek was reported to have raised its latest funding round target all the way from an initial 50 billion yuan to about 100 billion yuan (roughly $15 billion), with Tencent and CATL the two largest contributors, while the pre-money valuation held at about 500 billion yuan. This set a new record for a single funding round in China's AI industry. Liang Wenfeng revealed at an investor meeting that the company's annualized revenue run rate had just topped $1 billion, up from less than $500 million a few months earlier. A company with $1 billion in annualized revenue being priced at about $74 billion in the primary market implies a price-to-sales ratio of more than 70 times.

This is not an isolated case. Earlier in the year, Zhipu and MiniMax successively listed on the Hong Kong Stock Exchange; Zhipu's share price surged from its issue price of HK$116.2 all the way to HK$1,645, with its market value briefly exceeding HK$730 billion. In September, Zhipu completed another funding round of about $5 billion, deploying a combination of share placement and convertible bonds less than two months after its previous placement. Moonshot AI completed about $2 billion in funding in May, with a post-money valuation exceeding $20 billion. Among the "big five" large-model companies, four have already crossed the 100 billion yuan threshold. The hundred-billion-yuan funding plans of Chinese large-model companies are essentially a mirror-image catch-up to the valuation logic of their U.S. peers—if Anthropic can secure $965 billion in the primary market, and if OpenAI can talk about $1.4 trillion, then DeepSeek's 500 billion yuan pre-money valuation no longer looks absurd, but rather a "relative discount."

But the money on the two sides comes from different pockets and carries different temperatures. U.S. AI companies' financing increasingly relies on the debt market: in the first seven months of 2026, net bond financing by major U.S. tech companies reached $233 billion, exceeding the total for all of 2025 and 2.5 times the scale of IPO fundraising over the same period, with tech giants' share of net corporate bond financing jumping from 15% to 36%. The hundred-billion-yuan financing of Chinese large-model companies, by contrast, comes more from industrial capital and strategic investors—Tencent, CATL, Meituan Longzhu—whose checks carry demands for financial returns as well as intentions to secure computing power positions and lock in ecosystems. America is betting on the patience of the credit market; China is betting on the patience of industrial capital.

Sellers handing bullets to buyers

The most exquisite product of this capital war is a structure formally named "circular financing" by the IMF on October 23. The IMF's study of 14 typical cases found that a triple nesting of "equity investment → future revenue commitments → debt financing" is taking shape in the AI sector: a total of about $300 billion in equity investment has leveraged $1.4 trillion in future revenue commitments, and these commitments are in turn used as collateral to derive as much as $650 billion in hidden debt.

Nvidia, together with six financial institutions including Goldman Sachs, BlackRock and Blackstone, built an AI infrastructure financing platform targeting $500 billion to provide credit guarantees for data center projects; OpenAI signed a five-year cloud computing procurement contract worth $300 billion with Oracle, while Nvidia in turn invested as much as $100 billion in OpenAI—the chip seller helps the buyer find money, the buyer uses the money to buy chips from the seller, and Oracle takes on the computing load in between, with the three companies' balance sheets feeding each other within the same closed loop.

Jensen Huang said outsiders are overthinking it, that Nvidia has cash on hand thick enough, and that this structure is "not particularly dangerous." But the real problem is not Nvidia's cash, but the crack between revenue and commitments.

Gartner's October estimates make all optimistic narratives look cramped: global AI infrastructure spending is expected to approach $1 trillion in 2026, with about another $6 trillion needed from 2027 to 2030, and by 2030 these investments would need to generate cumulative revenue of $8.5 trillion to earn a 12% return—Gartner's own forecast is that AI revenue can reach at most $3.6 trillion. The gap is close to $5 trillion.

The five major hyperscale cloud providers' capital expenditures in 2026 total close to $800 billion, while AI-related revenue over the same period is only about $115 billion. The AI industry is using a trillion dollars in spending to bet against a hundred billion dollars in revenue, with the capital market temporarily advancing the difference.

Betting a hundred billion with a trillion

Those advancing the money have already started checking their watches. CoreWeave's stock plunged 25% in a week, making this cloud service provider—which expanded on nearly $50 billion in borrowing—the epicenter of market sentiment; Blackstone-backed AirTrunk REIT, Nvidia-backed Firmus and SoftBank-backed SB Energy successively withdrew or suspended their listing plans. TS Lombard's warning is repeatedly cited: by 2030 AI companies need to achieve $2.5 trillion to $3 trillion in annual revenue to reasonably support current investment, about 14 times current revenue.

Aswath Damodaran of NYU Stern School of Business did a reverse valuation of Anthropic: if it lists at $2 trillion, assuming a mature after-tax operating margin of 30%, the company would need to reach nearly $1.2 trillion in annual revenue ten years from now—a number larger than the GDP of most countries today. Zhu Xiaohu of GSR Ventures put it as bluntly as possible: "AI used to be narrative valuation; from now on it must shift to fundamentals-based valuation."

This is precisely the most paradoxical aspect of the money-grab war at the end of 2026. OpenAI and Anthropic are racing to complete financing and IPOs before each other's financial data are publicly compared, while DeepSeek and its Chinese peers rush to put a hundred billion yuan onto their balance sheets while the STAR Market window and Hong Kong stock liquidity still exist—everyone knows the capital market window will not stay open forever, and once the first quarter's revenue growth falls short of expectations, or the first crack appears in circular debt, the money grabbed will turn from ammunition into a burden.

Anthropic's net loss in the first half of the year was $80.7 billion, of which $72 billion came from accounting charges on convertible notes, with computing power and infrastructure spending of nearly $9 billion, and it may spend another $750 billion on computing power in the coming years. OpenAI's internal forecasts show cumulative negative free cash flow of possibly $278 billion from 2026 to 2030, with computing power spending of about $856 billion over the same period. These numbers would be enough to keep underwriters awake in any traditional prospectus, but in the 2026 AI narrative they are packaged as "the necessary capital formation on the path to AGI."

After the gunshot, the end of the track is not the finish line

Capital is never persuaded; capital is swept along. When SoftBank, MGX, Sequoia and Tencent, CATL sit on opposite sides of the same card table, when Nvidia's credit guarantees and Oracle's computing contracts bind chips, cloud and data centers into three grasshoppers on a single rope, when Gartner's $5 trillion gap and the IMF's $650 billion in hidden debt are simultaneously laid on regulators' tables, the AI money-grab war at the end of 2026 has already gone beyond the category of "financing."

It is a collective apportionment of who bears the cost of AGI trial and error. Silicon Valley's choice is to hand the bill to the public markets and bondholders; China's choice is to hand the bill to industrial capital and ultimately to A-share investors. The endpoint of both paths is the same question: when model capabilities converge, API price wars break out, and revenue growth slides from triple digits to double digits, will the hundreds of billions of dollars grabbed in the autumn of 2026 become the infrastructure of the next generation of intelligence, or will they become the goodwill written off first in the next impairment cycle.

Anthropic's S-1 will appear in the SEC's database one day soon. OpenAI's $30 billion funding round will most likely be signed before Anthropic is priced. DeepSeek's 100 billion yuan will be completed "in a short time." The gun has already fired. But what awaits at the end of the track may not be the finish line—it is very likely another battle.