Article | Lin Wanwan

On March 31, 2026, OpenAI announced the completion of $122 billion in financing, with a valuation of $852 billion, the largest private financing in human business history.

Amazon invested $50 billion in OpenAI. Of this, $15 billion was received immediately, while the remaining $35 billion will be paid out once a condition is met.

This condition is that OpenAI completes an IPO or achieves AGI.

One is going public, and the other is creating artificial general intelligence that surpasses human capabilities. The largest e-commerce company on Earth has staked an amount greater than the annual military budgets of most countries on an "or."

Let's break down the structure of all financing for OpenAI.

NVIDIA contributed $30 billion, and OpenAI happens to be one of NVIDIA's largest GPU customers.

OpenAI's CFO Sarah Friar has stated that most of the money will return to NVIDIA.

With Amazon's investment of $50 billion, OpenAI runs its models on AWS for inference, increasing AWS revenue and improving Amazon's financial reports. Microsoft has cumulatively invested over $13 billion, and OpenAI has committed to purchasing $250 billion worth of cloud services on Azure.

The money has cycled through a closed loop and returned. Wall Street calls this circular financing.

Bernstein analyst Stacy Rasgon mentioned that each of these transactions deepens market concerns about circular financing. Statistics from the CFA Institute suggest a disturbing total of nearly $1 trillion in various mutual investments and procurement commitments in the AI sector.

However, the topic of circular financing has been discussed for a whole year, and everything that needed to be said has been said.

The focus of this $122 billion financing is not on how the funds are circulating but on a more direct question: what exactly are these funds buying?

What is being bought with $852 billion

The answer is time. More accurately, it’s about buying time before the IPO.

OpenAI currently has a monthly revenue of $2 billion, resulting in an annualized revenue of about $24 billion. The $852 billion valuation corresponds to an approximate 35 times sales ratio. This multiplier indicates that the market is paying for OpenAI three to four years down the road.

Let's find some reference points to feel this out. NVIDIA, in the midst of making profits, has a price-to-sales ratio of around 20. Snowflake peaked at 100 times but quickly fell back below 30. Salesforce was about 10 times at its IPO.

A 35 times ratio for a company that is still losing money is already quite aggressive.

OpenAI's own plan projects revenues of $100 billion and profits of $14 billion by 2029. To grow from $24 billion to $100 billion, a compound annual growth rate of over 40% must be maintained for four consecutive years. I seriously considered the history of software companies maintaining this growth rate on a base of $10 billion in revenue and could not find even one.

The $852 billion valuation can only be justified by one condition: someone must be willing to take over at that price on the public market. This means the IPO must be successful.

Once this level of understanding is reached, the entire financing structure becomes comprehensible.

Of the $50 billion from Amazon, $35 billion is contingent on the IPO; without going public, the funds will not be disbursed. SoftBank's $30 billion is delivered in three phases; the first payment is made when the financing closes, with subsequent payments in July and October, precisely timed for critical points in the IPO preparation period.

OpenAI sold $3 billion worth of shares to retail investors through banks for the first time and plans to include itself in ARK Invest's ETF. Retail investors purchasing shares and entering the ETF will provide natural foundational buying when the IPO opens.

The wording in financing announcements has also become less like reporting to private investors. "We are the fastest platform to reach 10 million users, the fastest to reach 100 million users, and will soon be the fastest to reach 1 billion weekly active users," "Our revenue growth rate is four times that of Google and Meta during the same period." This set of rhetoric can be directly included on the first page of the prospectus without modification.

PitchBook has reported that among the three largest AI IPO candidates—OpenAI, Anthropic, and Databricks—OpenAI has the lowest business quality fundamental score but the highest valuation.

Every design detail of the $122 billion financing points in the same direction: to get this company public and let the public market absorb this valuation.

Two companies compete for the same faucet

OpenAI needs an IPO, but it is not the only one that does. This is the real spectacle of 2026.

First, let’s look at the queue. CoreWeave went public last March at an issue price of $40, now at $130, with a market capitalization exceeding $46 billion, setting an example for subsequent companies. Databricks has a valuation of $134 billion during its roadshow and an annualized revenue of nearly $5 billion. Cerebras has resolved CFIUS reviews and re-submitted its IPO application.

The real heavyweights are Anthropic and OpenAI. Anthropic is valued at $380 billion and has already engaged Wilson Sonsini for IPO legal preparations. Kalshi predicts that Anthropic is 72% likely to go public before OpenAI.

This odds situation is quite critical for OpenAI. The pool of funds looking to invest in AI targets is limited; if Anthropic secures this pool of funds and attention first, OpenAI’s IPO pricing will be compressed.

Indeed, Anthropic is encroaching on OpenAI’s territory. In the enterprise API market share, OpenAI dropped from 50% in 2023 to 25% by mid-2025, while Anthropic increased from 12% to 32% in the same period. Anthropic’s revenue growth rate is about three times that of OpenAI. Analysts projecting current trends expect Anthropic to surpass OpenAI’s annualized revenue by mid-2026.

Two years ago, OpenAI dominated the enterprise market, but now Anthropic has emerged as the leader in the enterprise API market. The Claude Code product has an annualized revenue of $2.5 billion, contributing 4% of global public GitHub submissions. This speed of shift is quite rare in the tech industry.

OpenAI certainly has its own trump cards. With 900 million weekly active users and 50 million paid subscribers, its ad business has crossed $100 million in annualized revenue in just six years. The brand recognition and user habits built by ChatGPT still represent the largest moat in the AI industry. However, the slowdown in the enterprise segment is very real.

Both companies are simultaneously spending at an astonishing rate.

OpenAI expects a loss of $14 billion in 2026, and by 2027, the annualized burn rate may reach $57 billion. The $122 billion financing sounds astronomical, sustaining the company for approximately 18 to 24 months. Anthropic expects to spend $19 billion in 2026, with $12 billion on training models and $7 billion on running inferences.

Whoever goes public first will extend their life. The private market money can no longer feed these companies adequately; the public market is the last faucet that has yet to be opened. Renaissance Capital predicts that there may be between 200 and 230 IPOs in 2026, and just the four companies OpenAI, Anthropic, Databricks, and Cerebras could surpass $200 billion in IPO financing.

This is the largest tech IPO window since 2000. The last time such a scale of IPO wave occurred was also in the year 2000.

Can the speed of making money outpace the speed of spending money?

All valuations, all financing structures, and all IPO plans ultimately rely on one judgment: whether AI can make money faster than it spends money.

If successful, the $122 billion financing is visionary, and the $852 billion valuation is a bargain.

However, there are also scenarios being modeled where this does not occur. Analysts refer to it as the CapEx Cliff, where capital expenditures plunge. Billions of dollars in data centers have been built, but the software running on them fails to generate enough revenue to cover costs, and efficiency revolutions will replace scale competition, leaving those who bet everything on "bigger is better" sitting atop a pile of expensive but underutilized hardware.

Advancements in efficiency are happening faster than most people realize. Training a model equivalent to GPT-4 cost about $79 million in 2023, but by 2026, with next-generation hardware and techniques like distillation and quantization, costs have dropped to $5 million to $10 million.

Last year, DeepSeek R1 trained a near-state-of-the-art inference model for under $300,000; this January, it published a new paper on training architecture, continuing to focus on efficiency. Google’s latest Gemini 3.1 Flash-Lite has reduced the inference cost to $0.25 per million tokens. IBM researchers have publicly stated that 2026 will be a year of divergence between leading large models and efficient small models.

If the efficiency route continues to outpace the scale route, the computing power empire that OpenAI is building with the $852 billion valuation might face depreciation before it is even completed.

After the 2000 bubble burst, the internet did not disappear; Google emerged from the rubble. What died were those companies that raised the most money at the peak of the bubble, built the most infrastructure, and never found a sustainable business model.

AI will also not vanish. But whether the $122 billion and $852 billion valuation can last until profitability remains far from certain.

The drum is still beating, and the rhythm is getting faster.

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。