OpenAI completes the largest funding round in history; Anthropic is more eager to go public than it is.

Author: Lin Wanyan

On March 31, 2026, OpenAI announced that it had completed a $122 billion fundraising round, valuing the company at $852 billion— the largest private funding deal in human business history.

Amazon invested $50 billion in OpenAI. Of that, $15 billion was deposited immediately, while the remaining $35 billion will be paid only after a certain condition is met.

That condition is that OpenAI completes an IPO, or achieves AGI.

One path is to go public. The other is to build a form of general intelligence that surpasses humans. Earth’s largest e-commerce company is putting a pile of money—larger than the annual military budgets of most countries—on a single “or.”

Let’s break down all of OpenAI’s funding rounds and look at the structure.

NVIDIA put in $30 billion, and OpenAI just happens to be one of NVIDIA’s biggest GPU customers.

OpenAI’s CFO, Sarah Friar, has said herself that most of the money will end up back with NVIDIA.

Amazon’s $50 billion investment goes in, OpenAI runs the models on AWS for inference, AWS revenue rises, and Amazon’s financial reports look better. Microsoft has cumulatively invested more than $13 billion, and OpenAI has committed to purchasing $250 billion in cloud services on Azure.

The money went around in a closed loop and came back. Wall Street calls this circular financing.

Bernstein analyst Stacy Rasgon says every one of these deals deepens the market’s concerns about circular financing. The CFA Institute’s data is even more unsettling: in the AI sector, the total commitments to invest in and purchase from each other have already approached $1 trillion.

But the topic of circular financing has been discussed for a whole year. Everything that could be said has already been said.

What truly deserves attention about this $122 billion round isn’t how the money circulates. It’s a more direct question: what exactly is all this money buying?

What $852 billion is buying

The answer is: time. More precisely, the time until before the IPO.

OpenAI’s current monthly revenue is $2 billion, or about $24 billion on an annualized basis. A $852 billion valuation corresponds to roughly 35x sales. This multiple means the market is effectively paying for OpenAI’s performance three to four years from now.

Let’s use a few reference points to feel it. Under the conditions of NVIDIA’s frenzy of profits, its PS is around 20x. At its peak, Snowflake reached about 100x, but soon dropped back below 30x. When Salesforce went public, it was roughly 10x.

Putting 35x on a company that’s still losing money is already aggressive.

OpenAI’s own plan is to reach $100 billion in revenue and $14 billion in profit by 2029. From $24 billion to $100 billion, it would require maintaining annual compounded growth of over 40% continuously for four years. I seriously thought about it—looking at software companies that sustained that growth rate on a base of more than $10 billion in revenue—and I couldn’t find a single example in history.

For a $852 billion valuation to hold, there’s only one condition: someone is willing to take over the company in the public markets at that price. In other words, the IPO must succeed.

Once you get past that layer, the entire financing structure starts to make sense.

Of Amazon’s $50 billion, $35 billion is contingent on an IPO—if it doesn’t list, the money doesn’t get paid. SoftBank’s $30 billion is paid in three tranches: the first at the time the financing closes, and the next two arriving in July and October, precisely placed on key moments in the IPO preparation window.

OpenAI’s first step was selling $3 billion worth of shares to retail investors through banks, and then also getting into ARK Invest’s ETF. Retail investors buy shares and enter the ETF; when the IPO opens, that becomes a built-in base order.

The wording in the financing announcement doesn’t sound much like it’s reporting to private investors anymore. “We are the fastest platform to reach 10 million users, fastest to reach 100 million users, and soon the fastest to reach 1 billion weekly active users.” “Our revenue growth rate is four times that of Google and Meta in the same period.” This pitch can be copied directly into the first page of a prospectus without changing a thing.

PitchBook has a study indicating that among the three biggest AI IPO candidates—OpenAI, Anthropic, and Databricks—OpenAI has the lowest score on business fundamentals but the highest valuation.

Every design detail of the $122 billion round points in the same direction: get this company to go public, and let the public markets absorb that valuation.

Two companies fighting over the same faucet

OpenAI needs an IPO, but it isn’t the only one that needs it. This is the real big show of 2026.

First, look at the lineup. CoreWeave already listed last March: its offer price was $40; it’s now $130, with a market cap above $46 billion—setting a template for the companies coming next. Databricks is valued at $134 billion and was in roadshows with annualized revenue close to $5 billion. Cerebras resolved the CFIUS review and resubmitted its IPO application.

The real heavyweight names are Anthropic and OpenAI. Anthropic is valued at $380 billion and has already retained Wilson Sonsini to prepare IPO legal work. Kalshi’s prediction market estimates the probability that Anthropic will list before OpenAI at 72%.

That kind of odds is deadly for OpenAI. The pool of capital looking to buy AI assets is limited. If Anthropic takes that money and attention first, OpenAI’s IPO pricing will get compressed.

And Anthropic really is eating into OpenAI’s turf. In the enterprise API market share, OpenAI fell from 50% in 2023 to 25% by mid-2025, while Anthropic rose from 12% to 32% over the same period. Anthropic’s revenue growth is roughly three times OpenAI’s. Some analysts extrapolate from the current curve and estimate Anthropic will exceed OpenAI’s annualized revenue in mid-2026.

Two years ago, OpenAI dominated the enterprise market outright. Now Anthropic has become the leader in the enterprise API market. Claude Code’s annualized revenue alone is $250 million, contributing 4% of the world’s GitHub public code submissions. This pace of reversal is rarely seen in the tech industry.

Of course, OpenAI has its own ace. It has 900 million weekly active users, 50 million paid subscriptions, and its advertising business trial has already surpassed $100 million in annualized revenue after six years. ChatGPT’s brand recognition and user habits are still the biggest moat in the AI industry. But the slowdown on the enterprise side is very real.

Both companies are also spending money at astonishing speed.

OpenAI is expected to lose $14 billion in 2026, and by 2027 the annualized cash burn rate could be as high as $57 billion. A $122 billion fundraising round sounds astronomical, but it would sustain them for only about 18 to 24 months. Anthropic is expected to spend $19 billion in 2026: $12 billion training models, and $7 billion running inference.

Whoever goes public first gets the longer runway. Private-market money has nearly run out to keep feeding these companies; the public markets are the last faucet still not turned on. Renaissance Capital predicts there could be 200 to 230 IPOs in 2026. Just combining OpenAI, Anthropic, Databricks, and Cerebras—those four alone—could put total IPO financing at more than $200 billion.

This is the biggest tech IPO window since 2000. And the last time an IPO wave at this level appeared was also in 2000.

Can the pace of making money outstrip the pace of spending money?

All valuations, all financing structures, all IPO plans—ultimately—rest on a single judgment. Can AI’s pace of earning money outrun its pace of spending?

If it can outpace it, then the $122 billion round is foresight, and the $852 billion valuation is a discounted price.

If it can’t, there are also models being built for the scenario. Analysts call it a CapEx Cliff, a capital expenditure cliff. Thousands of billions of dollars’ worth of data centers have been built. The software running on top doesn’t earn enough to cover the costs. Then an efficiency revolution will replace the scale race. Companies that bet everything on “the bigger the better” find themselves sitting on expensive hardware with insufficient utilization.

Progress in efficiency is faster than most people realize. Training a model at roughly the same level as GPT-4 cost about $79 million in 2023. By 2026, using next-generation hardware plus techniques like distillation and quantization, costs have already dropped to between $5 million and $10 million.

Last year, DeepSeek R1 trained an inference model at near-frontier levels for under $300,000. This year, in January, it published a new paper on a training architecture, continuing to push the efficiency frontier. Google’s latest Gemini 3.1 Flash-Lite has compressed inference costs to $0.25 per million tokens. IBM researchers have said publicly that 2026 will be a year where the frontier-maximum large-model route and the efficient small-model route diverge.

If the efficiency route continues to outperform the scale route, OpenAI’s “compute empire” built with a $852 billion valuation money may still face devaluation before it’s even finished being built.

After the bubble burst in 2000, the internet didn’t disappear. Google grew out of the ruins. What died were the companies that poured the most money into mergers and built the most infrastructure at the bubble’s highest point, yet never reached a sustainable business model.

AI also won’t disappear. But whether $122 billion and a $852 billion valuation can carry them all the way to the day they reach profitability is far from as clear-cut as it looks.

The drum is still being struck, and the beats are getting faster.