For $110 billion, OpenAI has turned its ideals into a business.
On the evening of February 27th, OpenAI dropped a bombshell.
It's a single - round financing of $110 billion. SoftBank invested $30 billion, NVIDIA $30 billion, and Amazon $50 billion. The pre - investment valuation was $730 billion, and the post - investment valuation is approaching $840 billion.
This is the largest single financing in the history of AI, without a doubt.
The official statement is quite decent. It aims to expand the infrastructure construction of artificial intelligence and accelerate the popularization of AI. Altman described the cooperation blueprint with the three companies on social media X. NVIDIA will provide computing power, Amazon will build the cloud, and SoftBank will serve as the capital backer.
However, if you put these three contracts together, you'll find something more thought - provoking than the numbers:
AGI, a concept once regarded as the ultimate human ideal, has been completely transformed into a hedging tool on the financial statements and has become a business.
Moreover, it is given completely different functions in different contracts.
1. The Three Faces of AGI
This story dates back to 2019.
That year, Microsoft made its first investment of $1 billion in OpenAI, and the cumulative investment has since exceeded $13 billion. There was a clause in the contract, later known as the "AGI clause". Once OpenAI announces the achievement of AGI, Microsoft will lose access to future models.
At that time, it seemed like an idealistic safety valve. OpenAI's original intention was to ensure that AGI benefits all of humanity and prevent any single company from monopolizing it.
It seemed quite romantic.
But as time has passed to 2025, romance is dead, and contracts reign supreme.
According to the leaked documents obtained by The Information, there is actually a secret agreement between OpenAI and Microsoft: The trigger condition for AGI is that OpenAI develops an AI system that can generate at least $10 billion in profit.
It's not a scientific consensus but a financial formula.
The new agreement in October 2025 introduced an independent expert verification mechanism, but the core logic remains the same. In Microsoft's contract, AGI is a termination switch. Its function is to change the distribution of rights and interests between the two companies once triggered.
Now, let's turn our attention to the protagonist of this round of financing: Amazon.
Amazon's $50 billion investment is divided into two parts: $15 billion is paid upfront, and the remaining $35 billion will be paid after "certain conditions are met".
According to information obtained by foreign media, these conditions include OpenAI going public before the end of the year or reaching a certain AGI milestone.
See? The same term, AGI, appears in the second contract, but its function is completely reversed.
For Microsoft, AGI is a sword hanging over its head. Once announced, Microsoft's rights will be restricted. For Amazon, AGI is a $35 - billion check. Once announced, the money will be in the account.
Announce AGI → Trigger the arrival of Amazon's $35 billion → But at the same time, trigger the restriction of Microsoft's rights
Don't announce AGI → Microsoft's revenue sharing continues → But Amazon won't get the $35 billion
OpenAI is caught between the two contracts. Altman has to walk a tightrope between two completely opposite incentives.
Of course, there is a third way: IPO.
If it goes public before the end of the year, it can also trigger Amazon's remaining payment without touching the AGI "bomb". However, this means that the public market will set its own price on "whether you have achieved AGI or not". By then, this issue will no longer be just a private game between two companies.
If AGI is really achieved, that day may not be a celebration but a large - scale legal and financial trial day.
As for the third face of AGI, it is still on OpenAI's official website: Our mission is to ensure that artificial general intelligence benefits all of humanity.
When the mission of benefiting all of humanity and the clause triggering $35 billion are associated with the same term, does this term belong to science, capital, or law?
2. AI Is No Longer Just Software; It Has Become Heavy Industry
Another strange thing about this round of financing is hidden in two words in the contract: Gigawatt (GW) and chips.
In the past, if you wanted to start an internet company, you could just rent a server and get started.
But now, OpenAI is more like building a nuclear power plant.
The contract clearly states that OpenAI has to promise Amazon to use 2 GW of computing power. What does 2 GW mean? It's the total power generation of two medium - sized nuclear reactors.
This computing power will run on Amazon's self - developed Trainium chips for training cutting - edge models.
The same goes for NVIDIA. OpenAI promised to deploy 5 GW of computing resources on NVIDIA's Vera Rubin system. This is the electricity consumption of about 5 million households, a digital black hole never seen in human history.
NVIDIA invested $30 billion, and OpenAI promised to use this money to buy NVIDIA's chips. Amazon invested $50 billion, and OpenAI promised to spend $100 billion on AWS in the next eight years.
This logic is equivalent to the supplier investing money in the customer, and the customer then using this money to buy the supplier's chips and cloud services. For the supplier, this money changes from investment expenditure to operating income on the financial statements. The stock price goes up, the inventory is cleared, and the customer is locked in.
For OpenAI, there is an additional amount of cash on the balance sheet, but a large part of it has already been earmarked for pre - determined consumption.
Some on Wall Street have started calling it circular financing. Reuters' analysis directly warns that this model of mutual corporate investment and signing supply transactions may artificially inflate demand and revenue.
Altman's response is: I understand everyone's concerns. All of this makes sense only when new revenue flows into the entire artificial intelligence ecosystem.
That's what he says, but the numbers don't lie. OpenAI's revenue in 2025 was $13.1 billion, but the loss reached $8 billion; the loss is expected to soar to $25 billion in 2026, with a burn rate as high as 83.3%. Internal forecasts suggest that it won't achieve profitability until the 2030s.
According to the financial documents submitted by OpenAI to its shareholders, the revenue in the first half of 2025 was $4.3 billion, but the net loss reached $13.5 billion. Internal documents show that the expected loss in 2026 is about $14 - 17 billion, and the cumulative cash consumption will reach $143 billion by 2029.
The valuation of $730 billion corresponds to continuously expanding losses. This valuation is not pricing today's performance but the future AGI.
When AGI will come, in what form, and who will get the profits after it arrives, all these are written into the contract and have become terms that can be traded, gamed, and arbitraged.
More importantly, this round of financing has completely changed the industry nature of AI.
Five years ago, when people talked about AI, they talked about the number of model parameters, the scale of training data, and the inference speed. Today, OpenAI's contracts talk about how many gigawatts of computing power and how much electricity consumption equivalent to that of nuclear power plants.
AI is no longer just software; it has become heavy industry. Its core resources are no longer algorithm engineers but chips, electricity, data centers, and cooling systems.
This is no longer a code competition; it's an infrastructure competition.
3. Microsoft's Embarrassment: Exclusive, but Not the Only One
There is another detail worth pondering in this round of financing: Microsoft.
As OpenAI's oldest investor and exclusive cloud service provider, Microsoft did not participate in this round of financing. The two sides issued a joint statement, strongly maintaining their core positions:
● Azure remains the exclusive cloud service provider for OpenAI's stateless API
● The revenue - sharing mechanism remains unchanged, covering the new revenue generated from OpenAI's cooperation with Amazon
● Microsoft continues to have the exclusive intellectual property license for OpenAI's models and products
But anyone with a discerning eye can see that this "marriage" has cracked. The word "exclusive" is being quietly diluted.
OpenAI's enterprise - level platform, Frontier, will also run on AWS. OpenAI will massively adopt Amazon's self - developed Trainium chips. It will spend $100 billion on AWS in the next eight years.
Microsoft has retained the API call line but has lost its status as the only cloud service provider. OpenAI is consciously reducing its dependence on a single cloud provider.
The reason is simple.
As the scale of model training approaches the computing power threshold required for AGI, a single supplier cannot meet all the needs. OpenAI needs more flexible computing power supply and more diversified sales channels. AWS's enterprise customer pool and sales team happen to fill the gap that Microsoft cannot cover.
Microsoft's role has changed from an exclusive partner to one of the core partners. This is a subtle but profound transformation.
4. Pursued from Behind and Constrained by Clauses
Looking at OpenAI's this round of financing in the context of the competitive landscape will make it clearer.
Two weeks ago, Anthropic just completed a $30 - billion Series G financing, with a post - investment valuation of $380 billion. SoftBank and NVIDIA are also investors. According to public data, Anthropic's annualized revenue has reached $14 billion, and it has a considerable market share in the programming field.
Google's Gemini once surpassed ChatGPT in several key benchmark tests, triggering a red alert within OpenAI.
ChatGPT's market share is declining. According to SimilarWeb data, it dropped from 86.7% at the beginning of 2025 to 64.5% at the beginning of 2026. Although the weekly active users have exceeded 900 million, the competition is no longer a one - sided game.
Meanwhile, OpenAI's "Stargate Project" has stalled. This project, jointly invested $50 billion by OpenAI, Oracle, and SoftBank, has not recruited formal employees and has not substantially built a data center in over a year due to differences in control rights and site ownership among the three parties. The original goal of signing a contract for 10 GW of computing power by the end of 2025 was only 6.8 GW.
The computing power dilemma is becoming an invisible shortcoming for OpenAI. It can only purchase temporary computing power from AWS and Azure, getting stuck in a cycle of relying on the outside - high cost - unable to build its own.
And the $110 - billion financing is to break this cycle.
[Beyond the Page] Words:
OpenAI has received $110 billion.
This is not an ordinary financing. It is a manifestation of the financialization of AGI, a milestone in the heavy - industrialization of AI, and a charge in the arms race.
But what's most thought - provoking is not the numbers themselves but a deeper question.
When AGI has changed from a scientific vision to a contract clause, from a research goal to an IPO trigger, and from a mission to benefit all of humanity to a switch for a $35 - billion check, how much of its original meaning is left?
The tech media Newcomer had a sharp comment on this round of financing:
Convince enough CEOs and rich people that you will change the world and get them to write huge checks.
And this may be the most realistic portrayal of the AI industry in 2026.
The competition of large models is no longer a technological one but a capital one. AGI is no longer a scientific proposition but a financial one. AI is no longer a light - asset software but a heavy industry that needs a nuclear power plant to drive.
Ideals are dead, and checks reign.
This article is from the WeChat official account "Beyond the Page", author: Huahua. Republished by 36Kr with authorization.