SAN FRANCISCO — OpenAI has told investors that its annualized revenue reached nearly $50 billion in September, roughly $20 billion below a figure that had been widely reported across Wall Street — triggering a fresh selloff in AI stocks and exposing just how sensitive markets have become to the economics behind the artificial-intelligence boom.
The headline initially looked devastating.
OpenAI had been widely associated with an annualized revenue figure approaching:
$70 billion.
Now the company was saying the number was closer to:
$50 billion.
That immediately created a disturbing question.
Had one of the fastest-growing technology companies in history suddenly fallen:
$20 billion short?
Not exactly.
The discrepancy appears to be largely about:
how the revenue is calculated.
And that distinction matters.
Because while OpenAI’s growth remains extraordinary, the episode has exposed a much bigger problem for the AI industry:
Investors still do not have standardized financial metrics for private AI companies that are already influencing:
trillion-dollar technology valuations
hundreds of billions in data-center spending
and
enormous amounts of Wall Street financing.
OPENAI SAYS SEPTEMBER ANNUALIZED REVENUE WAS NEAR $50 BILLION
The Financial Times reported that OpenAI recently told investors its annualized revenue was approaching:
$50 billion
at the end of September.
That is still an extraordinary number.
Annualized revenue, or revenue run rate, generally takes:
one month’s revenue
and assumes that pace continues for:
12 months.
So a $50 billion annualized rate does not necessarily mean OpenAI has already collected $50 billion during 2026.
It means its most recent monthly revenue pace would translate to roughly that amount over a full year.
That distinction is essential.
THE $70 BILLION NUMBER WAS NOT NECESSARILY COMPARABLE
The previously reported:
$70 billion
figure appears to have been derived partly from attempts by investors to compare OpenAI directly with:
Anthropic.
But the companies calculate annualized revenue differently.
Anthropic includes sales generated through cloud partners such as:
Amazon Web Services
and
Google Cloud.
OpenAI’s current reported figure does not include equivalent revenue in the same way.
That makes a simple:
$50 billion versus $70 billion
comparison misleading.
THIS IS MORE OF AN ACCOUNTING PROBLEM THAN A $20 BILLION COLLAPSE
The important point is that OpenAI did not appear to lose $20 billion of customer spending overnight.
The number changed because of:
what was being counted.
That means the immediate market reaction arguably exaggerated the operational impact.
But the confusion still matters.
If investors cannot easily tell what counts as revenue, comparing the biggest AI companies becomes much harder.
And valuation becomes more uncertain.
OPENAI IS STILL GROWING AT AN EXTRAORDINARY RATE
OpenAI reportedly started 2026 with an annualized revenue run rate around:
$20 billion.
By September:
nearly $50 billion.
That implies revenue momentum has more than doubled during the year.
By almost any normal corporate standard, that would be astonishing.
But OpenAI is no longer being judged by normal standards.
It sits at the center of a global infrastructure cycle costing:
hundreds of billions of dollars.
That means investors expect extraordinary revenue because the costs are extraordinary too.
THE MARKET REACTION WAS BRUTAL
The disclosure helped trigger a sharp decline across AI-linked stocks.
Nvidia fell roughly:
3%.
AMD declined.
Broadcom dropped sharply.
Micron also fell.
The Nasdaq lost more than:
1%.
Those companies did not suddenly lose contracts with OpenAI.
Instead, investors reacted to something broader:
confidence.
If OpenAI’s underlying revenue is lower than investors thought, then the enormous capital expenditures being made on its behalf look slightly harder to justify.
NVIDIA IS EXPOSED THROUGH COMPUTE DEMAND
Nvidia’s exposure is straightforward.
OpenAI needs enormous quantities of:
GPU computing power.
That computing demand ultimately supports sales of Nvidia processors.
More OpenAI users mean:
more inference
more training
and
more GPUs.
So investors view OpenAI’s revenue growth as one indicator of whether end-user demand is keeping pace with infrastructure spending.
If monetization slows, GPU demand may eventually become vulnerable.
Not today.
But eventually.
BROADCOM IS TAKING ON AN EVEN BIGGER ROLE
Broadcom is reportedly preparing to finance tens of billions of dollars of equipment tied to custom chips being developed for OpenAI.
The scale is enormous.
Wall Street Journal reporting indicates Broadcom has been exploring more than:
$50 billion
of financing related to AI-chip infrastructure.
That is a remarkable shift.
Technology companies once largely funded hardware internally.
Now AI infrastructure is becoming so expensive that companies are increasingly turning to:
banks
private credit
and
structured financing.
ORACLE IS DOING THE SAME THING
Oracle is also exploring external financing tied to AI hardware.
The company has been expanding data centers aggressively to serve major AI customers.
But Nvidia GPUs are expensive.
So are:
power systems
land
cooling
and
networking.
As infrastructure requirements grow, even profitable technology companies increasingly need financing partners.
This is turning AI into a capital-markets story.
EVEN SPACEX IS LOOKING FOR BILLIONS
SpaceX has also reportedly explored roughly:
$40 billion
of financing tied to Nvidia chip purchases.
That illustrates just how large the computing arms race has become.
AI hardware is no longer a normal corporate capital expenditure.
It increasingly resembles:
infrastructure finance.
The industry is moving closer to the economics of:
telecommunications
energy
and
utilities
than traditional software.
THIS IS THE BIGGER ISSUE: AI IS BECOMING EXTREMELY CAPITAL INTENSIVE
Software historically offered one of the best business models in corporate America.
Build once.
Sell repeatedly.
High gross margins.
Minimal physical infrastructure.
Artificial intelligence is different.
Frontier AI requires:
GPUs
data centers
power plants
cooling infrastructure
and
networking.
Those assets cost enormous amounts of money.
That changes the economics.
MORGAN STANLEY SEES A $1.5 TRILLION FINANCING NEED
Morgan Stanley estimates that AI infrastructure could require around:
$1.5 trillion
of external financing by:
2028.
That is a staggering figure.
It suggests internal corporate cash flows alone may not be enough.
Wall Street will increasingly need to fund the buildout through:
Debt
Private credit
Asset-backed structures
and
Project financing.
This connects AI directly to interest rates.
HIGH INTEREST RATES MAKE EVERYTHING HARDER
The problem is that borrowing is expensive.
U.S. Treasury yields remain elevated.
Long-term rates have reached levels not seen for decades.
That raises financing costs across the economy.
A data center financed when rates are:
2%
looks very different from the same data center financed when rates are:
6%.
The higher the borrowing cost, the more revenue an AI company needs to generate to justify the project.
That is why OpenAI’s revenue number matters so much.
THE ENTIRE AI ECONOMY DEPENDS ON SOMEONE PAYING AT THE END
The infrastructure chain looks like this.
Nvidia sells:
GPUs.
Data-center companies buy them.
Cloud providers rent compute.
OpenAI and Anthropic use that compute.
Consumers and businesses then pay for:
ChatGPT
Claude
APIs
and
Enterprise AI services.
The money ultimately has to travel backward through that chain.
If users do not generate enough revenue at the end, the economics weaken for everyone upstream.
OPENAI IS THE MOST IMPORTANT TEST
OpenAI has hundreds of millions of users.
Its products have become mainstream.
ChatGPT is used by:
Consumers
Developers
Corporations
and
Governments.
That makes it one of the strongest candidates in the world to monetize generative AI.
So if even OpenAI struggles to produce enough revenue to justify infrastructure costs, investors would have good reason to worry about smaller players.
That is why Wall Street watches its numbers so closely.
ANTHROPIC IS GROWING EVEN FASTER BY SOME METRICS
Anthropic has also reported spectacular revenue growth.
Its annualized revenue recently exceeded:
$65 billion
under its own accounting methodology.
Some forecasts suggest it could approach:
$100 billion
by year-end.
But again, comparison is difficult.
Anthropic includes revenue flowing through cloud partners more broadly.
OpenAI does not currently calculate the same metric in exactly the same way.
That makes headline comparisons dangerous.
ANTHROPIC’S CLOUD CHANNEL IS HUGE
A significant portion of Anthropic’s business comes through partners such as:
AWS
and
Google Cloud.
Those platforms resell access to Claude.
That means Anthropic’s reported annualized revenue includes business that flows through third parties.
OpenAI historically had a somewhat different arrangement with:
Microsoft Azure.
That relationship has also evolved over time.
So investors are not comparing identical business models.
MICROSOFT CHANGED THE ECONOMICS
For years, Microsoft was deeply intertwined with OpenAI.
It invested billions.
It provided computing infrastructure.
And Azure distributed OpenAI models.
But the companies have renegotiated parts of their commercial relationship.
OpenAI has gained more freedom to work with other cloud providers.
Microsoft has also diversified its AI partnerships.
That means the accounting around OpenAI-related revenue has changed.
And that helps explain why earlier annualized figures are difficult to compare with today’s numbers.
OPENAI ALSO SHARES REVENUE WITH PARTNERS
Another complication is revenue sharing.
OpenAI has contractual obligations to commercial partners.
Historically, Microsoft received a significant percentage of OpenAI’s revenue.
Those arrangements are changing over time.
That means:
gross customer spending
is not necessarily the same as:
revenue OpenAI keeps.
For investors trying to value the business, that distinction is critical.
REVENUE IS NOT PROFIT
This is perhaps the biggest mistake investors can make.
OpenAI may be generating tens of billions in annualized revenue.
But operating frontier AI remains enormously expensive.
The company pays for:
GPU capacity
Cloud infrastructure
Power
Researchers
and
Model training.
That means high revenue does not automatically produce high profits.
A company can grow very quickly and still consume cash.
AI MODELS ARE EXPENSIVE TO SERVE
Unlike traditional software, every interaction with an AI model requires computing resources.
Ask ChatGPT a question.
GPUs process it.
Generate an image.
More compute.
Run complex reasoning.
Even more compute.
That means usage growth directly increases costs.
In software, an additional user can sometimes cost almost nothing.
In AI, an additional user may create meaningful variable expense.
That affects margins.
MODEL EFFICIENCY IS THEREFORE CRITICAL
The AI companies are racing to reduce:
cost per token.
They are using:
Better chips
Smaller models
Distillation
Caching
and
More efficient architectures.
If inference costs fall faster than prices, margins can improve dramatically.
This is one of the industry’s biggest potential advantages.
But if model complexity rises just as quickly as efficiency improves, costs remain high.
OPEN SOURCE CREATES ANOTHER PRESSURE
OpenAI also faces growing competition from:
open-source models.
Companies increasingly have the option to run models themselves.
That can reduce dependence on premium API providers.
If open-source systems become sufficiently capable, OpenAI may have to lower prices.
That could increase usage but reduce revenue per unit of compute.
The entire industry is balancing:
growth
against
pricing power.
THE ENTERPRISE MARKET IS THE BIGGEST PRIZE
Consumers helped make ChatGPT famous.
Businesses may ultimately determine OpenAI’s financial success.
Enterprise customers can spend far more money.
They may use AI for:
Coding
Customer service
Research
Sales
Finance
and
Operations.
If OpenAI becomes embedded inside corporate workflows, recurring revenue could become much more durable.
That is one reason enterprise adoption matters more than simple user counts.
AGENTS COULD CREATE ANOTHER REVENUE BOOM
The next major growth opportunity is:
AI agents.
Instead of merely answering questions, agents can perform tasks.
Book travel.
Write software.
Analyze documents.
Operate enterprise systems.
Make purchases.
If agents become reliable, companies may pay substantially more for them.
That could create a much larger market than chatbots alone.
But agentic systems also consume more compute.
Again, revenue and infrastructure costs rise together.
OPENAI MAY EVENTUALLY GO PUBLIC
The accounting confusion matters because OpenAI could eventually pursue:
an IPO.
Public-market investors would demand much more standardized disclosure.
They would want audited figures for:
Revenue
Cloud costs
Gross margin
Cash flow
and
Capital commitments.
Annualized revenue estimates would no longer be enough.
Anthropic is moving toward public-market scrutiny too.
That could force the frontier AI industry to mature financially.
TODAY’S PRIVATE-MARKET METRICS ARE TOO FLEXIBLE
Private technology companies often emphasize metrics that make growth appear strongest.
Annualized revenue.
Bookings.
Committed revenue.
Remaining performance obligations.
Contracted spend.
Each can be useful.
But they are not interchangeable.
The OpenAI episode is a warning.
A number can appear precise:
$70 billion.
But if the definition is not standardized, precision is misleading.
WALL STREET IS FINALLY ASKING HARDER QUESTIONS
For much of the AI boom, investors focused on:
growth.
Now they are increasingly focused on:
returns.
How much cash does each data center generate?
How long does a GPU remain productive?
What happens when chips become obsolete?
How much debt is involved?
What are the margins?
Those questions are becoming unavoidable.
FIRMUS PROVIDED ANOTHER WARNING THIS WEEK
The same shift appeared in Australia.
Nvidia-backed AI infrastructure company:
Firmus
scrapped a planned IPO worth more than:
$5 billion.
The company had sought a valuation near:
$31 billion.
Investors resisted.
Concerns included:
valuation
debt
and
execution risk.
That is significant.
Nvidia backing and AI exposure were not enough.
Public-market investors demanded more discipline.
THE MARKET IS MOVING FROM FOMO TO FUNDAMENTALS
For several years, investors feared missing the AI boom.
Now some fear paying too much for it.
That does not mean AI demand is disappearing.
Far from it.
TSMC continues posting record revenue.
Nvidia remains highly profitable.
Cloud companies continue expanding.
But investors are increasingly differentiating between:
strong businesses
and
expensive stories.
That is a healthy change.
NVIDIA REMAINS EXTREMELY PROFITABLE
This matters because not every AI company faces the same economics.
Nvidia sells the scarce hardware everyone wants.
Its margins are enormous.
Demand remains strong.
That gives Nvidia a much clearer economic model than some downstream infrastructure companies.
The risk is indirect.
If AI customers eventually slow spending, Nvidia’s growth could decelerate.
But today’s OpenAI revenue confusion is not evidence that demand has already collapsed.
TSMC IS SEEING THE SAME HARDWARE BOOM
TSMC just posted record quarterly revenue driven heavily by AI.
That suggests physical chip demand remains powerful.
Advanced-node capacity remains tight.
Packaging demand remains elevated.
So the semiconductor supply chain is not currently signaling an AI crash.
The debate is about:
sustainability.
How long can spending continue at the current pace?
ENERGY IS BECOMING PART OF THE EQUATION
AI requires enormous electricity.
Technology companies are signing deals for:
Nuclear power
Natural gas
Renewables
and
Grid infrastructure.
Power costs therefore increasingly influence AI economics.
And right now energy prices are elevated because of geopolitical conflict.
That creates another pressure point.
HIGH OIL PRICES CAN INDIRECTLY HURT AI
The connection works through inflation.
Higher energy prices can keep inflation elevated.
Higher inflation keeps interest rates higher.
Higher rates increase borrowing costs.
Higher borrowing costs make AI infrastructure harder to finance.
So geopolitical shocks far from Silicon Valley can affect the economics of:
Nvidia
OpenAI
and
data-center projects.
That is why markets are so sensitive right now.
THE NASDAQ SELLOFF REFLECTED THAT COMBINATION
On October 8:
The Nasdaq fell 1.25%.
The S&P 500 declined:
0.47%.
The Dow edged slightly higher.
Technology and semiconductor stocks underperformed.
Energy stocks benefited from higher crude prices.
That is a classic environment where:
expensive growth stocks
struggle
while
commodity producers
outperform.
THE BIGGER STORY: THE AI BOOM HAS REACHED THE ACCOUNTING PHASE
The first phase of generative AI was about:
technology.
Could the models work?
The second was about:
adoption.
Would people use them?
The third became about:
infrastructure.
How many GPUs and data centers are required?
Now the industry is entering another phase:
financial accountability.
Investors want to know whether all the spending produces:
real revenue
and eventually:
real cash flow.
OpenAI’s annualized revenue approaching:
$50 billion
is still extraordinary.
But the fact that investors thought the number was:
$70 billion
shows how difficult it remains to understand the private AI economy.
The problem is not simply that one number was lower than expected.
The problem is that trillions of dollars of market value now depend partly on financial information that remains:
private
non-standardized
and
difficult to compare.
That will become harder to tolerate as the infrastructure bills grow.
Broadcom is exploring tens of billions of dollars in financing.
Oracle is borrowing for AI hardware.
SpaceX is exploring giant GPU financing.
Data-center developers need enormous amounts of capital.
And Wall Street may need to provide as much as:
$1.5 trillion
to fund the industry’s buildout.
That makes one question increasingly urgent:
How much real revenue ultimately exists at the end of the chain?
OpenAI’s latest figure does not prove the AI boom is collapsing.
But it does prove that investors are no longer satisfied with spectacular growth alone.
They want transparent numbers.
And that may be the most important transition the AI industry has faced yet.
OpenAI may still be growing toward a $50 billion annualized revenue business at extraordinary speed — but Wall Street is beginning to realize that before it finances another trillion dollars of AI infrastructure, it needs to know exactly what counts as revenue in the first place.
