OpenAI may still be the biggest name in AI, but the cracks are getting harder to ignore. ChatGPT’s share of the AI assistant market has fallen below 50% for the first time. Anthropic has taken a huge lead in enterprise coding, Gemini keeps gaining ground, and OpenAI is having a much harder time turning its massive free user base into real money. Its ad business is also tracking far below the huge targets the company once laid out. That would already be a problem on its own. But OpenAI is also burning through billions of dollars while locking itself into enormous compute deals with Oracle, Microsoft, Amazon, Nvidia, and Broadcom. In Q1 2026, it posted a $6.95 billion operating loss on just $5.7 billion in revenue, while its total commitments through 2030 have climbed into the hundreds of billions. And the biggest threat might not even be another company. AI models are starting to look more and more alike. Token prices keep falling, open models are catching up, and intelligence itself may be turning into a commodity. If that happens, OpenAI could end up stuck in the exact part of the AI industry where competition is brutal, costs are massive, and the real money gets made somewhere else.
AI Generated·May contain errors
Video Summary
AI Generated
OpenAI is facing a severe financial crisis, characterized by massive operating losses and a shrinking market share in critical high-revenue sectors. Despite a large user base, the company is struggling to monetize its services, missing ad revenue targets and losing ground in the enterprise and coding markets to competitors like Anthropic and Google.
OpenAI is losing more than $2 for every dollar it makes.
Their Q1 2026 operating margin was negative 122%.
They are on pace to undershoot its ad revenue projections by 90%.
ChatGPT's market share has dropped below 50% for the first time ever.
Their enterprise share has been cut nearly in half.
And Sam Altman just admitted it himself.
OpenAI has not had our best 12 months ever.
And it's mostly my fault.
How did this happen?
ChatGPT has over 900 million weekly users.
It's the company that started the AI race, which is now falling behind.
Can they still catch up?
To answer these questions, we need to look at the state of OpenAI now and how we got here.
The first tipping point wasn't actually Anthropic, but Gemini.
In November 2025, Gemini 3 beat GPT 5.1 across essentially every benchmark.
This caused Altman to issue a code red at OpenAI.
Though OpenAI still had a gigantic lead, over 1.1 billion users,
with Gemini at 662 million and Cloud with 245 million.
But these numbers are quite deceptive.
OpenAI has a lot of users, but their control on the markets that matter
is shrinking.
In May, ChatGPT's global assistance share dropped below 50% for the first time,
to 46%, with Gemini at 28% and Cloud at 10%.
It seems like OpenAI is still in the lead, but things get a lot worse for OpenAI
when we look at what actually makes money, enterprise spending.
And oh boy, is it bad.
We estimate Anthropic now earns 40% of enterprise LLM spend,
from 24% last year and 12% in 2023.
Over the same period, OpenAI lost nearly half of its enterprise share,
falling to 27% from 50% in 2023.
Google also saw significant gains,
increasing its enterprise share from 7% in 2023 to 21% in 2025.
This is one of the biggest foundational problems.
And a big part of this is coding,
where LLMs have seen some of the most successful return.
Anthropic now has 54% of the coding market, with OpenAI at just 21%.
And part of the challenge is the delay or restraint of computing power.
About 40% of US data centers planned for 2026 are already delayed,
transformer lead times have stretched to 5 years in some regions,
and OpenAI has cancelled sites in the UK, Norway and Lordstown, Ohio.
OpenAI is struggling in the deep high-revenue niches that matter.
But there's another part we need to talk about, ads.
OpenAI and others have claimed that ads are their saving grace,
and it very well could be the way to monetize the free tier,
because boy are there a lot of them.
Out of 900 to 910 million weekly active users, 95% of them are on the free service,
hence why OpenAI is at a target of $100 billion in ad revenue for 2030.
If they can't dominate coding or enterprise, maybe they can be the mass-market ad service?
Well, OpenAI's ad business is on pace to miss this by 90%.
In fact, the standalone AI chatbots like ChatGPT, Gemini, Copilot, Google AI Mode and more
will have cumulatively generated under $1 billion in ad revenue in 2026.
By 2030, it's expected to be just $5.41 billion across the entire market.
OpenAI's single-company 2030 target is roughly 20 times larger
than eMarketers' estimate for the entire US chatbot ad market.
But why?
They have so many users, why can't they monetize them?
Well, a few reasons.
It turns out, prompt replies are surprisingly hard to advertise on.
Only about 2% of ChatGPT prompts, aka what the person is searching,
involve purchasable products.
So, it's hard to sell something when the prompt doesn't relate in any way.
Hence, ChatGPT ads have a terrible click-through rate of 0.91 to 1.3%.
For context, Google search ads are around 29.2%,
where basically everything you search has some kind of relevant ad.
AI might be taking a lot of web traffic from search results,
but it sure doesn't seem to be taking away the ad revenue.
According to advertisers, the ad tech stack is also still primitive.
CPMs on the ad business have already collapsed from $60 to as low as $25.
To make matters worse, 80% of users sent fewer than 1,000 messages all year in 2025,
an average of fewer than 3 prompts per day.
So, the chance to see ads, let alone relevant ads, isn't great.
But with all that said, none of that matters if OpenAI is actually making money.
So, how is that looking?
Well, OpenAI has two gigantic problems to deal with before they can even start.
Let's look at things right now.
OpenAI has forecasted a $14 billion loss for all of 2026,
with losses totaling $44 billion from 2023 through to 2029.
But then, we got the Q1 2026 results, with a $6.95 billion operating loss
on $5.7 billion in revenue, which is a negative 122% operating margin.
So, for every dollar OpenAI made, they spent more than two.
They are one quarter into the financial year,
and have already used half of the loss they projected for all of it.
Once you add stock compensation of $2.3 billion,
the gap operating loss is as high as $9.3 billion.
FutureSearch now forecasts a $33 billion gap loss for 2026.
But as we'll soon find out, these losses aren't nearly the entire picture.
OpenAI needs to start making some real money, which is proving difficult.
OpenAI promised revenue would start booming eventually,
and revenue was and technically still is growing, but not so much anymore.
OpenAI grew from $6 billion to roughly $25 billion ARR over the 17 months to May 2026.
Growth then flattened.
Their run rate held nearly $25 billion from February through the spring.
OpenAI's internal target for mid-2027 is $62 billion.
My forecast median is $44 billion.
ChatGPT weekly active users have also plateaued
at around 900 million, with only 50 million being paying subscribers.
OpenAI can't seem to monetize their exorbitant user base,
which is important because of OpenAI's massive bills and capital commitments.
Starting in 2027, OpenAI will be paying Oracle $300 billion across 5 years for cloud capacity,
about 4.5 gigawatts, or the electricity of about 4 million homes.
They have a deal with AWS for $38 billion across 7 years,
for access to mass amounts of NVIDIA GB200 and GB300 GPUs.
Then, it expanded that agreement by $100 billion across 8 years.
OpenAI has also committed $250 billion in Microsoft Azure, aka cloud services,
as well as 10 gigawatts of custom AI accelerators from Broadcom.
Basically, OpenAI will design its own chip with Broadcom building it.
We don't know exactly how much this will be,
but the Financial Times estimated between $350 and $500 billion.
There are even more agreements, but in total, according to CFO Sarah Fryer,
OpenAI has current commitments of $750 billion through 2030.
And remember, there's way more spend beyond 2030.
Though, we also need extra context.
These are massive bills, but most of these companies are also funneling money into OpenAI.
They just announced a $110 billion funding round, twice the size of their last raise.
Amazon invested $50 billion, NVIDIA invested $30 billion,
and SoftBank invested $30 billion in the round.
Worldwide AI spending is estimated to reach $2.59 trillion in 2026,
a 47% increase over 2025.
The Magnificent Seven have planned more than $700 billion in AI capex for 2026,
up from roughly $400 billion in 2025.
This gives us insight into OpenAI's runway.
Their cash reserves and marketable securities are at $73 billion.
So, they do have cash on hand, and that's up from $40 billion at the end of 2025.
But they've also promised a lot of money to quite a few companies,
and they're burning through cash very quickly.
So, can OpenAI catch up?
One thing people point to for this is their IPO.
This has potential for a gigantic capital injection.
Sam Altman believes OpenAI has a valuation well over a trillion dollars,
but now they've pushed it back to next year, and it makes sense as to why.
OpenAI isn't winning the race anymore.
Well, they are in pure user counts, but not much else.
Which brings us to the big problem.
OpenAI signed contractual obligations against revenue that hadn't arrived yet.
But OpenAI was winning, so they committed big.
Though now, they're no longer winning the AI race, but still spent the winnings of said race.
Here's where things get really interesting though.
At the BlackRock Infrastructure Summit in March 2026, Sam Altman said something important.
Something massive is happening in AI, not too dissimilar to other industries.
It turns out Altman may be right, but maybe not in the way that he hopes.
This is bad for all AI model companies, not just OpenAI.
If you've watched this far, consider subscribing.
You probably enjoy our stuff, and it helps the channel too.
Thank you.
Let's give that earlier quote a bit more context.
Altman is referencing a phrase made by Louis Strauss,
chair of the Atomic Energy Commission in the 50s, where he said,
It is not too much to expect that our children will enjoy in their homes
electrical energy too cheap to meter.
We'll know of great periodic regional famines in the world only as matters of history.
We'll travel effortlessly over the seas, and under them,
and through the air with a minimum of danger and at great speeds.
And we'll experience a lifespan far longer than ours.
This was his prediction about nuclear power, which never came true.
But something else happened.
Electricity became one of the most tightly regulated industries in the developed world.
It still became a commodity with extremely thin margins.
A part of that ecosystem absolutely makes a ton of money.
The turbine makers, transformer manufacturers, and transmission networks are very profitable.
The value in electricity moved down the stack.
Or rather, it was always there anyway.
What makes all of this so ironic is that Sam Altman literally predicted all of this.
In his June 2025 essay, The Gentle Singularity, he wrote that,
As data center production gets automated,
the cost of intelligence should eventually converge to near the cost of electricity.
It's not just electricity either.
The internet eventually turned into a commodity.
A gigabit from brand A is the same as brand B,
so a customer has no reason to consider anything but price.
The part that makes real money accumulates where it can't move.
For a while, the company is selling the actual equipment, like Cisco and Corning.
Cisco was briefly the most valuable company in the world in 2000.
But when the build-out of internet infrastructure largely stopped, the value collapsed.
Now, the value is up the stack in the consumer end.
Companies like Google or Netflix, who have all your data,
information, and habits, have a high switching cost.
Just not a monetary one.
So how does this relate to AI?
Because the exact same thing is happening.
In 2023, both the GPUs and AI models were scarce.
But now, in 2026, GPUs are still scarce, but models are not.
And they're all getting closer and closer.
The gap between the best closed model and the best open rate model on chatbot arena
fell from 8.04% in January 2024 to 1.7% in February 2025.
The supply side is now effectively infinite.
The scarce asset becomes the ability to run them well.
Throughput, latency, cost per token, routing, caching, fine-tuning, evaluation, and guardrails.
The value in all of these areas accumulates where it is hardest to substitute.
The value accumulating down the stack.
At the bottom, with the most mass or value, are the semiconductors and compute.
NVIDIA's data center revenue hit $75.2 billion, up 92% year-over-year at a 75% gross margin.
Roughly a $300 billion annualized run rate.
The money is in companies selling GPUs, semiconductors, compute.
Not intelligence.
That might change if the supply for GPUs and compute outweighs demand.
There's also value near the top layer.
Not in chatGPT, but in third-party consumer-facing software.
But this is rather thin, as AI apps have lower margins than regular SaaS products.
Just look at Amazon.
They're cutting back their AI model, Nova, because, at least I believe,
they've figured out this part of the industry isn't the moneymaker.
It's the one below it.
This is why OpenAI is in such a tough spot.
They are stuck in the least valuable layer.
And as there are more comparable models and prices come down,
paying those massive bills might be tough.
What's interesting is how AI and intelligence have become a commodity,
which is in tokens, charged by use like electricity.
And now, token cost has even gone far beyond the costs of the employees.
But something weird has happened.
Tokens aren't actually getting more expensive.
The price of a single token has fallen by about 90% since 2023.
So if intelligence and tokens are cheaper, why is AI spend getting so much bigger?
Click here to learn the other side of the story.
But until then, I'm Hari, and I'll see you guys on the next one.