How Many Companies Is OpenAI Trying to Be?

Three things happened at OpenAI inside 72 hours this week.

Monday, Bloomberg reported a completed $7 billion tender offer, the company buying back employee shares at an $852 billion valuation, flat to its March round. Tuesday, Brad Lightcap, eight years in and four of them as COO, told staff he was leaving to start something new. Thursday, chief revenue officer Denise Dresser announced her exit after nine months, replaced by Dali Rajic, until recently president and COO of Wiz.

Add Fidji Simo, the company's number two, who stepped down in July when a chronic illness worsened. Add Barret Zoph, who ran enterprise sales and left in June, five months after rejoining. Add the three executives who went in April, including the head of Sora and the VP of OpenAI for Science.

The take writing itself right now is that OpenAI's revenue chief is fleeing just as the company bets everything on advertising. That reading doesn't survive the résumés, and what replaces it is worse.

Start with the résumés

Dresser spent more than a decade at Salesforce and ran Slack as CEO before joining OpenAI last December. Rajic was CRO and then COO at Zscaler, chief customer and revenue officer at AppDynamics, and president and COO at Wiz, the cloud security company Google bought this year for $32 billion.

There is no advertising anywhere in that. What there is, top to bottom, is expensive software sold to large companies with procurement departments, security reviews and multi-year contracts. Zscaler and Wiz are two of the better enterprise sales machines anyone has built this decade. If your problem is losing Fortune 500 accounts to a competitor, Rajic is close to the first call you make.

Ads at OpenAI belonged to Simo, who grew Instacart's advertising business to nearly a billion dollars a year and spent a decade at Facebook before that, along with the ads leadership she recruited underneath her, including a hire out of Meta. If you want the departure that genuinely threatens the ads roadmap, it already happened. It was July, and most of the coverage filed it as a health story instead of a business one.

The revenue seat at OpenAI is the enterprise seat. Bloomberg counts Rajic as the second chief revenue officer OpenAI has named in under a year, and the broader operating role that seat got fused with has turned over three times since April.

Everyone has the map upside down

Here's the version of the OpenAI and Anthropic rivalry that gets repeated: OpenAI owns consumer and is scrapping to defend enterprise.

The numbers point closer to the reverse. OpenAI owns consumer outright. ChatGPT's mobile apps passed a billion monthly users in May and nothing else is within shouting distance. What OpenAI is losing, and losing badly, is enterprise, to a company with a small fraction of its user base.

Menlo Ventures' State of Generative AI in the Enterprise put Anthropic at 40% of enterprise LLM spend against OpenAI's 27%, down from the 50% OpenAI held in 2023. Ramp's corporate card data, which tracks actual payments across more than 50,000 US businesses, showed Anthropic passing OpenAI on business adoption for the first time in April, 34.4% to 32.3%. By February, per the same dataset, Anthropic was already taking roughly 70% of head-to-head matchups among businesses buying AI for the first time. In coding, the one use case that has clearly broken out, the gap is wider still.

Consumer is where OpenAI's numbers have gone quiet. Similarweb had ChatGPT's daily mobile actives essentially flat through the first four months of this year, with web traffic up 7% year over year. Sensor Tower clocked its share of chatbot usage under 50% in March for the first time, down from 81% two years earlier.

Claude's growth off a small base is enormous, 627% year over year by one count, moving from 1% of AI app downloads in 2025 to 14% in Q2, and credit card data suggests it is winning paying consumers faster than anyone but ChatGPT. But a rounding error growing tenfold is still a small number. Anthropic hasn't won consumer. It has shown up in consumer, which is a different sentence.

What you're watching is two companies attacking out of their strength and into the other's home market. Axios reported this week that Greg Brockman is assembling an operating team meant to take OpenAI to number one in enterprise adoption, ahead of Anthropic. That's the brief. Rajic is the hire that fits it.

Why the seat keeps emptying

Which brings up the thing the churn is actually asking. Why has this role chewed through the people who take it?

The likeliest explanation is that OpenAI is running three companies with three incompatible metabolisms, and the revenue seat sits where they grind against each other.

A consumer advertising platform runs on scale and targeting and engagement loops. That's Meta physics. An enterprise software business runs on trust, compliance, procurement cycles and reference customers, a different sport played at a different speed. Underneath both sits a frontier research lab that posted a $20.9 billion operating loss on $13.1 billion of revenue in 2025, according to audited documents obtained by Ed Zitron and independently verified by the Financial Times.

Whoever holds the revenue seat has to speak all three languages while the org chart gets redrawn around them every quarter. Dresser absorbed Lightcap's operating scope in April, then Simo's when Simo went on leave. Three jobs collapsed into one, then vacated. That doesn't look like a hiring mistake. It looks like a job description nobody could satisfy.

Notice who ended up absorbing the vacuum. Not another professional operator. Brockman. A co-founder. An engineer. When a company that may list within a year answers operator churn by recentralizing around founders, you can read it two ways. Generously, it's founder mode before a hard sprint. Less generously, the operators kept leaving because the thing isn't operable by operators.

The ceiling OpenAI poured itself

Now the ads business, which has the stranger problem.

ChatGPT ads went live in the US in February and reached the UK, Mexico, Brazil, Japan and South Korea this week. AdClarity's estimate had average monthly ad spend around $109 million by May, pacing toward maybe $500 million this year against an internal target reported at $2.5 billion. So it's running at roughly a fifth of its own near-term goal while being cast as the lever that closes a twenty-billion-dollar hole.

Forget the shortfall for a minute. The design is the strange part.

OpenAI has walled ads off from paid tiers, from anyone under 18, from temporary chats and from sensitive categories including health, mental health and politics. Advertisers get no access to conversations. And the model itself is supposed to be blind to the ad's existence. Simo said publicly that if you ask ChatGPT about an ad sitting on your screen, it will tell you straight when the product is bad, and she acknowledged some brands would hate that.

Sit with it for a second. OpenAI has built an advertising business that promised in advance never to sell the only thing advertisers actually want, which is influence over the answer. Google's model works because the sponsored result is a result. OpenAI deliberately built the opposite, sponsored content parked beside an answer it cannot move.

Ethically it's the right call. Commercially it's a ceiling they poured themselves, and they've fenced it off from the highest-intent inventory in the building. The health questions. The financial ones. The users with a card already on file. What's left is the cheapest surface they own, monetized by the most constrained ad product on the market, funding the most expensive research operation ever assembled.

Every dollar above that ceiling has to be bought with trust. Trust happens to be the same currency they're bleeding in enterprise.

What the buyback tells you

Back to the listing. OpenAI filed confidentially on June 8. The Times then reported the company weighing a 2026 debut below a trillion against holding out to 2027 for a cleaner shot at it. Kalshi traders currently put about 59% odds on an announcement by March 2027.

The $7 billion tender at a flat valuation is the giveaway. Buybacks bleed off the internal pressure that builds when employees sit on large illiquid stakes, and that pressure is one of the main things that pushes a company public. You do this when you don't plan to list soon.

You also couldn't list comfortably right now. Every S-1 carries a dependence-on-key-personnel risk factor. Yours reads differently when three C-suite exits land inside five weeks and your incoming revenue chief has been in the building a week.

The case against all of this

Executive turnover at a company growing this fast is normal and often healthy. The job that existed at 500 people frequently doesn't exist at 5,000. Lightcap had already moved out of the COO role back in April, and an eight-year operator with Y Combinator history leaving to found something is the most ordinary move in the Valley. Simo's exit is a health matter and shouldn't be read as a business signal at all. Rajic is a strong hire by any measure. OpenAI's business revenue reportedly jumped 32% month over month in July, and the company expects enterprise to be half of total revenue by year end. A billion monthly users is an asset nobody else on the planet has.

Anthropic has real weakness of its own, and the most useful skeptic here is Ramp's own economist, who published three reasons he stays bearish in the same report that crowned Anthropic the leader: token-based pricing that misaligns its incentives with cost-conscious buyers, compute constraints, and service quality complaints. Add no consumer scale to speak of, and a June export-control episode that forced it to pull its most capable models off the market entirely.

Be careful with the revenue comparisons flying around, too. Most of the precise ARR figures for either company trace back to aggregators citing each other. Directional at best.

The actual question

Whether OpenAI is in trouble is the wrong question. A company with a billion users and $122 billion of fresh capital isn't in trouble in any sense that matters this year.

The real one is narrower. Can one company hold a consumer advertising platform, an enterprise software business and a frontier research lab inside a single org chart?

History suggests the tension gets resolved rather than sustained. Google let search advertising pay for everything and subordinated the rest to it. Meta walked away from enterprise and never looked back. Microsoft has never made a serious run at consumer advertising at scale, which is arguably why it still has both of its other businesses.

OpenAI is attempting all three at once. The layer of the company whose entire job is translating between them is the layer that keeps coming apart.

So watch what Rajic actually gets handed. If it's enterprise, cleanly, the hire makes sense and the churn was noise. If ads and consumer monetization land on his desk too, the seat empties again inside a year, and this stops being a story about personnel.