OpenAI's $70B Run Rate and the AI Deal Wars

A sales executive leaves a company he never officially worked for. He was an advisor, a board observer, someone the founder called every day. Then he takes the top sales job at a direct competitor, and the founder calls it a betrayal in public. It sounds like a small, petty story. It is actually a useful window into what the people building and funding artificial intelligence companies now consider normal, and what they still, grudgingly, consider a betrayal.

The executive is Chris Degnan, a veteran sales leader who had been a board observer and informal advisor at the AI coding startup Factory, run by founder Matan. While advising Matan, Degnan was also interviewing to become chief revenue officer, or CRO, at Cognition, a rival AI coding company (12:37). When Matan found out, he accused Degnan publicly of disloyalty. On this week's 20VC, host Harry Stebbings, SaaStr investor Jason Lemkin, venture investor Rory O'Driscoll, and guest Dave Ittycheria, the chief executive of the database company MongoDB, picked the episode apart to ask a bigger question: has loyalty actually changed in the AI industry, or does it just look different because everything is moving faster.

When Loyalty Breaks

Ittycheria, who has hired and fired plenty of sales leaders, drew a sharp line. A CRO who switches industries is normal. A CRO who defects to a direct competitor is not.

I've never seen a CRO flip from one competitor to another. I've seen a CRO go from one company to another in a different space, but I've never seen a CRO flip to a direct competitor. (14:08) Dave Ittycheria

His reasoning is not sentimental. A CRO recruits a sales team by selling them a story about the company's future. If that same person jumps to the enemy, everyone they hired is left holding a story that turned out to be disposable. O'Driscoll pushed back a little, noting that Degnan was only an advisor, not a full-time employee with board duties, so the rules were murkier than Matan's public anger suggested. Lemkin went further, arguing that in the current market, roughly 95 percent of CROs would see nothing wrong with taking the best offer on the table, competitor or not, because job turnover at that level has simply sped up.

Vinod Khosla, the founder of Khosla Ventures and an investor who had led Factory's most recent funding round, then made things worse. He posted publicly that Matan was a struggling second-tier competitor, an attack on the founder of his own portfolio company (23:03). The panel's reaction was closer to professional embarrassment than outrage.

Basically handed every other firm a weapon when they're competing on deals saying, is this the partner you want when things go bad. (24:00) Jason Lemkin

The point is not that Khosla was wrong about the company. It is that venture firms who back two competitors in the same category survive by staying quiet in public, and Khosla broke that unwritten rule in a single tweet, handing every rival firm a talking point for future pitch meetings.

The Money Behind the Noise

Underneath the personnel drama sits a much larger shift in the economics of the industry, and it starts with OpenAI. After a wobbly first half of the year, in which OpenAI's GAAP revenue, meaning revenue recognized under standard accounting rules rather than bookings or projections, grew just 18 percent quarter on quarter in Q2 while Anthropic was growing roughly tenfold, OpenAI's growth reaccelerated to 70 percent quarter on quarter in Q3 (08:12). That pushed its run rate, an annualized estimate of revenue based on the current quarter's pace, to nearly $70 billion (04:44), alongside reports of a $1.4 trillion valuation in a new funding round (04:29). Ittycheria pointed to a simple, overlooked reason: enterprise customers do not switch AI vendors as casually as individual developers do, because switching means retraining staff and rebuilding workflows, so once OpenAI is inside a company's procurement system, it is hard to dislodge (06:52). The timing matters too, because Anthropic is expected to file for an IPO in the middle of November, and its own Q3 numbers will land in that prospectus just as the market judges whether OpenAI's comeback came at Anthropic's expense (10:15).

Voice technology supplied its own outsized number. ElevenLabs, an AI voice company, doubled its valuation to $22 billion (33:55). Lemkin argued the case rests on a narrow but real technical bar: voice applications need response times fast enough that a caller never notices a delay, because if an automated system is picking up a phone for a florist's order line, it has to answer correctly in seconds or the product is worthless (35:30). Meanwhile Salesforce bought Listen Labs, an AI-powered market research company, for $2 billion (37:02), a deal Ittycheria said he intellectually understood but didn't quite believe in, given that $2 billion is a rounding error against Salesforce's roughly $60 billion in revenue (39:13). Ittycheria's framing was sharper: many AI applications are really just clever features sitting on someone else's platform, dressed up as companies, and the real test of durability is whether a product builds a data loop, where usage creates proprietary data that improves the product and attracts more usage, something a feature can't easily replicate (39:55).

The most forward-looking number came from Vercel, the application hosting company, where Lemkin noted that fully 50 percent of new business is now being driven by autonomous AI agents choosing the platform on their own, up from just 3 percent earlier this year, on $600 million of annual recurring revenue (46:30). This is not a human reading reviews. It is software deciding which vendor to use without being asked.

When agents pick you, it's a force of nature right now. And agents, what I've learned from them, have opinions, and they're really hard to argue with. (47:41) Jason Lemkin

Ittycheria connected this to twenty-five years of search engine optimization suddenly needing a sequel: companies now have to make sure AI agents, not just human searchers, can find and trust them, or risk becoming invisible in a market where the buyer is a piece of software (48:47).

Not every structure built for this speed survives contact with the law. Nvidia's $17 billion deal to license technology from and hire engineers out of the AI chip startup Groq, rather than acquire it outright, is now facing a shareholder lawsuit in Delaware from two former Groq engineers who argue the arrangement was an acquisition in substance, not form, and that it left ordinary shareholders worse off than those who crossed over to Nvidia (55:13). O'Driscoll explained why this format exists at all: it is a workaround for the slow pace of antitrust review at the Federal Trade Commission, letting a deal close overnight instead of waiting roughly nine months for approval (58:33). The catch, as Lemkin put it, is that calling something a license instead of a merger does not change what it functionally is.

Just 'cause you say it's not a duck, it don't mean it's not a duck. I think some of these are ducks. (60:00) Jason Lemkin

The episode closed on two smaller but telling signals. First, Meta's new consumer AI assistant, Muse, drew praise from the panel for its polish and resourcing, in sharp contrast to OpenAI's Dots launch, which Stebbings called tech-centric and underwhelming for ordinary users (61:21). Lemkin suggested Dots might simply be built for a different audience, developers running persistent coding agents, rather than consumers, and that judging it on consumer appeal might miss the point entirely (63:20). Second, the home security and personal safety company Aura pulled its IPO despite growing 74 percent on $1.2 billion in revenue, which Ittycheria guessed came down to a gap between what bankers had promised the board on pricing and what public investors were actually willing to pay (66:55). It is a reminder that even in a year of extraordinary valuations, the public markets still set their own price, no matter how fast the private story moves.