Anthropic's Revenue Grew 7x. OpenAI's Grew 18%.

The two companies racing toward IPOs are posting wildly different growth curves — and their executives are handling the gap in opposite ways

Monday, August 24, 2026 · 8 min read · Issue 246

OpenAI CFO Sarah Friar told employees at an all-hands meeting last Wednesday that the company "will be a public company in 2027," and asked them not to panic if Anthropic files first. "There is a chance they pull the cover off that confidential file in the coming weeks and become public in September. That's OK, we are running our own race," she said, according to two people familiar with her remarks. The reassurance was necessary because the race isn't close. Anthropic told investors over the weekend that its annualized revenue run rate hit $65 billion at the end of July — a sevenfold increase from a year earlier — with $11.5 billion in preliminary second-quarter revenue. OpenAI's second-quarter revenue was $6.7 billion, up 18% from Q1, pushing its annualized run rate past $40 billion. Both numbers are large. Only one of them is accelerating.

Friar's slides tried to reframe the comparison: OpenAI's overall revenue run rate is up 35% quarter-to-date, enterprise revenue run rate up 50%, and its coding products have hit 20 million weekly active users. Brockman, addressing a wave of executive departures — revenue chief Denise Dresser left after eight months, longtime executive Brad Lightcap ended an eight-year run, product chief Fidji Simo stepped back for health reasons — called the turnover "not actually that atypical," adding that OpenAI's scrutiny is a function of being "so much in the spotlight." That's a defensible point about departures. It doesn't address why Anthropic's revenue is compounding at a rate OpenAI's isn't, with OpenAI carrying an $852 billion valuation into an IPO it needs Wall Street to believe in.

The two companies are also diverging on how they treat customer data, and the timing is not a coincidence. Anthropic last week told enterprise customers it will require 30-day data retention on its most capable models — Claude Fable 5 and Mythos 5 — reversing the zero-retention default. Anthropic's own risk report called this "a decision we believe will be unpopular with customers... and pose real risks to our business success," but said it's "essential to detect and prevent sophisticated attacks that span multiple requests." OpenAI moved the opposite direction the same week, previewing "Private Safety Processing" — a technique it says lets it flag misuse patterns across related conversations while preserving zero data retention, with data staying on customer infrastructure or encrypted with customer-held keys. Both companies are responding to the same problem: frontier models are now capable enough that risk shows up across sessions, not within one prompt. They've reached opposite conclusions about whether solving it requires giving up the privacy guarantee enterprises were sold on.

Two frontier labs, two growth curves, two answers to the same safety question, both racing the same IPO clock. Everything else this week — a Chinese e-commerce giant's stock cratering on AI capex, a US governor turning on the industry that courted him, a London lab quietly out-scoring both companies at a narrow task — is downstream of that same tension: capability is compounding, and nobody has fully worked out who absorbs the cost.

Briefs

Meta hires OpenAI veteran Luke Metz for Superintelligence Labs. Metz, who left OpenAI for Mira Murati's Thinking Machines in 2024 before rejoining OpenAI earlier this year, starts at Meta this week reporting to Alexandr Wang. It's the latest move in what Axios calls the ongoing AI talent wars, following Meta's Scale AI acquisition that brought Wang aboard as head of AI efforts. (Axios)

Google's Gemma family passes one billion downloads. Over two years, developers have published more than 100,000 Gemma variants; NASA, Satlyt, and Nvidia-backed Starcloud are running Gemma models in orbit for satellite image analysis and intersatellite routing. Yale and Google researchers built C2S-Scale on Gemma to identify a novel cancer therapy pathway, verified in living cells — the first time, Google says, an AI system produced a mechanistic therapeutic hypothesis confirmed experimentally. Google is launching an "Awesome Gemma" GitHub repository to catalog community projects. (Google Blog)

A London lab founded by DeepMind alumni says its 27-billion-parameter agent beat Claude Opus 4.8 and GPT-5.5 at a specific task. Inherent's Faraday agent, trained via reinforcement learning rather than instruction on scientific method, outperformed both frontier models at independently replicating published research findings without being told the answer. Cofounder Edward Hughes said the result mattered less than the training approach — teaching "research taste" through reward signals rather than rules. Faraday runs on a Qwen 3.6-based model roughly an order of magnitude smaller than the frontier systems it beat. (TechCrunch)

Texas Governor Greg Abbott says data center companies "dug their own grave." Abbott, who called Texas the "epicenter of AI development" just nine months ago while announcing a $40 billion Google investment, has since ordered regulators to make data centers pay full infrastructure costs and audit grid-connection requests. He joins Pennsylvania's Josh Shapiro and New York's Kathy Hochul in reversing course. An Annenberg Public Policy Center survey found 61% of Americans now oppose new data centers in their area, up from 49% in March — including 54% of Republicans. President Trump has publicly pushed back on Abbott's approach. (Axios)

Nvidia is playing matchmaker between GPU buyers and Nordic data-center capacity. Sources tell CNBC that Nvidia is connecting companies holding its chips with data-center operators that have available capacity in Sweden, Norway, Finland, and Denmark, extending its influence beyond chip sales into deal-making as the region's AI infrastructure boom accelerates. (CNBC)

Dispatch: China / East Asia

Alibaba shares fell as much as 10% in Hong Kong on Monday after the company priced an HK$80 billion ($10.2 billion) placement of new shares to non-U.S. investors, with all net proceeds earmarked for AI infrastructure. The placement lands four days after Alibaba reported a 75% year-over-year profit drop for the June quarter, driven by capital expenditure that jumped 75% to 67.7 billion yuan. Alibaba is issuing 710 million new shares at HK$112.70 — an 8.4% discount to Friday's close — to fund what it calls "full-stack AI capabilities." The company committed last year to at least 380 billion yuan in AI and cloud infrastructure spending over three years; Tencent's capex rose 65% quarter-over-quarter to 52.8 billion yuan over the same period chasing the same buildout. UBP senior equity advisor Vey-Sern Ling told CNBC Alibaba is "well positioned to chase that growth" given its cloud and model strength, but that "profits might weaken in the near term, while capex might rise." The market's answer on Monday was to sell the stock anyway — a live test of how much AI capex investors will fund on faith before demanding it show up in earnings. (CNBC)

Dispatch: India

India's National Health Authority has integrated Google's Gemma 4 model and open-source Medical Data Toolkit into Aarogya Setu 2.0, an app with more than 100 million Android downloads, letting citizens convert complex medical reports into standardized digital formats they can securely share across providers. Separately, Google's MedGemma model — built on Gemma — is supporting outpatient triage at the All India Institute of Medical Sciences, one of the country's largest public hospitals. Both deployments are part of what Google calls the "Gemmaverse": more than 100,000 community-published Gemma variants running everywhere from edge devices to orbital satellites. The throughline for India specifically is scale — a single open-weight model family now processing medical data for a population larger than most countries have citizens. (Google Blog)

Dispatch: Europe

Nvidia's push into Nordic data-center dealmaking (see Briefs) and Inherent's Faraday result (see Briefs) are two sides of the same regional story: Europe's AI infrastructure and research base are both quietly compounding while attention stays fixed on the US and China. Inherent's twelve-person team works entirely in-person out of London's King's Cross, the neighborhood DeepMind's presence helped turn into a top AI hub — and Hughes has publicly pushed to end the UK's "garden leave" practice, which bars departing employees from joining competitors for months, calling it a structural disadvantage against US labs that face no such restriction when poaching talent. Meanwhile Nordic grid capacity and renewable power are becoming the physical backbone for AI buildout that the US is now fighting itself over — the political friction playing out in Texas and Pennsylvania hasn't yet arrived in Scandinavia. (TechCrunch; CNBC)

The View

Every story this week is a variation on the same question: who pays for the gap between AI's promised trajectory and its current unit economics? Alibaba's investors just answered for China — sell the stock, fund the capex anyway, and revisit in a year. Texas and Pennsylvania voters are answering for the US — 61% opposition to local data centers isn't a rounding error, it's a policy constraint that Governor Abbott, who personally courted a $40 billion Google investment nine months ago, now says the industry earned by ignoring. And OpenAI and Anthropic are answering for themselves by making opposite bets on what enterprises will tolerate: Anthropic is betting customers will accept mandatory data retention in exchange for security guarantees ahead of an IPO it needs to look disciplined for; OpenAI is betting it can preserve zero-retention as a competitive differentiator while still catching threats, precisely because its revenue growth is the one metric where it's currently losing. Neither bet is obviously right. What's notable is that the industry no longer has one dominant business model everyone is converging toward — it has several competing ones, tested in public, with real capital and real voters keeping score.

The Miss

Coverage of Alibaba's placement led almost universally with the stock drop — a clean, single-number headline. Buried underneath: Tencent's capex also jumped 65% the same quarter, meaning the two largest Chinese consumer tech companies are now spending at rates that outpace their own profit growth simultaneously, not as an isolated Alibaba decision. That's a sector-wide capital allocation shift happening in China at the same moment US governors are actively trying to slow the equivalent buildout domestically — a genuinely interesting divergence in how two governments and two markets are absorbing the same infrastructure bet, and it got almost no comparative treatment anywhere this week.

Pull Quotes

"The IPO is not a finish line, it is a milestone, another fundraise." — Sarah Friar, OpenAI CFO

"I actually think that the difference between OpenAI and other organizations is that we are so much in the spotlight, so every departure gets scrutinized in a way that it doesn't otherwise." — Greg Brockman, OpenAI President

"We have recently announced our plan to require 30-day data retention on our most capable models — a decision we believe will be unpopular with customers... but which we believe is essential to detect and prevent sophisticated attacks that span multiple requests." — Anthropic risk report

"They basically dug their own grave for the problem that's been caused for them. And that's why they got the backlash they deserve." — Greg Abbott, Texas Governor

"What was most interesting to us about this was not so much the result of beating those frontier agents — which of course we liked — but was actually the way we went about building this." — Edward Hughes, Inherent cofounder

Reads & Links

• OpenAI 'will be a public company in 2027' or sooner, CFO Friar tells employees — CNBC
• OpenAI previews zero-retention safety system as Anthropic requires data logs — Axios
• Alibaba plunges after announcing $10.2 billion share placement to fund AI push — CNBC
• Meta hires OpenAI veteran Luke Metz — Axios
• Data centers "dug their own grave," says Texas Gov. Greg Abbott — Axios
• Nvidia plays matchmaker in Nordics as AI data center deals boom in region — CNBC
• Inherent, founded by DeepMind alumni, says its AI teammate outperformed Anthropic and OpenAI at replicating research — TechCrunch
• Inside the Gemmaverse: celebrating one billion Gemma downloads — Google Blog

Out

That's the briefing. Issue 247 tomorrow.