OpenAI Chose Ads. Anthropic Chose Users. The Score Flipped.

The most expensive Super Bowl punchline in history

On February 9, 2026, OpenAI flipped the switch on advertising inside ChatGPT. Two days earlier, Anthropic had aired its first-ever Super Bowl campaign — four darkly comic spots directed by Jeff Low via Biscuit Filmworks, each depicting an ad-funded AI assistant derailing conversations about health, parenting, and work with jarringly irrelevant product pitches. The tagline landed like a verdict: “There’s a time and a place for ads. Your conversations with AI should not be one of them.” The spots won the Clio jury’s Most Creative Commercial award, pushed Claude into the top ten free apps on the Apple App Store, and boosted Anthropic’s site traffic 6.5 percent overnight. Two months later, the scoreboard tells the rest of the story: Anthropic has overtaken OpenAI in revenue for the first time in history.

The numbers are unambiguous. Anthropic’s annualized revenue run-rate hit $30 billion in April 2026, up from $14 billion in February and $9 billion at the end of 2025. OpenAI sits at roughly $24 billion in annualized run-rate, generating about $2 billion per month. The growth trajectories tell an even starker story: Anthropic doubled from $14 billion to $30 billion in eight weeks while OpenAI’s consumer base was hemorrhaging. Eight of the Fortune 10 are now Anthropic customers. More than a thousand companies spend over $1 million annually on Claude. The enterprise mix is brutal for OpenAI’s thesis — 80 percent of Anthropic’s revenue comes from business customers, compared to OpenAI’s more consumer-heavy composition where enterprise only recently crossed 40 percent.

The irony cuts deeper than market share. OpenAI’s decision to monetize through advertising was supposed to diversify its revenue streams ahead of a planned IPO targeting a $1 trillion valuation in late 2026 or early 2027. The company told investors to expect $2.5 billion in ad revenue this year, scaling to $11 billion in 2027, $25 billion in 2028, $53 billion in 2029, and $100 billion by 2030. Those projections assume ChatGPT’s weekly active users will reach 2.75 billion — a figure that exceeds Instagram’s current user base and would make ChatGPT the most widely used software product in human history. Instead of building toward that future, the ad launch detonated a user revolt that may have permanently altered the competitive landscape.

The backlash was immediate and viral. A grassroots campaign called QuitGPT surged to 2.5 million participants, ChatGPT uninstalls spiked 295 percent in a single day, and Claude hit number one on the U.S. App Store for the first time. ChatGPT’s market share cratered from an estimated 60 percent to under 45 percent, with users scattering to Claude at 18 percent, Gemini at 15 percent, and DeepSeek at 7 percent. The QuitGPT movement was not solely about ads — a Pentagon military deal and political donation disclosures compounded the anger — but the advertising integration became the most visceral symbol of a company that had stopped working for its users and started working for its shareholders.

Sixty dollars per thousand impressions and 2.5 million goodbyes

The economics of ChatGPT advertising reveal a product designed for maximum extraction. OpenAI charges approximately $60 per thousand impressions — roughly three times Meta’s average CPM rate — with a minimum buy-in of $200,000 to participate in the beta program. Two ad formats launched in February: a shopping product carousel with Etsy and Shopify integration for direct checkout, and a conversational banner where users can ask ChatGPT questions about the advertised product. The ads use the current conversation topic and, if the user has personalization enabled, past chat history and ChatGPT Memory to select relevant placements. Within six weeks, the pilot crossed $100 million in annualized revenue and attracted more than 600 advertisers.

The targeting is where the model gets uncomfortable. Unlike Google, which infers intent from a search query, or Meta, which builds profiles from social behavior, ChatGPT’s ad engine mines something more intimate: the content of your conversations. Users discussing medical symptoms see pharmaceutical ads. Users working through financial stress see loan offers. The ad system is indexing the most private dialogue many people have with any piece of software, and doing so with the explicit trust that comes from an AI assistant framed as working in the user’s interest. Anthropic’s Super Bowl campaign understood this intuitively — the humor in those spots worked precisely because the premise felt like a betrayal.

OpenAI restricted ads to the free tier and the $8-per-month Go subscription tier, keeping Plus ($20), Pro ($200), Team, Business, Enterprise, and Education accounts ad-free. With 900 million weekly active users in February and only 50 million paying subscribers, that leaves over 840 million users on ad-eligible tiers — a massive addressable audience by any measure. But the segmentation created a perverse incentive structure: the free product that attracts users and builds the funnel now actively degrades their experience to generate revenue, while the premium product remains pristine. Every user who upgrades to escape the ads is a user OpenAI earns more from through subscriptions. Every user who leaves for Claude is a user Anthropic earns from instead.

The revenue comparison between the two companies illuminates a deeper divergence in business philosophy. OpenAI projects $14 billion in losses on $25 billion of revenue this year, with cumulative losses of $44 billion projected before the company reaches profitability around 2029. It has committed to over $500 billion in disclosed cloud capacity, including a $250 billion Azure contract with Microsoft, and its compute spending is expected to consume 75 percent of total revenue. Anthropic, by contrast, projects peak training costs around $30 billion annually — roughly four times less than OpenAI’s projected $121 billion in compute spending by 2028. Anthropic is targeting profitability in 2028-2029 without relying on advertising as a revenue stream.

Cross-referencing these figures yields a proprietary insight that neither company states directly: OpenAI’s ad gamble is not a growth strategy — it is a survival strategy. The company is burning $17 billion this year, has committed to infrastructure contracts that assume exponential user growth, and is preparing an IPO that requires a narrative of unstoppable momentum. Advertising revenue provides the second revenue line that makes the story work on a roadshow slide. But if the ads drive users away faster than they generate revenue, the entire financial model inverts: costs stay fixed because the infrastructure commitments are locked in, while the user base shrinks and takes both subscription and advertising revenue with it. OpenAI needs advertising to work and users to tolerate it. Those two requirements may be fundamentally incompatible.

The global digital advertising market provides sobering context for the $100 billion target. Meta is projected to generate $243 billion in ad revenue in 2026, overtaking Google’s $240 billion for the first time. The entire global digital ad market will surpass $1 trillion for the first time this year. OpenAI’s 2030 target of $100 billion would make it the third-largest advertising platform on earth, behind only Meta and Google, and ahead of Amazon. That projection requires not only 2.75 billion weekly users but an ad product sophisticated enough to compete for budgets currently flowing to platforms with decades of targeting infrastructure, measurement tools, and advertiser relationships. OpenAI has six weeks of data and a user base that staged a mass exodus the moment the first banner appeared.

The three ways this ad empire crumbles

The most dangerous assumption in OpenAI’s advertising thesis is that user growth will continue despite the ads, not because of them. Google Search succeeded with advertising because users came for search results and tolerated ads as a minor tax on an otherwise unmatched product. Facebook succeeded because the social graph created switching costs that made leaving painful even as the ad load increased. ChatGPT has neither of those structural advantages. AI chatbots are commoditizing rapidly — Claude, Gemini, DeepSeek, and a growing roster of capable alternatives offer comparable functionality without advertising. The switching cost for a ChatGPT user is approximately the time it takes to create an account on a competitor’s website. When OpenAI’s ads appeared, 700,000 users demonstrated exactly how low that switching cost is in a single week.

The second vulnerability is that OpenAI’s ad targeting is simultaneously too powerful and not powerful enough. Too powerful because the intimate nature of conversational data triggers privacy revulsion in a way that behavioral targeting on social media does not — users feel surveilled differently when their one-on-one conversation with an AI is visibly shaping the ads they see. Not powerful enough because ChatGPT lacks the purchase intent signals that make Google Search ads so valuable, and the social context that makes Meta’s ads so targetable. A user asking ChatGPT to explain quantum computing is not necessarily in the market for a physics textbook. A user venting about a bad day is not necessarily ready to book a vacation. The conversational context is rich but ambiguous, and the early $60 CPM pricing suggests OpenAI is charging a premium for novelty rather than proven ROI. As the novelty fades and advertisers demand conversion data, the CPM will either drop or the advertiser base will shrink. Both outcomes undercut the $100 billion projection.

The third risk is reputational, and it compounds over time. OpenAI had already fumbled an earlier attempt at monetizing the ChatGPT interface. In December 2025, the company pulled “app suggestion” promotional messages after paying Plus and Pro subscribers discovered Peloton and Target promotions embedded in their conversations. Chief Research Officer Mark Chen acknowledged the company “fell short” and disabled the feature. CEO Sam Altman declared a “code red” and paused advertising plans to focus on product quality. Two months later, the company launched a formal ad program anyway. The pattern — announce, retreat, relaunch — erodes the trust that an AI assistant fundamentally depends on. Users who survived the December incident and the February launch are priced in for the next encroachment. Users who left are now building habits and workflows on competitor platforms, and habit formation in software is nearly irreversible.

Skeptics of the bearish case will point to OpenAI’s raw scale. Nine hundred million weekly active users is an audience that most advertising platforms would pay billions to access, and even a modest revenue per user would generate substantial returns. The ads pilot reaching $100 million in ARR within six weeks validates demand from advertisers if not from users. And OpenAI’s model improvements — GPT-5.4, Codex, the forthcoming reasoning models — continue to set benchmarks that justify the company’s technical lead. The counter-argument is that technical leadership and monetization strategy are different games, and OpenAI is currently winning one while losing the other. A company can have the best model in the world and still alienate the users it needs to pay for it.

There is also the geopolitical dimension. As this blog covered last week, DeepSeek V4 runs entirely on Huawei Ascend chips at a fraction of the cost of Western models, and China’s open-source competitors are closing the capability gap faster than the investment differential would suggest. If user trust drives the next wave of AI market share — and the QuitGPT movement suggests it does — then OpenAI faces competitive pressure not just from Anthropic’s ad-free positioning but from Chinese models that are both free and unencumbered by the advertising baggage that now shadows ChatGPT’s brand. The window for OpenAI to establish advertising as a durable revenue stream is narrower than the company’s ten-year projections imply.

The fork in the road every AI operator needs to see

The OpenAI-Anthropic revenue inversion is not just a corporate rivalry story. It is a structural signal about how the AI industry will monetize, and every company building on or buying AI products should be adjusting their plans accordingly.

The lesson from the first quarter of 2026 is that AI trust is a depletable asset. Users extend trust to an AI assistant because they believe it is working for them, not for an advertiser. That trust enables deeper engagement, longer conversations, and the kind of data sharing (personal health questions, financial planning, career decisions) that makes AI genuinely useful. The moment advertising enters the conversation, the user’s mental model shifts from “tool that serves me” to “tool that serves someone else.” OpenAI’s user data confirms this shift: ChatGPT’s share of the AI chatbot market dropped fifteen percentage points in two months. Rebuilding that trust will cost more than the ads generated.

Anthropic’s counter-strategy — explicit, public rejection of advertising — is working precisely because it is legible. The Super Bowl campaign did not merely mock OpenAI. It made an implicit promise: Claude will never show you ads. That promise is now a competitive moat, because breaking it would trigger the same backlash OpenAI is experiencing, amplified by the hypocrisy premium. Anthropic has boxed itself into an ad-free model, which constrains future monetization options but strengthens the enterprise sales motion that already accounts for 80 percent of its revenue. The company’s growth from $1 billion ARR in December 2024 to $30 billion in April 2026 — a thirty-fold increase in sixteen months — is the strongest validation yet that enterprise customers will pay premium prices for AI they believe is aligned with their interests rather than an advertiser’s.

The IPO math tells the final chapter. OpenAI is targeting a $1 trillion valuation and has hired Cynthia Gaylor, former CFO of DocuSign, as head of investor relations to prepare the filing. At $25 billion in annualized revenue and $14 billion in projected losses, the company needs a growth narrative that justifies a 40x revenue multiple. Advertising provides that narrative — if investors believe $100 billion in ad revenue by 2030 is plausible. But every month of user decline makes the projection harder to defend, and the IPO timeline creates a perverse urgency: OpenAI needs to demonstrate advertising momentum to justify the valuation, but pushing ads harder accelerates the user attrition that undermines the growth story. It is a feedback loop with no obvious exit.

For operators and decision-makers navigating this moment, the takeaways are concrete:

• Audit your AI vendor stack for monetization alignment. If your organization relies on ChatGPT for customer-facing interactions, the presence of ads in the free tier changes the calculus. Evaluate whether your use case requires ad-free guarantees and whether your contract provides them.
• Watch the enterprise revenue split. Anthropic’s 80-percent enterprise mix versus OpenAI’s 40-plus percent signals where serious business buyers are placing bets. Enterprise AI procurement is a lagging indicator — the contracts being signed this quarter reflect decisions made months ago, which means the shift has been building longer than the revenue numbers suggest.
• Price in the trust premium. The cheapest AI model is not the best AI model if the platform’s incentive structure conflicts with your users’ expectations. ChatGPT’s $60 CPM means OpenAI values each free user’s attention at roughly $60 per thousand interactions. That is the price at which your users’ trust is being sold. Decide whether that trade-off is acceptable for your product.
• Diversify your model dependencies now. The QuitGPT movement proved that platform risk in AI is real and sudden. Companies that built solely on OpenAI’s API faced a brand risk they did not anticipate. Multi-model strategies — running Claude for customer-facing interactions, GPT for internal tooling, open-source models for cost-sensitive workloads — are no longer paranoid over-engineering. They are prudent architecture.
• Follow the revenue, not the benchmarks. Technical superiority and commercial success are decoupling. Anthropic passed OpenAI in revenue while spending four times less on training. The market is telling us that users and enterprises value trust, reliability, and alignment at least as much as they value marginal benchmark improvements. Build accordingly.

The AI industry’s first genuine platform war is being fought not over model quality but over business models. OpenAI bet that users would tolerate ads in exchange for free access to the world’s most capable AI. Anthropic bet that users would pay — or that enterprises would pay on their behalf — for an AI that never commoditized their attention. As of April 2026, the market has rendered its verdict. The company that chose its users is winning.

In other news

Google expands agentic restaurant booking to eight countries — Google’s AI Mode in Search now handles restaurant reservations in the UK, Australia, Canada, India, Singapore, and three additional markets, extending beyond the U.S. launch in August 2025. The feature processes natural language dining requests and books tables directly, part of a broader push into agentic search that will extend to hotel and flight bookings later this year.

Half of American workers now use AI on the job — For the first time in Gallup’s measurement, 50 percent of employed U.S. adults report using AI at work at least a few times a year, with 13 percent using it daily. Leaders adopt far faster than individual contributors — 67 percent of executives use AI weekly versus 46 percent of ICs — highlighting a growing AI fluency gap within organizations.

Meta set to overtake Google in global ad revenue for the first time — eMarketer projects Meta will generate $243 billion in digital ad revenue in 2026 versus Google’s $240 billion, dethroning Google as the world’s largest ad platform. Meta’s 24.1 percent growth rate, fueled by AI-driven ad targeting on Reels and Instagram, outpaces Google’s 11.9 percent as the $1 trillion global digital ad market reshuffles.

Google’s TurboQuant compresses LLM memory sixfold with zero accuracy loss — Presented at ICLR 2026, Google Research’s TurboQuant algorithm reduces KV cache memory for large language model inference to 3-4 bits per element without retraining, achieving roughly 6x compression. On H100 hardware, 4-bit TurboQuant delivers up to 8x speedup on attention computation versus full precision, a breakthrough the internet has already dubbed the “Pied Piper” of AI.

Anthropic’s always-on Conway agent surfaces in code leaks — Evidence of an unreleased persistent agent platform called Conway has appeared in Anthropic’s iOS app code behind hidden feature flags. Conway reportedly runs continuously in the background, responds to webhooks, executes Claude Code, and accepts third-party extensions — positioning it as a 24/7 autonomous worker that goes beyond the session-based interactions current chatbots offer. No public release date has been announced.