OpenAI is using Google’s AI architecture running on Google’s browser infrastructure to compete with Google.
OpenAI launched ChatGPT Atlas on October 21, 2025, during a livestreamed event featuring CEO Sam Altman and Ben Goodger—a former Chrome engineer who helped build Google’s browser. “We think that AI represents a rare once-a-decade opportunity to rethink what a browser can be,” Altman said, directly challenging Chrome’s 15-year reign.
A persistent ChatGPT sidebar follows users across every webpage, automatically understanding context without copy-pasting. Browser Memories let the AI recall sites visited weeks ago and identify patterns. Agent Mode autonomously completes multi-step tasks: booking restaurant reservations, ordering groceries based on recipes, filling out forms, even planning entire vacations across multiple websites.
Atlas is built on Chromium—the same open-source browser engine powering Chrome. This means Atlas inherits Chrome’s rendering speed, web standards compliance, and extension compatibility because Google open-sourced the foundation. It’s available free on macOS with Windows, iOS, and Android versions coming soon. Agent Mode remains exclusive to Plus, Pro, and Business subscribers ($20-30/month), creating a premium tier that could drive OpenAI’s subscription revenue while the free version maximizes adoption among its 800 million weekly ChatGPT users.
“T” is for Transformer
The transformer architecture powering every major AI system today—ChatGPT, GPT-4, Claude, Gemini, DALL-E—was created at Google in 2017. Eight researchers detailed this breakthrough in “Attention is All You Need,” which has since accumulated over 173,000 citations. Yet when Llion Jones, one of those eight authors, explains his work, he says: “No one knows my face or my name, but it takes five seconds to explain: ‘I was on the team that created the T in ChatGPT.'” Not “the T in Google’s products.” The T in ChatGPT.
When we say Google “built” the browser challenging Chrome, we mean it literally. The transformer neural network architecture processing your conversational queries: designed at Google. The browser engine rendering the web pages: coded by Google. The only thing Google didn’t build was the willingness to combine them.
Google had everything needed to build ChatGPT first. By January 2020, the company unveiled Meena, a 2.6 billion parameter conversational chatbot that could “chat about anything” in open-ended exchanges, scoring 79% on sensibleness metrics versus 86% for humans. The team proposed limited public release—exactly the staged approach OpenAI used with GPT-2. Leadership rejected it, citing “AI principles around safety and fairness.”
The cycle repeated in 2021 with LaMDA, a 137-billion-parameter model that exhibited striking contextual intelligence. Engineers Noam Shazeer and Daniel De Freitas successfully integrated it into Google Assistant and proved it was ready for real-world use. Once again, leadership declined to release it. After 21 years at the company, a disillusioned Shazeer left in October 2021 to found Character.AI, which soon achieved a $1 billion valuation. In a twist rich with irony, Google spent $2.7 billion in August 2024 to bring him back.
Every one of the eight researchers who created the transformer architecture eventually left Google. Illia Polosukhin departed in 2017 to found NEAR Protocol. Łukasz Kaiser joined OpenAI in 2021, becoming a key architect of GPT-4 and the o1/o3 reasoning models. Ashish Vaswani and Niki Parmar co-founded Adept AI. Jakob Uszkoreit started Inceptive. Aidan Gomez built Cohere, now valued above $2 billion. Llion Jones launched his own venture. Collectively, their companies are worth more than $4 billion—a diaspora of talent that transformed Google’s foundational breakthrough into a constellation of competitors.
The timeline is damning. Google had Meena in early 2020. OpenAI launched ChatGPT in November 2022—nearly three years later—using an architecture Google invented. Had Google released Meena publicly in 2020, ChatGPT would have been the “Google clone.” Instead, Google found itself declaring “code red” in December 2022, scrambling to respond to a product built on its own research.
A Culture of Caution Suffocated the Advantage of Invention
Multiple factors explain Google’s paralysis, converging on a single theme: the liabilities of incumbency. With $175 billion in annual search advertising revenue and 90%+ search market share, Google had everything to lose.
Bureaucracy compounded conservatism. Google had over 7,000 employees working on AI by 2023, compared to OpenAI’s approximately 150 researchers. That scale created layers of sign-offs, committee approvals, and coordination costs. Multiple executives told The New York Times the company suffered from “paralyzing bureaucracy, a bias toward inaction and a fixation on public perception.” At least 36 vice presidents left between 2020 and 2023, many citing decision-making frustration.
Illia Polosukhin, who left in 2017, explained the dynamic bluntly: “If you want to move really fast and put something in front of a user, Google is a big company with a lot of processes and security protocols. Google doesn’t move unless [an idea is] a billion-dollar business.”
Revenue protection fears paralyzed product launches. Leadership worried that conversational AI would cannibalize the ad-laden search results generating Google’s profits. OpenAI, with no legacy business to defend, could afford aggressive experimentation.
Concerns over reputation eclipsed urgency. Google’s executives hesitated to release a chatbot that might fabricate facts, fearing the reputational fallout of imperfection. Yet when OpenAI introduced ChatGPT—warts and all—users embraced it as an experiment in progress, offering feedback that accelerated its evolution. Google’s pursuit of flawlessness devolved into paralysis.
Sundar Pichai embodied this culture of caution. Colleagues often described him as a leader who prized consensus over speed—a virtue in stable markets, but a liability in an era defined by rapid AI disruption. In the race to define the future, Google’s deliberation became its undoing.
The $100 Billion Demo Failure
On November 30, 2022, OpenAI launched ChatGPT publicly. Within five days it reached 1 million users. Within two months, 100 million—the fastest-growing consumer application in history.
December 2022: “Code Red.” CEO Pichai convened emergency meetings with co-founders Larry Page and Sergey Brin, who had stepped down in 2019 but were called back for crisis management. Multiple teams were reassigned to AI efforts under a 100-day deadline.
February 6, 2023: Bard announcement. Just one day before Microsoft revealed Bing integrated with ChatGPT, Google announced Bard, an “experimental collaborative AI service” powered by LaMDA. The rushed timing was obvious—Google was reacting, not leading.
February 8, 2023: Catastrophic failure. In Bard’s demo video, the chatbot claimed the James Webb Space Telescope took “the very first pictures of a planet outside of our own solar system.” The statement was false—the European Southern Observatory accomplished this in 2004. Alphabet’s stock plummeted 7.7%, erasing $100 billion in market value in a single day. Analyst Gil Luria summarized the market’s verdict: “Google has been scrambling over the last few weeks to catch up on search and that caused the announcement yesterday to be rushed and the embarrassing mess up.”
The disaster crystallized Google’s predicament. Having blocked release of superior chatbot technology for years over quality concerns, the company rushed out an error-filled demo when competitive pressure became unbearable.
March 21, 2023: Bard’s limping launch. Google released Bard to 10,000 “trusted testers” in the US and UK—16 months after ChatGPT’s public debut. User reception was lukewarm.
The company eventually rebranded Bard as Gemini in December 2023, merged Google Brain and DeepMind into a unified organization, and invested billions in AI infrastructure. But the narrative was set: Google as the sleeping giant caught off-guard.
Microsoft Saw Google’s Vulnerability and Struck
In 2019, Microsoft CTO Kevin Scott sent an alarmed email to CEO Satya Nadella and Bill Gates after attempting to replicate Google’s BERT model. “As I dug in to try to understand where all of the capability gaps were between Google and us for model training, I got very, very worried,” Scott wrote. “Even though we had the template for the model, it took us ~6 months to get the model trained because our infrastructure wasn’t up to the task.”
Scott’s fear of Google’s AI dominance directly catalyzed Microsoft’s $1 billion investment in OpenAI in July 2019, which eventually grew to $13 billion by 2023. The irony: Microsoft invested in OpenAI specifically because it feared Google’s technological superiority—a superiority Google then failed to deploy.
Microsoft’s strategic approach contradicted Google’s at every turn. With Bing holding only ~3% search market share, Microsoft had nothing to lose. The company avoided internal bureaucracy by partnering with OpenAI rather than building everything in-house. It launched products despite imperfections, learned from public feedback, and iterated rapidly.
When ChatGPT succeeded, Nadella declared it “a fantastic day” for Microsoft. Within months, ChatGPT was integrated into Bing, Edge, Windows, Office 365, GitHub, and Azure. Meanwhile, Google was hemorrhaging talent, rushing half-baked demos, and watching the technology it invented become synonymous with its competitor’s brand.
Former Google CEO Eric Schmidt’s 2015 comment about Noam Shazeer haunts the narrative: “If there’s anybody I can think of in the world who’s likely to” achieve human-level AI, “it’s going to be him.” Google had that person. Then it blocked his chatbot release. Then he left. Then Google paid $2.7 billion to bring him back.
Chrome Faces Its First Serious Threat—Built on Chrome’s Foundation
The Atlas browser represents OpenAI’s most direct assault on Google’s business model infrastructure. Chrome’s 3 billion users and 65-72% global market share make it the gateway to Google Search, the data collection mechanism for ad targeting, and the distribution channel for Google’s ecosystem. If Atlas captures even 10% of Chrome users, it redirects billions in search queries away from Google’s monetization engine.
Historical precedent suggests disruption is possible. When Google launched Chrome in 2008, Microsoft’s Internet Explorer commanded over 60% market share. Industry observers deemed meaningful competition impossible. Yet Chrome’s superior speed, simplicity, and features steadily captured users. By 2012 it was dominant.
But open source cuts both ways. By making Chromium freely available, Google gave any competitor the ability to build a Chrome-quality browser without the decade of engineering investment. OpenAI simply needed to wrap Chromium in an AI-first interface—exactly the innovation Google could have implemented but didn’t.
This is strategic jujitsu. Atlas inherits Chrome’s speed and compatibility, eliminates the switching friction that typically prevents browser adoption, and adds the differentiator Google was too cautious to deploy aggressively—conversational AI as the primary interface. Users aren’t choosing between incompatible ecosystems. They’re choosing Chrome with traditional search versus Chrome with ChatGPT integrated.
U.S. District Judge Amit Mehta cited this history in his September 2025 antitrust decision allowing Google to keep Chrome rather than forcing divestiture. The judge noted that AI-driven competition from OpenAI and others “already is reshaping the competitive landscape,” making structural remedies unnecessary. Atlas validates that assessment.
The competitive dynamics favor paradigm shift. About 60% of Americans now use AI to find information “at least some of the time,” rising to 74% for those under 30, according to a summer 2025 AP-NORC poll. Atlas users don’t search and click through results—they converse with the browser, which autonomously completes tasks. It’s the difference between using a map and having a driver.
Google’s September 2025 Chrome upgrades demonstrate defensive awareness—integrating Gemini, announcing coming “agentic browsing” features, adding AI Mode in the address bar. But these are incremental enhancements to a fundamentally search-oriented architecture, not the ground-up reimagining that Atlas represents.
The financial stakes are existential. Google’s $175 billion search advertising business depends on users seeing ads alongside results. AI answers that bypass websites eliminate those impressions. If Atlas enters advertising—and analyst Gil Luria calls it a “precursor for OpenAI to start selling ads”—OpenAI gains superior first-party browsing data that could outperform Google’s targeting.
OpenAI’s Path to Profitability Runs Through the Gateway Google Built
For OpenAI, Atlas solves a critical business problem: the company currently loses $5-7 billion annually despite $12 billion in subscription revenue. Atlas provides three monetization vectors that transform OpenAI’s financial outlook.
Near-term: subscription conversion. Atlas’s free tier maximizes adoption among 800 million weekly ChatGPT users, while Agent Mode exclusivity for Plus/Pro/Business subscribers creates upgrade incentives.
Medium-term: advertising foundation. Atlas gives OpenAI unprecedented data access—every page view, click, scroll, search query, and conversational intent signal. This first-party data positions OpenAI to launch a native advertising business by 2027-2028.
Long-term: platform economics. As the browser becomes an AI agent completing transactions—booking travel, ordering food, scheduling appointments—OpenAI can take platform fees on commerce flowing through Atlas.
The privacy tradeoff fuels skepticism. Security expert Simon Willison captured many observers’ concerns: “The security and privacy risks involved here still feel insurmountably high to me—I certainly won’t be trusting any of these products until a bunch of security researchers have given them a very thorough beating.”
OpenAI addresses these concerns with opt-in training data policies, site-specific visibility controls, incognito modes, and user-controllable memory deletion. But the fundamental tension remains: AI-powered browsing requires data access that many users may resist providing.
What Went Wrong Is What Always Goes Wrong with Incumbents
Google’s failure to commercialize transformers fits a pattern Silicon Valley repeatedly demonstrates: inventing the future doesn’t guarantee owning it. Xerox PARC invented the graphical user interface, mouse, and Ethernet, then watched Apple and others commercialize them. Kodak invented digital photography, then went bankrupt as digital cameras destroyed film.
The structural factors that make companies successful—scale, resources, market dominance—become liabilities when paradigms shift. Google’s 7,000 AI employees created coordination costs that 150-person OpenAI avoided. Google’s $175 billion search business created revenue protection fears that OpenAI never faced. Google’s reputation concerns became paralysis when imperfect products proved sufficient for market leadership.
Clayton Christensen called this the “innovator’s dilemma”—incumbent companies rationally serve existing customers and protect profitable businesses, leaving them vulnerable to disruptors targeting emerging markets with initially inferior products. ChatGPT launched with acknowledged limitations. But it was available, and availability beat perfection.
The talent exodus compounds strategic errors. All eight transformer inventors leaving Google represents a failure of organizational culture. Engineers who wanted to build products found Google’s processes intolerable. Illia Polosukhin captured the dynamic: Google’s “high bar for turning ideas into products” meant researchers focused on “career advancement and visibility on research papers” rather than solving user problems.
Leadership matters enormously. Sundar Pichai’s consensus-driven, deliberate style served Google well in stable competitive environments. But in the rapid transformer revolution, that same style meant blocked product launches, lost talent, and strategic whiplash. The $100 billion market cap evaporation after Bard’s demo failure quantifies the cost.
Compare Microsoft’s Satya Nadella, who saw Kevin Scott’s alarmed email and immediately authorized OpenAI investment. Or Sam Altman, who navigated OpenAI’s complex governance structure, survived a board coup attempt, and relentlessly shipped products. Both demonstrated decisiveness that Pichai’s committee culture couldn’t match.
The Browser Google Built, Then Forgot to Launch
Atlas’s debut comes amid a fundamental upheaval in how people access information. The familiar ritual of typing queries, scanning links, and judging credibility feels increasingly antiquated in the age of conversational AI. What once defined efficiency now looks like friction—and when that friction is multiplied across billions of searches, the implications for Google’s business model are existential.
For publishers, the threat is no less dire. When AI systems answer questions directly, users never reach the sites that once relied on search traffic for survival. News outlets, blogs, and review platforms—all cornerstones of the open web—see their audiences siphoned away by AI intermediaries. Google’s search empire depends on this ecosystem; Atlas’s AI-first paradigm does not.
Competition is intensifying. Perplexity launched its Comet browser earlier in 2025, later making it free for all users. Opera has reoriented itself around AI features. The Browser Company introduced Dia. Microsoft’s Edge now embeds Copilot throughout. The AI browser wars have begun—and every major contender runs on transformer architecture, the very technology Google invented.
Even antitrust pressures have twisted into irony. The U.S. Department of Justice’s long-standing scrutiny of Google’s search dominance culminated in Judge Amit Mehta’s September 2025 decision allowing the company to retain Chrome, citing the rise of AI-driven competition as proof that regulation was working. In effect, Google’s own stumbles became its best defense.
The story of ChatGPT Atlas is, at its core, one of squandered opportunity. Google created the transformer architecture that powers today’s AI revolution. It built Chromium, the very engine beneath Atlas. It had the research, the resources, and the lead. What it lacked was the nerve to deploy its own breakthroughs.
Whether this ends as a historical footnote or a strategic catastrophe depends on what Google does next. If Chrome’s Gemini integration delivers comparable AI capabilities and leverages Google’s vast ecosystem, Atlas may remain a compelling curiosity rather than a mass-market shift.
But the opening exists because Google left it. The company invented the future, then smothered it under layers of caution and bureaucracy. Silicon Valley loves to mythologize invention as the hardest part. Google’s story exposes a harder truth: innovation is meaningless without the courage to act.
In technology, being first counts for little if you’re too afraid to move. Google held the future in its hands—and dropped it. OpenAI picked it up, turned it into a browser, and may now own the gateway to the internet.