Signals Inbox·July 28, 2026·Frontier AI

Is Google winning the AI race?

Google is ahead in the broad AI race because it combines strong models with unmatched distribution, cloud growth, custom chips and several ways to make money. It still does not own the best model or the strongest standalone assistant.

Q1What would it actually mean for Google to win the AI race?

Google is winning only if we define the AI race as more than a weekly model leaderboard: it must combine strong models, mass adoption, enterprise spending and workable economics.

A company can briefly own the best benchmark score without building a durable business. It can also reach hundreds of millions of people through default placement while remaining their second choice for serious work. We therefore need to judge Google across four connected contests.

The first is model quality, especially in reasoning, coding and agents. The second is consumer preference: do people deliberately open Gemini, or do they mainly encounter it through Google products? The third is enterprise adoption, where real contracts and production workloads count more than trials. The fourth is economics, including chips, infrastructure costs and the ability to turn AI usage into revenue.

Google currently leads the infrastructure and distribution parts of this race. Its commercial momentum is also unusually strong. Anthropic and OpenAI still hold important advantages in frontier models, coding and the standalone assistant market.

Current AI race scorecard

Q2Why does Google suddenly look like the AI winner?

Google suddenly looks like the AI winner because several businesses that were supposed to be threatened by generative AI are now accelerating together.

Alphabet's latest quarterly results showed Search revenue growing 17%, Google Cloud revenue rising 82% and the Gemini app reaching 950 million monthly users. AI Mode also crossed one billion monthly users, while Gemini APIs processed 22 billion tokens per minute.

The change is the synchronization of those results. Earlier in the AI boom, Google often had one encouraging announcement surrounded by obvious weaknesses. Bard struggled at launch, Microsoft appeared to control the enterprise narrative, and ChatGPT became the consumer brand people associated with AI.

These days, Google is gaining across the whole chain. Search is still growing. Cloud is winning larger workloads. Gemini usage is rising inside and outside Google products. TPUs are becoming an external product rather than just internal infrastructure.

No single number proves that Google has taken the lead. Put together, they show that its AI strategy has moved from defensive experimentation to measurable commercial expansion.

Q3Are Google's AI models the best today?

No. Google does not currently have the best all-round frontier AI model.

Artificial Analysis's current Intelligence Index places Claude Opus 5 at about 61, Claude Fable 5 at 60 and OpenAI's GPT-5.6 Sol at 59. Google's Gemini 3.6 Flash scores around 50, while Gemini 3.1 Pro scores about 46.

That gap is large enough to take seriously. Anthropic and OpenAI currently perform better across the index's mix of professional work, science, reasoning, coding and agentic tasks. Google has also publicly acknowledged that coding and agentic coding need improvement.

Google is moving quickly, though. The company says Gemini 3.6 Flash improved by more than ten points on its Deep SWE coding benchmark compared with the previous Flash model after only six weeks of work. Gemini 3.5 Pro is being tested, and Google has started its largest pretraining run so far for Gemini 4.

Flash competes differently from the most expensive frontier models. Artificial Analysis measures it at roughly 235 output tokens per second, several times faster than the leading Claude and GPT models, with a much lower estimated cost per task. Google is trading some maximum intelligence for speed, scale and cost efficiency.

That can be an excellent business choice. It does not give Google the technical crown today.

Current independent model comparison

Q4Is Gemini actually catching ChatGPT?

Gemini is catching ChatGPT quickly, but ChatGPT still owns the stronger consumer habit today.

Google reported 650 million monthly Gemini users in the third quarter of 2025, 750 million by the end of that year and 950 million in its latest results. That is an increase of roughly 46% across three reported quarters. Daily usage has also tripled over the past year.

Independent traffic data points in the same direction. Similarweb estimated that Gemini's mobile monthly users grew by more than 1,100% between September 2024 and March 2026. ChatGPT grew by roughly 156% during the same period, although it started from a much larger base.

OpenAI still reports more than 900 million weekly ChatGPT users and over 50 million paying consumer subscribers. A weekly user figure is usually more demanding than a monthly one, so comparing Google's 950 million directly with OpenAI's 900 million would flatter Gemini.

Google also benefits from placement inside Android, Search and other products. Some users actively choose Gemini; others reach it because Google has already placed it in their normal workflow. ChatGPT usage more often begins with a deliberate visit to ChatGPT itself.

So this is a real race, not a completed takeover. Gemini has become a mass-market assistant at exceptional speed. ChatGPT still appears to have deeper consumer loyalty and a much stronger record of turning that loyalty into subscriptions.

Q5Is Google winning because AI is already inside Search?

Yes. Google is winning the distribution fight because it can put AI in front of people without asking them to adopt a completely new product.

AI Mode has already passed one billion monthly users since Google expanded it globally. Google has also placed Gemini-powered conversations inside YouTube, where Ask YouTube reached more than 140 million users in one recent month.

The scale comes from ordinary behavior. People search for a restaurant, watch a tutorial, write an email or open a document. Google can insert an AI interaction directly into each of those moments.

OpenAI and Anthropic usually need users to form a new habit around ChatGPT or Claude. Google can start with habits that already exist across Search, Chrome, Android, YouTube and Workspace.

This gives Google more chances to become useful, although it does not guarantee that people will use Gemini for their hardest tasks. Someone may accept an AI summary in Search and later open Claude for coding or ChatGPT for research.

Still, controlling the first interaction is valuable. Google can distribute a new AI feature to hundreds of millions of people faster than any standalone AI company can acquire them one by one.

Q6Is AI helping Google Search or eating its business?

For now, AI is helping Google Search earn more money while changing how people use it.

Google Search and related advertising revenue reached $63.3 billion in the latest quarter, up 17% from a year earlier. That is a strong result for a business that was widely expected to lose queries and advertising demand to chatbots.

Google says AI features are encouraging people to make more searches, including longer and more detailed ones. Those queries also give the advertising system more context about what someone wants.

Advertiser adoption is becoming large enough to measure. Half a million advertisers now use AI Max. According to Google, advertisers using AI Max or Performance Max generate an average of 15% more conversions or conversion value at a similar return on advertising spending.

AAA Auto Club provides a more concrete example. Google says its AI Max campaign increased conversion volume by 17% while reducing the cost per lead by 11%. Google also reported a 20% improvement in the relevance of Shopping ads shown for complex queries.

These are Google's own measurements, so they do not settle the separate argument about publishers receiving less traffic from search results. They do show that Google has found ways to place AI inside its advertising machine without weakening revenue so far.

The immediate threat to Search has been exaggerated. The harder question is what happens if users increasingly prefer complete answers and transactions that leave fewer openings for traditional advertising.

Q7Is Google Cloud taking the AI lead from AWS and Microsoft?

Google Cloud is gaining ground quickly, but AWS and Microsoft still lead the wider cloud market.

Synergy Research Group estimated that Google held 12% of global cloud infrastructure spending at the end of 2024. Its share had risen to 14% by the first quarter of 2026. AWS fell from 30% to 28% during that period, while Microsoft stayed close to 21%.

A two-point gain sounds small until we consider the size of the market. Quarterly cloud infrastructure spending reached about $129 billion in the first quarter of 2026. Two percentage points of that market represent roughly $2.6 billion in quarterly spending.

Google's own results now show even faster growth. Cloud revenue rose 82% to $24.8 billion, while its operating margin increased from 20.7% to 35.6%. Operating income more than tripled to $8.8 billion.

The order book has expanded even faster. Google Cloud's backlog went from $155 billion in the third quarter of 2025 to $240 billion at the end of that year and $514 billion in the latest quarter. It has grown by more than three times in roughly nine months.

Alphabet expects to recognize just over half of that backlog during the next two years. The contracts include infrastructure, Workspace, security, data and other cloud products, so we cannot label the entire amount as AI revenue.

Google is currently winning the cloud growth contest. AWS still has twice its market share, and Microsoft remains comfortably ahead.

How cloud infrastructure shares have moved

Q8Are enterprises really using Gemini at scale?

Enterprise Gemini has clearly moved into production-scale use.

Google says nearly 90% of the Fortune 100 now use Gemini Enterprise. That figure alone remains vague because a company can technically be a customer while running a limited deployment. The consumption numbers are more revealing.

Nearly 500 Google Cloud customers each processed more than one trillion tokens during the past year. One trillion tokens averages about 2.7 billion tokens per day. Across those 500 customers, the minimum implied annual volume exceeds 500 trillion tokens.

More than 2,000 enterprises separately crossed 100 billion tokens. Google also says existing customers are using over 50% more capacity than they originally committed to, while transactions through Google Cloud Marketplace have increased more than sevenfold.

The named deployments cover several kinds of work. HSBC is using Google AI in wealth management, Intel is applying it to internal processes, Bell Canada to customer engagement, Macy's to commerce and SIGNAL IDUNA to knowledge management. PepsiCo is using both AI and analytics services.

Those examples do not prove that Gemini controls each company's entire AI strategy. Large businesses regularly use Google, OpenAI, Anthropic and open models at the same time.

They do show that Gemini is processing large, repeated and paid workloads. The enterprise argument can no longer be dismissed as a collection of trials.

Q9Are developers choosing Gemini over OpenAI and Claude?

Gemini has become a major developer platform, while OpenAI and Anthropic still dominate some of the most valuable use cases.

Google's API volume has climbed from seven billion tokens per minute in the third quarter of 2025 to 16 billion in the previous quarter and 22 billion now. Usage has more than tripled in less than a year.

More than nine million developers build with Google models each month. Its Antigravity agent platform has 2.4 million weekly users, the Agent Development Kit has nearly 70 million downloads and the Gemma open-model family has passed 900 million downloads.

OpenAI remains powerful in coding. It says Codex has 1.6 million weekly users, more than three times the level seen at the beginning of the year. OpenAI also has over nine million paying business users across its products.

Anthropic has turned Claude's coding reputation into unusually strong revenue. Claude Code reached a $2.5 billion annualized revenue run rate earlier this year, while business subscriptions quadrupled. An outside estimate cited by Anthropic suggested that Claude Code was already responsible for 4% of public GitHub commits.

Developers are choosing different tools for different jobs. Claude has a clear pull in coding, OpenAI has the largest general AI platform, and Google is becoming difficult to avoid when a project also needs cloud infrastructure, data, security or Workspace integration.

The old view that developers ignore Gemini is outdated. Calling it their universal first choice would go too far.

Q10Do Google's TPUs create an advantage rivals cannot easily copy?

Yes. TPUs give Google a real cost and supply advantage that OpenAI and Anthropic cannot reproduce on their own.

Google has spent more than a decade developing custom AI chips. Its latest TPU 8t is designed for training and offers nearly three times the raw processing power of the previous generation. TPU 8i focuses on inference and delivers about 80% better performance per dollar than its predecessor, according to Google.

Google can tune the chip, networking, software, model and data center together. This reduces its reliance on Nvidia and gives its engineers more ways to cut the cost of serving products such as Search and Gemini.

The strongest outside validation comes from Anthropic. The company has agreed to use multiple gigawatts of new Google and Broadcom TPU capacity starting in 2027. Anthropic specifically highlighted the price, performance and resilience of Google's infrastructure.

Google has also started delivering complete TPU systems to customer-owned data centers. Alphabet recognized revenue from these sales for the first time in its latest quarter, with most of the existing agreements expected to contribute more heavily in 2027.

Amazon and Microsoft design their own AI chips too, so Google is not alone among large cloud companies. Its position is more unusual among the companies training frontier models: Google owns the lab, the chip, the cloud and the consumer distribution.

That control should make Google cheaper and harder to disrupt, even during periods when Gemini is not the top-ranked model.

Q11Can Google afford the AI race better than everyone else?

Google can fund the AI race longer than most rivals, but the bill is already straining even Alphabet's cash flow.

Alphabet spent $44.9 billion on capital expenditure in its latest quarter, up sharply as it bought servers and built data centers. That spending exceeded the company's $39.1 billion of operating cash flow, leaving free cash flow at negative $5.9 billion.

Full-year capital expenditure is now expected to reach $195 billion to $205 billion. At the midpoint, that would be almost four times Alphabet's $53.3 billion of free cash flow over the previous 12 months.

Alphabet still has an enormous cushion. It ended the quarter with $242.5 billion in cash and marketable securities, while Search, YouTube, subscriptions and Cloud continue to generate operating profit.

The company has nevertheless started using outside financing much more aggressively. It raised $49.6 billion through equity and another $20.3 billion from senior notes during the quarter. Long-term debt has climbed to roughly $98 billion, compared with about $16 billion one year earlier.

That financing decision shows how strongly Google believes in future AI demand. It also shows how expensive this has become.

OpenAI and Anthropic depend more heavily on investors and infrastructure partners. Google can finance a large part of its expansion internally, but it no longer behaves as though internal cash alone will cover every ambition.

The size of Google's AI investment

Q12Is Google winning multimodal AI?

Google currently has the strongest multimodal product system, even though individual image and video rankings still move around.

Gemini can work with text, images, speech and video inside the same model family. Google also controls the places where those media formats naturally appear: YouTube for video, Photos for images, Android for cameras and microphones, and Workspace for documents and meetings.

The company recently introduced Omni, which lets people create content from different kinds of input. Google said daily Gemini users creating videos increased by 40% after the launch.

A video model becomes more useful when people can generate a clip, edit it, add music, publish it and analyze the audience without leaving the same ecosystem. Google can connect Gemini, Veo, Lyria, YouTube and its advertising tools in ways a standalone model provider cannot easily match.

OpenAI and other laboratories can still lead particular image or video evaluations. Creative leaderboards move quickly because users care about different things: realism, control, consistency, speed, editing and price.

Google's advantage comes from breadth and integration. It can turn multimodal models into features across products already used by billions of people rather than waiting for one creative application to build its own audience.

Q13Does DeepMind give Google an edge outside chatbots?

DeepMind gives Google an edge that ordinary chatbot rankings miss.

AlphaFold has been used by more than three million researchers across over 190 countries. Its database contains predictions for more than 200 million protein structures, covering almost every protein known to science.

That work has influenced research into disease, drug discovery, agriculture and plastic recycling. Demis Hassabis and John Jumper also shared the 2024 Nobel Prize in Chemistry for the AlphaFold breakthrough.

DeepMind has since expanded into weather forecasting, materials, mathematics, robotics and genome research. Alphabet's Isomorphic Labs, which grew out of DeepMind's drug-discovery work, recently raised more than $2 billion to develop its platform and clinical pipeline.

None of this makes Gemini the best coding assistant. It does give Google several paths into valuable AI markets that have little to do with consumer chatbots.

A company that leads in scientific discovery could create intellectual property, cloud demand and new businesses that are invisible in today's assistant market shares. DeepMind improves Google's long-term position even when Claude or GPT wins a current model comparison.

Q14Does Google have the strongest AI stack?

Google has the strongest end-to-end AI stack today.

It designs AI chips, operates data centers, trains frontier models, runs a major cloud platform and owns global products that can distribute those models. It can then monetize usage through advertising, subscriptions, cloud consumption and hardware sales.

OpenAI has a stronger standalone consumer brand, although it relies on outside companies for much of its compute. Anthropic has exceptional models and enterprise momentum while running them across Google, Amazon and Nvidia infrastructure.

Microsoft owns cloud, workplace software and operating-system distribution, but much of its frontier-model strategy still depends on external laboratories. Amazon has enormous cloud infrastructure and custom chips, yet it lacks a consumer AI product with ChatGPT- or Gemini-level reach.

Google's layers feed one another. Search and Android provide users. Those users create demand for inference. TPUs lower part of the serving cost. Cloud sells the same infrastructure to external customers. Advertising and subscriptions help pay for the next generation.

A superior stack cannot rescue a consistently poor product. Google's models are close enough to the frontier that the surrounding system becomes decisive.

This is Google's strongest argument for winning the wider race. A competitor may beat Gemini on a model release and still struggle to match everything Google can place around it.

Q15Can regulation break Google's AI distribution advantage?

Regulation can make Google's AI distribution less automatic, but it cannot erase the products Google already owns.

The U.S. Department of Justice has secured restrictions on exclusive distribution agreements involving Google Search, Chrome, Assistant and the Gemini app. Google cannot make the placement of one product dependent on a partner also carrying another Google product.

It also cannot prevent device makers and other partners from distributing competing search engines, browsers or generative AI products at the same time. Compliance with the final judgment is now being monitored.

These rules target one of Google's most useful advantages: the ability to secure default placement and combine several products in one commercial agreement. Competing assistants should find it easier to reach Android users and hardware partners.

The remedies leave Google with Search, Chrome, Android, YouTube, Workspace, Cloud and its existing user relationships. A rival receiving equal permission to appear on a device still has to persuade people to use it.

Regulation will make product quality and user preference more important. It may weaken Google's control over distribution without removing its enormous natural reach.

Q16Is Google turning AI into more revenue than OpenAI and Anthropic?

Google has the largest AI-linked revenue base, although the exact amount is impossible to separate from Search and Cloud.

Google Cloud generated $24.8 billion in the latest quarter, with AI infrastructure and AI products identified as major growth drivers. Search produced $63.3 billion, while Google says AI features are increasing queries and advertising performance. Subscriptions, platforms and devices added another $12.9 billion, partly helped by demand for Google's AI plans.

We cannot add those figures and call the result Google's AI revenue. Most Search revenue would still exist without Gemini, and Cloud includes databases, storage, security, Workspace and many other services.

Anthropic offers a cleaner comparison because nearly all its revenue is tied directly to Claude. The company says its annualized revenue run rate has passed $30 billion, up from about $9 billion at the end of 2025. More than 1,000 customers now spend at least $1 million each on an annualized basis.

OpenAI has disclosed more than 900 million weekly users, 50 million consumer subscribers and nine million paying business users, but it does not publish standard quarterly financial statements.

Google's advantage is the number of places where AI can create value. It earns from customers using Gemini directly, companies renting Google infrastructure and even competitors such as Anthropic buying TPU capacity.

That gives Google more ways to make money from the same AI boom. The downside is that outsiders cannot see precisely which revenue came from AI or whether the return justifies the immense capital spending.

Q17What could still stop Google from winning the AI race?

Google can still lose if it becomes the default AI for easy tasks while competitors own the work customers value most.

Coding is the clearest warning. Google has admitted that it needs to improve agentic coding, while Claude Code already generates billions of dollars in annualized revenue. OpenAI's Codex is also building a large weekly audience.

Consumer loyalty creates another risk. ChatGPT has more than 50 million paying subscribers. Google may reach more people through Search while converting fewer of them into customers who actively prefer Gemini and pay for it.

The infrastructure bill leaves less room for mistakes. Alphabet is planning around $200 billion of capital spending in one year. If model prices fall faster than demand grows, the return on those data centers could disappoint.

Large companies may also keep several AI suppliers. They can use Gemini in Workspace, Claude for coding, OpenAI for agents and open models for sensitive internal systems. Google could build a huge business in that market without achieving the control it once enjoyed in web search.

The final threat is execution. Google has more products, teams and interfaces than its younger competitors. That breadth creates opportunities, but it can also produce confusing names, overlapping tools and slower decisions. We have seen Google do this before.

Google's current position is powerful. Victory still depends on turning its scale into products people choose for important work.

Q18Is Google winning the AI race?

Mostly yes. Google is currently winning the AI race as a business, but it has not won the contest for the best model or the strongest standalone assistant.

Gemini has reached 950 million monthly users, AI Mode has crossed one billion, and Google Cloud is growing by 82%. Those results show that Google can turn AI into search activity, advertising performance, enterprise contracts and infrastructure demand at the same time.

No rival has an equally complete combination of chips, data centers, frontier research, cloud services, consumer distribution and existing monetization. That gives Google the strongest position if the AI race lasts many years and requires hundreds of billions of dollars in infrastructure.

Anthropic currently leads several frontier model rankings and has built an exceptional coding business. OpenAI retains the strongest standalone consumer habit and the largest disclosed base of paying AI subscribers. Those are substantial leads.

Google's advantage lies elsewhere. It can finish second in a model comparison and still place that model in front of more users, serve it more cheaply and earn revenue through more channels.

The fairest judgment is that Google is ahead in the broader race while the most visible individual races remain open. The model crown still moves between laboratories. The commercial machine around those models is increasingly tilting toward Google.