Alphabet said on October 28, 2026, that its Gemini app had passed one billion monthly active users, up from 650 million a year earlier, and that its first-party model APIs were processing nearly 32 billion tokens a minute, 40% more than three months earlier and more than four times the level of a year ago. On the same day, Google announced Gemini 4 Carbon, which chief executive Sundar Pichai called its strongest frontier model, and restricted it at first to a group of trusted cyber defenders. Pichai said Google is bringing agentic features to Search next and described its approach to AI in Search as building the experience first, "followed by monetization." Shared AI research costs rose 71% to $7.7 billion in the quarter.

In Short

Google's AI chatbot app, Gemini, is now used by more than a billion people a month, and companies that plug Google's AI models into their own software are using them four times as much as a year ago. Google also released a new, more capable model that it is giving first to cybersecurity defenders, and said its AI helpers will soon do tasks like bookings and purchases inside Google Search. For marketers, the key line is that Google says it is building these AI features first and working out how to make money from them afterwards.

The AI figures in Alphabet's results

Alphabet published its third-quarter results on October 28 for the quarter ended September 30, 2026, together with an earnings presentation, and Pichai posted an edited transcript of his prepared remarks for the analyst call. The table below sets the AI usage figures against those PPC Land reported in earlier quarters.

Metric Q3 2025 Q4 2025 Q1 2026 Q2 2026 Q3 2026
Gemini app monthly active users 650m 750m+ not stated 950m 1bn+
First-party API tokens per minute 7bn 10bn+ 16bn 22bn nearly 32bn

The earlier figures come from PPC Land's reports on the third quarter of 2025, the fourth quarter, the first quarter of 2026 and the second.

Gemini app: past one billion

The earnings release says the Gemini app "has reached one billion users," and calls it the fastest-growing product in Google's history. The earnings presentation is more specific, listing "1B+" Gemini app monthly active users. The increase from 950 million in July is about 50 million, or 5%, in three months. Over twelve months, the app added about 350 million monthly users, a rise of 54%, PPC Land's calculation.

"People are already using the Gemini app to do more than ask questions; they're using it to do things, including complex, multi-step agentic queries," Pichai said.

Outside data points to the same direction of travel in the United States. Comscore's AI Intelligence Report, released on September 22, put Gemini's share of US prompt volume across seven assistants at 30% in June 2026, up from 17% in January, while ChatGPT fell from 70% to 50%. Comscore did not disclose panel sizes or margins of error for those series.

Access to the strongest models in the app is limited by plan. From October 9, free Gemini accounts kept only Flash-Lite, losing Flash and Pro; Google gave no reason in its help documentation. Gemini 4 Argon, released on October 13, is available to consumers only on the paid AI Pro and Ultra plans, in the Gemini app and in AI Mode.

The earnings documents give no figures for paid AI subscribers, for revenue from the Gemini app, or for daily use. They do not mention advertising in the Gemini app.

32 billion tokens a minute

Alphabet said its first-party model APIs are "now processing nearly 32 billion tokens per minute," which its presentation describes as up 40% quarter on quarter and measured "across 1P models, including Gemini via direct API usage." That is the volume developers and companies send to Google's models through its programming interfaces, not usage inside Google's own consumer products.

The growth has been steep and steady. PPC Land reported 7 billion tokens a minute in October 2025, more than 10 billion in February, 16 billion in April and 22 billion in July. At 32 billion a minute, the API alone would process about 1.38 quadrillion tokens in a 30-day month, PPC Land's calculation, slightly more than the 1.3 quadrillion monthly tokens Alphabet reported in October 2025 across all of its surfaces combined.

Argon contributed. "On our developer platforms, usage in the first week ramped up three-and-a-half times faster than Gemini 3, even with capacity constraints," Pichai said. In an interview on October 21, he said Argon processed almost 25 trillion tokens in its first 72 hours, about three times Gemini 3.0 over the same period, and that Google remained capacity-constrained. Google priced Argon at an introductory $2 per million input tokens and $10 per million output tokens.

Gemini 4 Argon and Carbon

Pichai's remarks describe Argon as strong "in areas like finance, legal, cybersecurity, and science." He said developers and Cloud customers are using it "for things like untangling legacy codebases, accelerating scientific research, robotics, discovering product improvements, and domain-specific tasks requiring in-depth reasoning."

Google's own teams are among the users. "And at Google, just in the last few weeks, teams across Chrome, Ads, and Search have used Argon to write and optimize code much faster, shipping 35% more code with 30% fewer errors," Pichai said. He gave no baseline or time window for the comparison. When Argon launched, Google claimed over 35% more agentic code submissions in its Antigravity tool, also without a baseline.

Carbon

Carbon, announced on October 28, is the second frontier model in the Gemini 4 series. Pichai called it "our strongest frontier model, with state-of-the-art cybersecurity and long horizon capabilities" and said it ranks first on the recently updated Artificial Analysis Intelligence Index with a score of 59. "Carbon outperforms Argon across almost every domain, like full-stack web development, multimodal understanding, complex knowledge work, and scientific reasoning," he said. The ranking and the comparisons are Google's; the remarks do not name the models Carbon was measured against.

Access is restricted. "We're rolling Carbon out first to a subset of trusted cyber defenders before we expand access, and it's being tested by the U.S. government," Pichai said. He gave no date for wider availability or pricing.

The restriction follows a pattern in Google's published safety rules. Version 3.1 of Google's Frontier Safety Framework, published in April, raised recommended security to level 2+ for models reaching a critical capability level for cyber uplift, among other domains. At Argon's launch, Google said partners in its Fairwind Program could use a version without cyber guardrails.

The release pace

Alphabet's presentation lists the quarter's releases from July to October: Gemini 3.6 Flash and 3.5 Flash Cyber, Gemini 3.7 Flash, Gemini 3.8 Flash and 3.8 Flash Cyber, Gemini 3.5 Transcribe, Omni 1.1 Flash, Lyria 3.5, Gemini 3.8 Live and Extended Thinking, Nano Banana 2.1, Gemini 4 Argon and Gemini 4 Carbon. On October 27, the day before the results, Google announced Genie 4, which Pichai called "our most advanced world model yet that simulates safe, realistic, interactive worlds with native audio in real time," and "an important building block for training robots and agents."

Pichai said more is coming. "We have more models coming soon, including more efficient versions to help developers optimize for speed and cost and to deploy across our billions of users." Smaller versions of this kind are cheaper to run at inference, the stage at which a trained model answers requests. In his October 21 interview, he said distilled Gemini 4 versions were planned, without a date.

He also linked the pace to AI itself. "We're at the point where AI is helping us increase our training and launch velocity. I would expect the next 12 months to see continued acceleration because of RSI." The remarks do not define RSI, an abbreviation commonly used for recursive self-improvement, in which AI systems contribute to building their successors.

AI in Search: experience first

For advertisers, the most consequential AI passage in the remarks concerns Search. "Search is bringing the benefits of AI to more people than any other product in the world," Pichai said. "Just like our shift to mobile over a decade ago, we are focused on building the best Search user experience, followed by monetization."

He described the product as a single surface. "We are building one seamless experience across AI Overviews, AI Mode, and traditional Search. And people are using our AI features more, searching in entirely new ways. Their queries are longer, personalized, and multimodal." Queries in AI Mode are on average three times longer than traditional searches, he said, and one in five AI Mode searches globally use voice, images or documents. Google has crossed 50 billion visual searches per month.

Alphabet credits AI with lifting Search usage. "Our AI investments are driving overall Search query growth, including commercial queries," Pichai said in the release, adding that year-over-year query growth was the highest in five years. Search & other revenue grew 12% to $63.35 billion, against 17% in the second quarter, and was $79 million higher than in the second quarter. The documents give no query counts, paid clicks or cost-per-click figures that would show how AI queries are monetized.

Ads already run inside AI surfaces. Google formally announced shopping ads in AI Mode on February 11, and said in May that AI Mode had passed 1 billion monthly active users. In July, Alphabet said half a million advertisers were using AI Max. The third-quarter documents update none of these figures.

Pichai devoted a section of his remarks to agentic AI, in which systems carry out tasks on a person's behalf. "One area I'm excited about is helping people get things done, be it booking reservations, filling in forms, scheduling appointments, or making purchases," he said. "As AI becomes more personal and starts taking action for people and businesses, trust and security matter more than ever."

Google started with business customers. "We focused first on getting these agentic experiences right with the Gemini agent for enterprise because the bar for security, safety, and governance is extremely high," Pichai said. "As a next step, we're bringing these features to Search. And also, of course, to our billion users on the Gemini app."

The groundwork in Search is already visible. At I/O on May 19, Google introduced information agents in Search that monitor the web for changes on a question a user defines, first for AI Pro and Ultra subscribers. On June 12, they became available to Ultra subscribers in every AI Mode language and market. On October 15, Google began rolling out Universal Cart in US Search, including AI Mode, and in the Gemini app, with checkout through Google Pay or a handoff to merchants through the Universal Commerce Protocol. The tasks Pichai named in his remarks, bookings, forms, appointments and purchases, go further than monitoring.

Pichai described agents as the layer that ties Google's products together. "Agents will operate as a unifying layer across our products, so you'll see us extend these agentic experiences beyond Search and the Gemini app to Chrome, Android, and our Workspace products as well."

The remarks do not explain how advertising would work when an agent books or buys on a user's behalf. In his October 21 interview, Pichai said the long-term answer was to put users first while making the arrangement work for businesses, and did not address ads directly.

What AI costs Alphabet

The release separates the cost of Google's general AI models from its business segments. "Alphabet-level activities," which "primarily reflect expenses related to our shared AI research and development," cost $7,655 million in the third quarter, up 71% from $4,467 million. The release says these include employee compensation and "technical infrastructure usage costs associated with the development of our general AI models." Research and development across the company rose 36% to $20,640 million.

Infrastructure is the larger bill. Purchases of property and equipment reached $51,355 million, up 114%, and free cash flow was negative $9,065 million for the second quarter in a row. In July, Alphabet raised its 2026 capital spending guidance to $195 billion to $205 billion; the October documents do not update it. Depreciation rose 49% to $8,369 million.

"We are working hard to bring more capacity online to meet this long-term demand," Pichai said of Cloud.

TPUs as a product

Google's own AI chips are now sold as well as used internally. The release describes Google Cloud as generating "product revenues primarily from the sale of TPU systems," and says "TPU systems sales were also a meaningful contributor to revenue growth" without giving a figure. Google Cloud revenue rose 103% to $30,709 million, which the release attributes to demand for AI infrastructure and AI solutions as well as core Google Cloud Platform services, and Cloud operating income rose to $11,121 million from $3,594 million.

The earnings presentation sets out what Google sells: trays with four proprietary chips on one liquid-cooled board, racks of 16 TPU machines with battery backup, pods of 144 TPU racks with three optical switch racks, optical circuit switches and high-bandwidth memory, supported by JAX, PyTorch and the XLA compiler, the vLLM and SGLang serving engines, and post-sales support and warranty. "We're seeing strong demand for our TPU systems across a wide range of customers, and we expect that to continue as we scale," Pichai said.

One of the largest customers is visible from outside. Anthropic's filing for its initial public offering shows at least $114.5 billion of compute commitments to Google Cloud running from April 2026 to July 2033. In August, UBS estimates cited by Ed Zitron projected that OpenAI and Anthropic would account for more than 48% of Google Cloud revenue in 2027, a projection rather than a reported figure.

Pichai also said Google had "launched our TPUs into space with Project Suncatcher." When Google announced the project in November 2025, it planned to launch two prototype satellites with Planet in early 2027.

AI beyond Google's products

Pichai cited other uses of the models. Isomorphic Labs is using Argon to support drug design research. Google demonstrated "magic state cultivation" in quantum computing for the first time. Waymo, which he said has driven more than 300 million fully autonomous miles and operates in 15 major US cities, sits in Other Bets, whose operating loss widened to $2,047 million.

Why this matters for the marketing community

The AI figures in Alphabet's results are usage figures: users, tokens, query length, visual searches. The revenue figures attached to AI are in Cloud, where AI infrastructure, AI solutions and TPU systems drove growth of 103%. For advertising, the documents offer intent rather than numbers. Pichai's statement that Search is following the mobile pattern, "building the best Search user experience, followed by monetization," places ad products behind the build-out of AI Mode and agents in the order of priorities.

Three developments in the remarks bear on how marketers reach people. The Gemini app, now past a billion monthly users, has no advertising mentioned in the documents, while OpenAI said on October 21 that ChatGPT ads were available in 89 countries. Agents in Search would complete tasks that currently begin with a click on an ad or a result, and Google has not said how ads or merchant listings would be chosen when an agent acts. And the code behind Google's ad systems is now being written with Argon, which Pichai said lets teams including Ads ship more code with fewer errors.

What the documents do not say

  • Advertising in the Gemini app or in agentic tasks.
  • AI Mode and AI Overviews user figures, last given as more than 1 billion monthly users for AI Mode in May.
  • Paid AI subscribers on the AI Pro and Ultra plans.
  • Revenue from TPU systems or from Gemini API usage.
  • Carbon's pricing and wider availability, or the models it was compared against.
  • Updated capital spending guidance.

Timeline

Summary

Who: Alphabet Inc. and Google, with chief executive Sundar Pichai quoted in the earnings release and his published call remarks. The developments affect advertisers and publishers who depend on Google Search, developers and companies using Gemini models, and Google Cloud customers buying AI infrastructure and TPU systems.

What: The Gemini app passed one billion monthly active users, up from 650 million a year earlier. First-party model APIs processed nearly 32 billion tokens a minute, up 40% in a quarter. Google announced Gemini 4 Carbon and restricted it at first to trusted cyber defenders. Pichai said agentic features are coming to Search, and that Google is building the best Search experience "followed by monetization." Shared AI research costs rose 71% to $7.7 billion, capital spending reached $51.4 billion, and TPU systems sales contributed to Cloud revenue growth of 103%.

When: Results and the Carbon announcement came on October 28, 2026, for the quarter ended September 30, 2026. Genie 4 was announced on October 27 and Argon released on October 13.

Where: Alphabet is based in Mountain View, California. The figures are global; Universal Cart and the earliest Search agents launched in the United States.

Why: Google is building AI usage at scale across its app, its APIs and Search, and selling the infrastructure behind it through Cloud. For marketers, the results show usage growing faster than the disclosed means of monetizing it: the documents describe agents and AI search experiences in detail, but say nothing about how advertising fits into them.