Google’s Gemini app is now close to a billion monthly active users. In Alphabet’s second-quarter 2026 remarks, chief executive Sundar Pichai put the figure at more than 950 million, up from 750 million in the previous quarter. He also said AI Mode in Search had passed one billion monthly active users, developers were processing more than 22 billion tokens a minute through Gemini APIs, and customers had created more than 90 million custom “Gems”.
Those numbers make Gemini one of the largest consumer AI products in the world. They do not, however, mean that 950 million people pay for Gemini, use it daily, or have abandoned Google Search for a chatbot. Each metric measures a different layer of Google’s AI distribution system. The useful question is not simply “How many users does Gemini have?” It is what those users are doing, how Google reaches them and whether scale is turning into durable behaviour.
Alphabet’s Q2 AI scale dashboard
| Metric disclosed for Q2 2026 | Reported scale | What it measures | What it does not prove |
|---|---|---|---|
| Gemini app monthly active users | More than 950 million | People active in the Gemini app during a month | Daily use, paid subscribers or task completion |
| AI Mode monthly active users | More than 1 billion | Reach of Google’s conversational Search experience | Exclusive use or replacement of standard Search |
| Gemini API throughput | More than 22 billion tokens per minute | Developer and enterprise processing volume | Number of unique developers or economic value per token |
| Custom Gems created | More than 90 million | User-created specialised assistants | Continued use, quality or business adoption |
| Google Search revenue growth | 17% year on year | Commercial strength of Search and related advertising | Revenue caused specifically by AI Mode |
| Google Cloud revenue growth | 82% year on year | Demand across cloud infrastructure, platform and AI services | Gemini-only revenue |
The figures come from Alphabet’s official Q2 2026 remarks. They are best read together. The app number shows consumer reach; AI Mode shows distribution through Search; token throughput shows the developer layer; Search and Cloud growth show the commercial context.
The 950 million figure is large because distribution is large
Standalone AI products usually need people to discover a new service, create an account and form a new habit. Google can introduce Gemini through products that already sit inside daily routines: Android, Search, Chrome, Workspace and the Gemini app itself. That distribution advantage matters as much as model quality.
A user may encounter Gemini while summarising a long email, asking a follow-up question in Search, generating an image on a phone or using a Workspace feature. These interactions can all strengthen the ecosystem without creating the same kind of engagement. A monthly active user who opens the app once is not equivalent to a professional who uses Gemini every working day. Alphabet did not disclose that split in the remarks.
The quarter-on-quarter change is still meaningful. Moving from 750 million to more than 950 million implies an increase of at least 200 million monthly users in one quarter. That is not a small-product growth curve. It indicates that Google has found multiple distribution channels capable of putting Gemini in front of mainstream users rather than only AI enthusiasts.
The strongest interpretation is therefore about reach: Gemini is no longer an experimental companion to Search. It is a mass-market interface woven through Google’s consumer and enterprise estate.
AI Mode is not the same audience as the Gemini app
AI Mode’s one-billion-user figure can easily be added to the Gemini app figure to produce a misleading headline. The audiences overlap. Someone who uses Gemini on Android may also encounter AI Mode in Search. Alphabet did not provide a deduplicated total.
The two products also answer different needs. The Gemini app is an open-ended assistant: users can draft, analyse, create, plan and maintain longer conversations. AI Mode is attached to information seeking. It helps people explore a query through follow-up questions and synthesised answers while retaining links to the web.
That distinction matters for publishers and brands. AI Mode can change how people reach source pages, but it does not eliminate the need for credible information. When an AI answer cites a product comparison, scientific source or first-party announcement, the underlying page still has to be useful enough to earn selection and clicks. In practice, the growth of AI Mode raises the standard for original evidence and clear value objects rather than rewarding generic summaries.
For users, the key difference is intent. Use Search or AI Mode when the task depends on current web information and source discovery. Use the Gemini app when the work involves a continuing context, a document, an image, a personal workflow or a custom Gem. Treating both as “the same chatbot” hides why Google is operating more than one AI surface.
What 22 billion tokens per minute tells developers
Token throughput is a measure of activity across Gemini APIs. At more than 22 billion tokens per minute, it signals that the model family is supporting substantial machine-to-machine and enterprise workloads. Tokens can represent input, output, cached context and internal application traffic; the figure is not a count of prompts.
High throughput has three implications. First, Gemini is being used beyond the consumer app. Businesses and developers are embedding the models into software, data workflows and agents. Second, Google needs enormous serving capacity. The economics of inference—chips, data centres, power, networking and model efficiency—become as important as benchmark scores. Third, small changes in price, latency or token efficiency can have large effects when multiplied across this volume.
The metric does not tell buyers which Gemini model to choose. A fast, low-cost model may account for many more tokens than a reasoning-heavy model while producing less revenue per request. Developers still need workload tests covering accuracy, latency, context, tool use, failure modes and total cost. Our Kimi K3, Gemini 3.6 Flash and GPT-5.6 Sol comparison uses that decision-based approach rather than treating one benchmark as a universal answer.
Ninety million Gems show experimentation, not retention
Gems are custom versions of Gemini configured for a recurring purpose. A user might build one for meeting preparation, travel research, editorial review or a company-specific style guide. More than 90 million created Gems demonstrate a low barrier to experimentation.
Creation is only the first stage. The important unanswered measures are how many Gems are used repeatedly, how many are shared, what percentage connect to external tools and whether they save measurable time. A gallery full of one-off experiments is different from a set of trusted assistants embedded in work.
Anyone building a Gem should apply a simple retention test:
Name one repeatable task, not a broad role such as “be my assistant”.
Define the approved inputs and data boundary.
Specify the expected output format.
Add a source or verification requirement where factual accuracy matters.
Measure whether the Gem is used again within seven days.
If it fails the final test, the configuration may be interesting but not valuable. Alphabet’s creation number shows appetite; recurring use will show whether custom assistants become a durable interface.
Search and Cloud growth make the adoption story more credible
Alphabet reported 17% year-on-year growth in Google Search revenue and 82% growth in Google Cloud revenue. The numbers matter because they place AI adoption inside businesses that are growing, not inside a separate experiment funded without commercial support.
Search growth suggests that AI experiences have not automatically destroyed the existing advertising engine. It does not prove that AI Mode caused the growth. Advertising demand, query volume, pricing and product changes all contribute. Cloud growth likewise includes infrastructure, platform and software revenue beyond Gemini.
The useful conclusion is narrower: Alphabet is scaling consumer AI, developer APIs and infrastructure while its two most relevant commercial engines remain strong. That gives Google more freedom to invest aggressively, serve expensive workloads and distribute AI through products with established revenue.
This is also why Alphabet’s higher capital-spending plan deserves its own analysis. The user figures describe demand; the 2026 AI spending plan describes the supply needed to serve it.
Five questions the headline numbers cannot answer
The Q2 disclosure leaves several questions open:
Engagement: How many Gemini app users return daily or weekly?
Monetisation: How many pay directly, and how much revenue comes from bundled subscriptions?
Overlap: How much duplication exists between Gemini app, AI Mode, Android and Workspace audiences?
Quality: Are task-completion and user-satisfaction measures improving as reach expands?
Economics: How quickly are model efficiency and pricing offsetting the cost of inference?
These are not reasons to dismiss the disclosed scale. They are the next metrics needed to distinguish broad exposure from deep adoption.
What the numbers mean for users and businesses
For consumers, Gemini’s scale means faster product integration and a larger ecosystem of tutorials, extensions and compatible services. It also increases the importance of checking account settings, activity retention and connected-app permissions. A mass-market assistant can touch email, files, browsing and device functions; convenience should not become invisible access.
For developers, the API figure confirms a mature distribution channel but raises the need for model portability. A production workflow should separate prompts, tools, retrieval and evaluation from one provider’s interface where possible. High adoption is not a reason to skip fallback plans, cost controls or data classification.
For organisations, the most useful signal is not the consumer headline. It is the combination of app reach, API throughput, Search distribution and Cloud growth. Google can place Gemini in front of employees and customers through several routes. Governance therefore needs to cover both centrally purchased AI and consumer tools that appear inside existing accounts.
The verdict
Gemini’s 950-million-plus monthly audience is a genuine scale milestone. It shows that Google has moved AI from a specialist product into a mainstream consumer layer. AI Mode’s one-billion-user reach and the API’s 22-billion-token-per-minute throughput make the story broader than a single app.
The figures should not be turned into claims Alphabet did not make. They do not identify daily users, paying customers, deduplicated reach or Gemini-specific revenue. Read correctly, they show a company with unusual distribution, accelerating adoption and the infrastructure challenge of serving AI at Google scale. The next decisive evidence will be retention, task completion, monetisation and cost—not another larger monthly-user headline.
