Google Reveals 350M Monthly Gemini Users, Trails ChatGPT
During day three of the U.S. Department of Justice’s antitrust remedies trial, Google revealed that its flagship large language model, Gemini, has surged to 350 million monthly active users as of March 2025. The figure was presented on a slide by Sissie Hsiao, former head of Google AI’s consumer products, highlighting Google’s rapid user growth even as it continues to lag behind OpenAI’s ChatGPT.
Background and Model Evolution
Gemini debuted in late 2023 built on Google’s next-generation Pathways infrastructure. It originally shipped at tens of billions of parameters, but the rollout of Gemini 2.0 in late 2024 introduced a 120-billion parameter model with low-latency inference optimizations. Earlier this year, Google launched Gemini 2.5, which employs a mixture-of-experts (MoE) routing layer to dynamically allocate compute, reducing inference compute by up to 30% for simple queries while ramping up to full capacity on complex tasks.
Key technical specifications of Gemini 2.5 include:
- Parameter count: 140B (base) + MoE experts
- Context window: 128k tokens
- Average inference latency: 120ms (on TPU v4 Pods)
- Throughput: 200 tokens/sec per TPU v4 chip
- Support for retrieval-augmented generation (RAG) via Vertex AI
Deployment and Scalability
Google has progressively integrated Gemini across its ecosystem, including Search Labs, Google Workspace (Docs and Gmail smart compose), and Bard in Android 15 beta. These services run on Google Cloud’s TPU v4 pods and leverage Kubernetes autoscaling on GKE to handle unpredictable demand. According to internal benchmarks, scaling from 50 to 200 million daily queries requires spinning up an additional 10 TPU pods within 30 seconds.
Comparison with ChatGPT
Despite Gemini’s growth, Google’s own analysis estimates OpenAI’s ChatGPT at approximately 600 million monthly active users as of Q1 2025—up from around 400 million earlier in the year. OpenAI’s recent release of GPT-4 Turbo with a 128k token context window and more efficient quantization has further boosted ChatGPT’s appeal for enterprise API usage.
Variations in traffic measurements complicate direct comparisons:
- Weekly active users: OpenAI cites up to 400 million (selected 7-day windows).
- Monthly active users: More conservative metric, trusted in legal filings.
- API vs. consumer interface: Google’s slide counts both Bard and API calls, whereas OpenAI may emphasize web usage.
Economics of Generative AI
Generative AI services are compute-intensive and incur substantial costs. Current industry estimates place inference costs around $0.0006 per 1,000 tokens on TPU v4, while GPU-based providers report $0.001–$0.002 per 1,000 tokens on A100 hardware. Both Google and OpenAI acknowledge negative margins for high-volume tiers:
- OpenAI’s $200/month “Pro” plan loses an estimated $50–$80 per user in compute spend.
- Google has not publicly detailed Gemini subscription economics but internal memos suggest sub-$0.10 revenue per user per month relative to infrastructure costs.
Antitrust Implications and Legal Strategy
These usage figures play a pivotal role in Google’s defense against DOJ claims that Alphabet unfairly bundles its AI services with core products. By showcasing rapid adoption, Google argues that users freely choose Gemini over competitors. However, regulators may scrutinize deep integrations—such as default Bard prompts in Search—as potential barriers to entry for rivals.
Expert Perspectives and Future Outlook
AI industry observers offer varied takes on Google’s trajectory. Dr. Emily Bender (University of Washington) notes, “Gemini’s large context window and MoE design are impressive, but true benchmarks are still emerging against GPT-4 Turbo in zero-shot tasks.” Meanwhile, former OpenAI researcher Andrej Karpathy highlights that “model efficiency and ecosystem lock-in will determine which platform dominates the next wave of AI applications.”
Looking ahead, Google plans to introduce a specialized Gemini Nano model for on-device inference in Pixel 9, aiming to reduce latency and address privacy concerns. The competition with OpenAI and Microsoft, which has embedded GPT-4 in Office 365, promises continued innovation in model architectures, cost optimizations, and regulatory scrutiny.