Data shows that Google is narrowing the gap with OpenAI in multiple dimensions. Gemini was downloaded 100.8 million times in November, compared with 67.8 million for ChatGPT. Users now spend more time chatting on Gemini than chatbot competitors like ChatGPT or Claude. In the two weeks since the release of Google Gemini 3, the seven-day average number of daily unique active users of ChatGPT has dropped by 6%.
One of the big news in the artificial intelligence (AI) field this week is that OpenAI CEO Sam Altman announced to all employees on Monday the launch of a "red alert" to focus all resources on optimizing ChatGPT to deal with the fierce competition from Google Gemini. This strategic adjustment reflects the profound changes in the AI competitive landscape and also reveals the potential threat to the dominance of Nvidia’s chips by Google’s self-developed chip TPU.
Media reports stated that OpenAI decided to postpone the development of other products including advertising business, health and shopping AI agents and personal assistant Pulse, and reallocate core resources to improving the daily use experience of ChatGPT. Altman said that OpenAI still needs to improve the daily experience of ChatGPT, including improving personalization capabilities, speed and reliability, and expanding the range of questions that can be answered.
UBS technology analyst Tim Arcuri pointed out in the latest research report that Google's new generation TPU chip Ironwood and its TPU ecosystem are posing a substantial challenge to Nvidia. Nvidia's stock price performance has significantly lagged behind Google.
Google user time exceeds the limit, ChatGPT daily activity declines
Market data shows that Google is narrowing the gap with OpenAI in multiple dimensions. According to Sensor Tower data, monthly downloads of Gemini reached 100.8 million times in November, while ChatGPT had 67.8 million downloads.
Even more notably, users now spend more time chatting on Gemini than chatbot competitors like ChatGPT or Claude.
According to Deedy Das statistics, in the two weeks since the release of Google Gemini 3, the number of daily unique active users of ChatGPT (seven-day average) has dropped by 6%, showing the direct impact of competitive pressure. Although OpenAI still has more than 800 million weekly active users and dominates overall chatbot usage, users are losing to Google.
Nick Turley, OpenAI's director of ChatGPT, posted on social media on Monday night that search is one of the biggest areas of opportunity. ChatGPT currently accounts for about 10% of global search activity and is growing rapidly.
He also said the company is focused on making ChatGPT stronger, continuing to grow and expand global access, while making it more intuitive and personal.
UBS: Google TPU chips pose a threat to Nvidia
Behind the competition in AI models, the competition at the chip level is equally fierce. UBS technology analyst Tim Arcuri pointed out in a research report that the advancement of Google TPU chips is changing the market structure.
According to Arcuri analysis, Google first disclosed the latest generation TPU chip Ironwood in April this year and officially launched it in November. The chip is optimized for large language models (LLM), mixture of experts (MoE) and advanced inference, supporting training, fine-tuning and inference workloads, in contrast to the narrow customization of previous TPUs.
Ironwood has not yet been submitted to MLCommons' MLPerf v5.1 data center training benchmark, but Arcuri expects its single-chip performance to significantly exceed Trillium's given more computing resources, FP8 support and far higher bandwidth memory than its predecessor.
Arcuri pointed out that Google's previous generation Trillium chip was specifically optimized for inference workloads and had a lower HBM capacity (32GB vs 95GB). In contrast, Ironwood has more computing resources, FP8 support and a significantly increased HBM capacity, and is expected to significantly surpass Trillium in single-chip performance. Ironwood also expanded the TPU scale to a maximum of 9216 TPU domains, far exceeding v5p's 8960 and Trillium's 256.
Arcuri pointed out that this is why Nvidia's entire ecosystem is significantly lagging behind Google, which is enjoying the surge in attention brought by TPU products. Koray Kavukcuoglu, chief technology officer of Google DeepMind, said that by using Google’s own custom chips to train AI models, the company has “significantly improved performance.”
UBS believes that while Google may consider expanding the TPU ecosystem over time, any such efforts must limit the potential cannibalization of Google Cloud Platform (GCP) revenue. From this perspective, both Meta and Apple are prime candidates for in-house TPU deployments because they have large AI projects supporting internal workloads, large internal AI clusters, and relatively little dependence on GCP.
OpenAI faces multiple competitive pressures
The background to OpenAI’s launch of the red alert is the pressure it faces from multiple competitors. A new version of the Gemini AI model released by Google last month surpassed OpenAI in industry benchmarks, sending the stock price of Google parent company Alphabet soaring. Last week, Alphabet's stock price rose by more than 14% in a week. Since the release of Gemini 3 two weeks ago, it has also risen by more than 10% in less than two weeks as of last Friday.
Gemini's user base has continued to grow since the release of image generator Nano Banana in August. Google revealed that monthly active users increased from 450 million in July to 650 million in October.
OpenAI also faces pressure from Anthropic, which is growing in popularity among enterprise customers. Although OpenAI still dominates overall chatbot usage with more than 800 million weekly active users, users are gradually being attracted to Google.
Nvidia responds to TPU challenge
Faced with the rise of Google TPU, Nvidia emphasized its strong relationship with Google Cloud Platform in its communication with UBS, pointing out that Google uses both TPU and GPU in Gemini inference workloads.
Nvidia believes that cloud service providers are unlikely to run TPUs in their cloud stacks because significant workload optimization is required to achieve total cost of ownership (TCO) benefits on application-specific integrated circuits (ASICs). Nvidia also said that its performance advantage over peers has not narrowed so far.
Looking to 2026, Nvidia pointed to Anthropic's 1 gigawatt (GW) capacity and HUMAIN's 600,000-unit expansion as additions to its $500 billion order book for 2025-2026, providing potential upside.
Nvidia's CPX chips are targeting advanced programming applications that require context windows of more than 1 million tokens. Nvidia has not officially announced market size, but has previously suggested that contextual window applications account for about 20% of the inference market.
Altman said last month that OpenAI's data center projects have committed investments totaling about $1.4 trillion over the next eight years. In other words, OpenAI has $1.4 trillion in committed capital to maintain its industry leadership.
Overall, OpenAI does have reason to be nervous, but the turmoil remains within the company for now. As for whether Nvidia, the world's most valuable company, also faces a similar "red alert," the market is still watching closely.