Google Q4 2025 财报：Gemini token 处理量同比增长 52 倍，服务成本下降 78%
Google's 52x AI Growth
Google Q4 2025 财报电话会披露，Gemini 经直接 API 每分钟处理超 100 亿 token，同比增长 52 倍，同时 Gemini 服务单位成本下降 78%，相当于每 GPU 小时 token 产出提升 4.5 倍。
作者把 Google 财报数据与 Microsoft 同口径对比，读者可借此横向判断各家云厂商的 AI 增长与成本效率差异。
In short : Google's Gemini processes 10B tokens/minute, up 52x YoY, while cutting serving costs 78%. Cloud revenue hit $17.7B, growing 48%.
Google’s Q4 2025 earnings call revealed a company in the midst of a spectacular AI acceleration.
“Our first-party models like Gemini now process over 10 billion tokens per minute via direct API used by our customers, up from 7 billion last quarter.”
This represents a staggering 52x increase year-over-year, up from ~8.3 trillion tokens/month in December 2024 to an annualized run rate of over 430 trillion.
For context on the scale :
“Nearly 350 customers each process more than 100 billion tokens.”
Microsoft reported over 250 customers projected to process more than 1 trillion tokens annually - a 10x higher threshold, suggesting their largest customers are consuming significantly more tokens per account.
While volume is exploding, costs are plummeting. Google announced :
“We were able to lower Gemini serving unit costs by 78%.”
This means a 4.5x improvement in tokens per GPU hour.
Compare this to Microsoft’s update last year, where they highlighted a 90% increase in tokens per GPU. Google’s 4.5x (or 350%) improvement suggests they are finding massive efficiencies in their TPU infrastructure and model architecture.
The AI boom is translating directly to revenue :
“Backlog grew 55% to $240 billion.”
This compares to Microsoft’s RPO of $625 billion, 45% of which comes from OpenAI.
Gemini Enterprise has sold more than 8 million paid seats just four months after launch. Google Cloud revenue grew 48% to $17.7 billion, outpacing Azure’s 39% growth.
To fuel this growth, Google is committing capital at an unprecedented scale.
“Our 2026 CapEx investments are anticipated to be in the range of $175 to $180 billion.”
If Google alone is spending ~$175B, the hyperscalers collectively (Google, Microsoft, Amazon, & Meta) could drive $500B to $750B in data center CapEx this year. This level of investment signals their conviction that the demand for tokens is only just beginning.
Currently, AI infrastructure spending is ~1.6% of GDP, compared to the peak of the railroad era at 6.0%. At this rate, AI data center buildouts would be equivalent to the national highway system as an investment percentage of GDP.
Google’s AI business is growing at 48% while reducing serving costs by about 80%. The efficiency of the business is unparalleled.
Get the next one in your inbox
The 1-minute read that turns tech data into strategic advantage.
Read by 150k+ founders & operators.
GP at Theory Ventures. Former Google PM. Sharing data-driven insights on AI, web3, & venture capital.
来源：Tomer Tunguz 博客（VC 分析） · tomtunguz.com