At a recent company-wide meeting, Google officially launched its latest AI large model - the Gemini3 series. This series is hailed as the strongest AI model currently available, even surpassing the products of market competitors such as OpenAI, which has drawn widespread attention from the technology industry, and Google's stock price has also risen significantly as a result.
Amin Vahdat, head of Google Cloud AI Infrastructure, detailed the future development goals at the meeting. He stated that Google plans to achieve a 1000-fold improvement in computing power, storage, and network capabilities under the same cost and energy consumption conditions within the next 4 to 5 years. This means that Google hopes to provide stronger technical support in the face of increasing AI demand. In addition, Google also plans to double its computing capacity every 6 months to maintain continuous technological progress.
To achieve this goal, Google has launched its seventh-generation TPU platform, codenamed Ironwood. The core configuration of this platform includes 192GB of HBM memory and 7.4TB/s bandwidth, with a single chip performance of 4614TFLOPS, which is 10 times that of the fifth-generation TPU, and a performance per watt improvement of about 6 times. These innovations have placed Google at the forefront of AI computing power, providing strong support for the Gemini3 series.
When talking about future technology layout, Vahdat pointed out that Google has already cooperated with companies such as Broadcom to start developing the eighth-generation TPU. Although specific specifications have not been disclosed yet, it can be anticipated that the new TPU will bring more significant performance improvements. At the same time, Google's capital expenditure plan is gradually increasing, and it will significantly increase next year, further supporting the construction of AI infrastructure.
Key points:
🚀 Google launched the Gemini3 series large model, becoming the strongest AI model currently available.
📈 In the next 4-5 years, it will achieve a 1000-fold improvement in computing power, storage, and network capabilities under the same cost and energy consumption conditions.
💻 The seventh-generation TPU platform Ironwood has a single-chip performance of 4614TFLOPS, significantly improving computing power.
