Amazon’s newest server packs up to 144 of the company’s custom-built Trainium3 chips, able to output 4.4 times more compute than the second generation, with 4x the efficiency and almost 4x the memory bandwidth of its second-generation server.
This chip follows the new 7th-generation “Ironwood” AI chip from Google and Nvidia’s recently introduced Blackwell Ultra.
An exposed dataset from the license plate surveillance company Flock, which is known to work with the US Border Patrol and ICE via local police, showed that some of the AI annotators paid to classify American license plates are located in the Philippines.
After 404 Media contacted Flock for comment, the dataset disappeared.
Microsoft, just like OpenAI, Google, and Perplexity, has announced new AI shopping updates, as US users of Copilot in Edge will have access to Cashback, Price Comparison, Price History, Product Insights, and Price Tracking features. It also says Copilot’s AI will now proactively tell you if there’s a cheaper deal elsewhere.
Huang said this during a company meeting with employees last week, Fortune reports. Apparently some managers have told Nvidia employees to use less AI? Not on Huang’s watch. He said he wants AI to automate “every task” that can be automated with AI – and employees shouldn’t worry about AI taking their jobs in the process.
The infrastructure buildout will add nearly 1.3 gigawatts of capacity for AI and cost up to $50 billion, the company said. US government customers will have access to both AWS Trainium AI chips and NVIDIA chips, and Amazon said it plans to start building the data centers in 2026.
