|
|
|
Oct 24, 2025
|
|
|
|
|
|
TGIF! Anthropic plans to use up to 1 million Google Tensor processing units. President Donald Trump calls off plans to send the National Guard on Saturday into San Francisco. Globalstar's shares surge on The Information's scoop.
|
|
|
|
Anthropic on Thursday said it plans to use up to 1 million Google Tensor processing units to power its artificial intelligence, implying the startup would spend tens of billions of dollars renting TPU servers. The announcement is a show of confidence in Google’s chip as an alternative to Nvidia’s graphics processing unit, which dominates the AI server market. Anthropic didn’t provide more details except to say that next year, the TPUs in its deal with Google will power a gigawatt of data center capacity. Developing a 1 GW data center costs tens of billions of dollars. It isn’t clear whether the TPUs for Anthropic would be run by Google Cloud or at data centers operated by other cloud providers. Recently, Google struck a deal to put a large number of TPUs in a New York data center run by cloud provider
Fluidstack, The Information reported, but it wasn’t clear who would use the TPUs. TPUs historically only ran inside Google Cloud providers, and the most advanced version of the chip was restricted to Google’s in-house AI teams. Google offers servers powered by the chip as an alternative to Nvidia’s GPUs, though Google Cloud is also one of the biggest buyers of GPUs, as many cloud customers prefer using those chips. Anthropic says it uses a mix of TPUs, GPUs, and Amazon’s Trainium AI chips. OpenAI at one point was testing TPUs but later said it didn’t have plans to use them at scale. Both Google and Amazon own significant stakes in Anthropic and have invested billions of dollars in the company. Anthropic spends billions of dollars a year renting servers from Google and Amazon’s cloud units.
|
|
|
|
President Donald Trump has called off plans to send the National Guard on Saturday into San Francisco, after he said friends who live in the area urged him not to go ahead with it. “Great people like Jensen Huang, Marc Benioff, and others have called saying that the future of San Francisco is great,” Trump said in a post on Truth Social. “They want to give it a “shot.” Therefore, we will not surge San Francisco on Saturday.” The mention of Benioff is notable. In a recent interview with The New York Times, Benioff praised Trump and said he
supported deploying National Guard troops in San Francisco. The comments sparked backlash and led prominent Silicon Valley venture capitalist Ron Conway to resign from the board of Salesforce’s philanthropic arm. Benioff later apologized for the comments but not before the Times published a separate report revealing that Salesforce had been pitching Immigration and Customs Enforcement (ICE) for government contracts.
|
|
|
|
Satellite company Globalstar shares surged nearly 8% on Thursday following The Information’s report that the company has been looking to sell itself for more than $10 billion. The company has been owned by investor James Monroe’s Thermo Companies since 2004 when he acquired the then beleaguered satellite company out of bankruptcy. After turning the company around and establishing a partnership with Apple, Monroe has been looking for a potential exit. Apple can be a tricky partner though. Globalstar’s interest in selling itself
could be one of several developments that increase the likelihood of a satellite partnership between Apple and Elon Musk’s SpaceX, The Information also reported. In Globalstar’s most recent quarterly earnings filing, it wrote about Apple: “The loss of the Customer would likely have a material adverse impact on our financial condition, results of operations and cash flows.”
|
|
|
|
Synthesia, a London-based maker of artificial intelligence software that generates video avatars of people, is in talks to raise a new round of funding led by early investor GV, an investment arm of Google-owner Alphabet, at a $4 billion valuation including the investment, according to two people with knowledge of the discussions. The startup recently passed $150 million in annual recurring revenue, up 50% from April. The fundraising discussions follow M&A talks with Adobe, which considered buying the startup for $3 billion, as well as talks about a potential sale to Meta Platforms, The Information reported Wednesday. Neither talks advanced. Synthesia last raised venture funding at the start of the year at a $2.1 billion valuation in a fundraising led by NEA. The current fundraise hasn’t closed and terms could still change.
|
|
|
|
Amazon launched a new artificial intelligence-powered product recommendation feature Thursday, the e-commerce giant’s latest move to incorporate more AI features into both its consumer-facing shopping experience as well as the systems that power its search and recommendation features. The new feature, known as Help Me Decide, will analyze a users’ browsing activity, as well as search and shopping history, to recommend a product. It will appear on product pages if a user has been browsing items but hasn’t made a purchase. Help Me Decide is available in beta to U.S. users. Amazon says that Help Me Decide uses AI models, including one from Amazon’s cloud service Bedrock, to understand what users are searching for and why by looking at their shopping history and preferences. It then matches that information with product details and reviews to make a suggestion. Amazon has been using generative AI to refine and improve its search results more generally, The Information previously reported, and has been working to better understand the intent behind shoppers’ queries rather than just looking for key search terms they’re using. Amazon is also testing new metrics for sellers that better assess the detail and accuracy of product listings to improve them for AI search.
|
|
|
|
On Intel’s third-quarter earnings call, CEO Lip-Bu Tan and CFO David Zinsner said the chipmaker is expecting a boost as more companies shift their spending from training AI models to inference, the process of using trained models to power AI applications. “It is increasingly clear that CPUs play a critical role today and will going forward within the AI data center as AI usage expands, and especially as inference workloads outpace tha | | | |
