|
|
|
Jun 25, 2026
|
|
|
|
|
Supported by
|
|
|
|
|
|
|
Happy Thursday! OpenAI and Broadcom unveil a new server chip for running AI. Qualcomm agrees to acquire AI infrastructure startup Modular for nearly $4 billion. Anthropic accuses Alibaba of illicitly accessing Claude AI models.
|
|
|
|
OpenAI and Broadcom unveiled a new inference server chip the two companies have developed for running AI, a step forward in OpenAI’s effort to reduce its reliance on Nvidia chips and control its own hardware. Talks between OpenAI and Broadcom date back to 2024, and the companies were recently figuring out Broadcom’s financing of the effort, which initially will cost around $18 billion, and which cloud provider would commit to buying the chip. OpenAI’s blog post implied that Microsoft, one of OpenAI’s biggest cloud providers, is a buyer of the chip, called Jalapeno, and it said in a blog post that “early testing” showed it would “deliver performance per watt substantially better than current state-of-the-art.” If successful, the chip is also a big step forward for Broadcom, whose AI chip design business has taken off in the past 12 months thanks to customers such as Google, which works with Broadcom on its TPU AI chips.
|
|
|
|
Qualcomm has agreed to pay around $3.9 billion in stock for Modular, a software startup that allows developers to write software to run on different chips without having to rewrite the code for each chip. The deal price, based on Qualcomm’s last closing price on Tuesday, more than doubled the startup’s last private valuation of around $1.6 billion last September. Founded in 2022, Modular aimed to challenge Nvidia’s proprietary CUDA, the dominant software for training and running machine-learning models on its chips. Qualcomm, which generates most of its revenue from selling chips for smartphones, has been looking for ways to compete better with rivals in the AI data center chip market. A deal with Modular could help Qualcomm to build out its own software to serve different chips from different suppliers. It has also been in talks with AI chip startup Tenstorrent, The Information reported.
|
|
|
|
Anthropic accused Chinese tech giant Alibaba Group of illicitly accessing its Claude AI models to extract their capabilities in violation of the U.S. company’s terms of service. In a letter to U.S. senators dated June 10, Anthropic said Alibaba and its Qwen AI lab generated more than 28.8 million exchanges with Claude between April 22 and June 5 through almost 25,000 fraudulent accounts in a practice known as distillation. Anthropic described Alibaba’s campaign as “the largest known distillation attack on Anthropic to date.” Anthropic also urged Congress to expand information sharing between U.S. AI labs, close loopholes that allow Chinese companies to access advanced U.S. chips and impose penalties on Chinese AI labs involved in distillation attacks. The latest accusation follows Anthropic’s earlier claim in February, when the U.S. company accused three Chinese AI developers—DeepSeek, Moonshot AI and MiniMax—of using Claude to train and improve their own models. An Alibaba spokesperson didn’t comment on Anthropic’s accusation.
|
|
|
|
OpenAI has hired Amazon Web Services veteran Chris Grusz to serve as managing director of cloud partnerships, a role that involves working with the startup’s cloud provider and software partners. Grusz, whose recently updated LinkedIn profile reflects his new role, spent nearly 11 years at AWS overseeing partnerships with Salesforce, SAP, Snowflake, and Databricks, among others. He is now reporting to Colleen Kapase, OpenAI’s vice president of strategic global partnerships and ecosystems, according to a person familiar with the matter. Kapase joined OpenAI from Google Cloud in April, and OpenAI earlier this month also hired 14-year Salesforce veteran Brian Landsman to serve as vice president of global partnerships. OpenAI has committed to spend more than $600 million across several large cloud providers in the coming years to develop its technology. At AWS, Grusz worked on software partnerships that have helped the cloud provider attract more business customers, and OpenAI is following a similar playbook as it tries to broaden the market for its AI models and other products. These agreements can be mutually lucrative: Snowflake announced last month that it sold more than $2 million of its products through AWS last year, and that it had committed to spend $6 billion on AWS chips and other products over the next five years.
|
|
|
|
Ornn, a startup that tracks the cost of AI tokens and computing power, raised $33 million in a funding round led by Andreessen Horowitz. The money will be used to build a marketplace for trading compute, executives at the company told Bloomberg. Ornn is one of several companies seeking to turn the buying and selling of compute into a new asset class. While there are already several marketplaces for compute, and many companies like neoclouds offer it by the hour, founders are betting that there is an opportunity to consolidate buy and sell orders. Earlier this year, the Intercontinental Exchange, which owns the New York Stock Exchange, said it would offer futures contracts that rely on Ornn’s index. CME Group is also working on its own contract. The contracts are still awaiting U.S. regulatory approval, and there are still questions about how useful actual compute buyers and sellers will find them or if a deep enough market will develop. Still, Goldman Sachs and JPMorgan are exploring ways they might trade in the nascent market, underlining broader interest in ways to track and hedge against compute price fluctuations.
|
|
|
| |
