|
|
|
Mar 17, 2026
|
|
|
|
|
Supported by
|
|
|
|
|
|
|
Happy Tuesday! Jensen Huang says Nvidia expects $1 trillion in revenue from its Blackwell and Rubin AI chips between 2025 and 2027. OpenAI’s applications chief tells staff to focus on business customers. Alibaba consolidates its AI businesses in a major reorganization.
|
|
|
|
Nvidia CEO Jensen Huang said the company now expects $1 trillion in revenue from its Blackwell and Rubin AI chips between 2025 and 2027, a huge amount for a company that in the year through January reported $216 billion in revenue. In making the projection, Huang was updating a projection from last year when he said that Nvidia expected $500 billion in revenue from sales of its latest AI chips, the Blackwell and Rubin, between 2025 and 2026. At that time, the projection created some confusion, as it wasn’t entirely clear what the $500 billion figure represented and precisely what period. The number turned out
to cover the calendar years 2025 through 2026 even though Nvidia doesn’t report on a calendar-year basis but instead on a fiscal year ending in January. On Monday, a slide behind Huang showed the $1 trillion figure covered the period between 2025 and 2027 (presumably the calendar years, if he’s being consistent with the previous projection). That’s roughly consistent with the $1.05 trillion that analysts are estimating Nvidia will earn in revenue for the three years through January 2028, according to S&P Global Market Intelligence.
|
|
|
|
OpenAI’s applications chief Fidji Simo told staff in a meeting last week that the company needed to refocus on business customers and cut down on side quests that were becoming a distraction, according to a person with knowledge of the comments. Simo made the comments as archrival Anthropic has made gains with businesses with AI for coding and other white-collar work and is quickly closing a revenue gap with OpenAI. Simo implied the company would move staff toward enterprise-focused products and away from consumer ones, this person said. The company on Monday also
said it was looking to form a joint venture with private equity firms to sell its technology to businesses, a move that followed a report in The Information last week about Anthropic’s similar effort to form a JV with Blackstone, Hellman & Friedman and other PE firms. The Simo comments resembled the kind of “code red” declaration CEO Sam Altman made at the end of last year as pressure was rising to keep up the growth of OpenAI’s core ChatGPT product and subscriptions in the face of a threat from Google. The Wall Street Journal earlier reported on the Simo comments.
|
|
|
|
Alibaba Group has moved all of its major AI businesses under one new division managed directly by the company’s CEO, in a drastic reorganization aimed at facilitating more cooperation among teams that work on AI models, apps and services. The new division, Alibaba Token Hub, comprises Tongyi Laboratory, a unit responsible for research and development of AI models, as well as Alibaba’s Qwen AI app for consumers. It also includes Alibaba’s new AI agent for enterprise users called Wukong, which was unveiled Tuesday. In addition, the new division will manage Alibaba’s Model Studio, a marketplace that distributes AI models to enterprise customers. Alibaba Cloud, which managed Model Studio before this reorganization, will continue to be in charge of the marketplace’s sales and customer service, while Alibaba Token
Hub will handle the technical work of running the platform. Revenue from Model Studio’s sales of AI models will continue to belong to Alibaba Cloud, not the new division. Alibaba CEO Eddie Wu said in a statement that Alibaba Token Hub will drive “strategic coordination” across the company’s AI businesses. He said the new division’s mission is to “create tokens, deliver tokens and apply tokens.” (Tokens are basic units of data that AI models process.) Wu also said the new structure will enable Alibaba to seize business opportunities stemming from the rise of AI agents. “Billions of AI agents are poised to take on an ever-greater share of digital work, each powered by tokens generated by models, and these agents will increasingly become the primary interface between people and the digital world,” he said. Alibaba’s announcement of the reorganization follows the recent resignation of top AI researcher Junyang Lin, a key architect of the firm’s highly regarded Qwen open-source large language models. Lin, who departed earlier this month, was leading the LLM team at Tongyi Laboratory, which is now coming under the new Alibaba Token Hub division.
|
|
|
|
Nvidia CEO Jensen Huang announced a new AI server system based on technology the company licensed last year from chip startup Groq, designed to make its AI servers more energy- and cost-efficient for tasks such as AI coding. This is the first time Nvidia has integrated another company’s AI processor into its server racks, and it’s a tacit admission that its graphics processing units, on their own, may not always be the best option for certain AI workloads. The system, Groq LPX, will be an optional add-on for Nvidia’s upcoming AI server chip system, Vera Rubin. The Groq part of the system will handle some of the computing tasks that would otherwise run on the Vera Rubin servers. Huang said during his keynote at Nvidia’s annual developer conference that the Groq system will begin shipping in the third quarter of
this year. He also confirmed The Information’s earlier reporting that the main chip used in the Groq system will be manufactured by Samsung. (Nvidia’s chips are produced by Taiwan Semiconductor Manufacturing Co.) Unlike Nvidia’s graphics processing units that underpin the Vera Rubin system, Groq’s design integrates memory directly onto the chip, which can help deliver faster responses for tasks like software coding that benefit from quicker response times. While Groq’s chips have historically been limited by the small amount of memory that can be placed on each chip, Nvidia plans to address this by linking up to 256 Groq chips so they work together in unison.
|
|
|
|
Meta Platforms struck a deal to spend $27 billion on Nebius Group’s AI data centers over the next five years, the cloud provider announced Monday. It follows a $3 billion deal Meta struck with Nebius last November. In the latest deal, Dutch firm Nebius will supply Meta with $12 billion of dedicated capacity starting early next year, which will include deployment of Nvidia’s latest generation of AI chips, known as Vera Rubin. The parent company of Facebook, Instagram and Threads will buy up to $15 billion in additional capacity over a five year period. The deal underscores Meta’s appetite for greater compute capacity to support the development of its AI services. The company has projected it will spend between $115 billion and $135 billion in 2026 on capital expenditures, nearly double that of 2025, driven mostly
by investments in AI infrastructure.
|
|
|
|
| |
