Elon Musk: 1M Nvidia GPUs? Nah, My Supercomputers Need the Power of 50M
Musk today tweeted that xAI aims for compute power that's on par with 50 million Nvidia H100 GPUs, the enterprise-grade graphics chip widely used for AI training and running chatbots. "The xAI goal is 50 million in units of H100 equivalent-AI compute (but much better power-efficiency) online within 5 years,' he said.
Musk's tweet comes a day after rival Sam Altman, the CEO of OpenAI, wrote in his own post about plans to run 'well over 1 million GPUs by the end of this year,' with the goal of exponentially scaling up the compute power by '100x.' Meta CEO Mark Zuckerberg, meanwhile, has a similar goal; he wants mega data centers devoted to developing AI super intelligence.
These growing AI investments underscore how expensive it is to scale up (and attract top talent). Musk's tweet doesn't mean he'll try to buy 50 million GPUs, though. The H100 was introduced in 2022 before Nvidia began offering more powerful models, including in the GB200, which can reportedly deliver an up to 2.5 times performance boost.
Nvidia has also released a roadmap that outlines two additional GPU architectures, Rubin and Feynman, which promise to unleash more powerful AI chips in the coming years with improved power efficiency. Still, Musk's xAI will likely need to buy millions of Nvidia GPUs to reach his goal.
In the meantime, Musk said in another tweet that xAI's Colossus supercomputer in Memphis, Tennessee, has grown to 230,000 GPUs, including 30,000 Nvidia GB200s. His company is also building a second Colossus data center that'll host 550,000 GPUs made up of Nvidia's GB200s and more advanced GB300 chips.
This compute power requires enormous amounts of electricity; xAI is using gas turbines at the Colossus site, which environmental groups say are worsening the air pollution in Memphis.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
24 minutes ago
- Yahoo
OpenAI eyes $500 billion valuation in potential employee share sale, source says
(Reuters) -ChatGPT maker OpenAI is in early talks for a potential secondary stock sale that would allow current and former employees to sell shares, valuing the company at around $500 billion, a source familiar with the matter told Reuters on Tuesday. Bloomberg was first to report the news. The Microsoft-backed company aims to raise billions through the sale, with existing investors, including Thrive Capital, expressing interest in buying some of the employee shares, the source said. Thrive Capital declined to comment on a Reuters request. Separately, OpenAI is still in the process of raising $40 billion in a new funding round led by SoftBank Group at a $300 billion valuation to advance AI research, expand computational infrastructure and enhance its tools.
USA Today
25 minutes ago
- USA Today
It's not you, it's me. ChatGPT doesn't want to be your therapist or friend
In a case of "it's not you, it's me," the creators of ChatGPT no longer want the chatbot to play the role of therapist or trusted confidant. OpenAI, the company behind the popular bot, announced that it had incorporated some 'changes,' specifically mental health-focused guardrails designed to prevent users from becoming too reliant on the technology, with a focus on people who view ChatGPT as a therapist or friend. The changes come months after reports detailing negative and particularly worrisome user experiences raised concerns about the model's tendency to 'validate doubts, fuel anger, urge impulsive actions, or reinforce negative emotions [and thoughts].' The company confirmed in its most recent blog post that an update made earlier this year made ChatGPT 'noticeably more sycophantic,' or 'too agreeable,' 'sometimes saying what sounded nice instead of what was helpful.' OpenAI announced they have 'rolled back' certain initiatives, including changes in how they use feedback and their approach to measuring 'real-world usefulness over the long term, not just whether you liked the answer in the moment.' 'There have been instances where our 4o model fell short in recognizing signs of delusion or emotional dependency,' OpenAI wrote in an Aug. 4 announcement. 'While rare, we're continuing to improve our models and are developing tools to better detect signs of mental or emotional distress so ChatGPT can respond appropriately and point people to evidence-based resources when needed.' Here's what to know about the recent changes to ChatGPT, including what these mental health guardrails mean for users. ChatGPT integrates 'changes' to help users thrive According to OpenAI, the 'changes' were designed to help ChatGPT users 'thrive.' 'We also know that AI can feel more responsive and personal than prior technologies, especially for vulnerable individuals experiencing mental or emotional distress,' OpenAI said. 'To us, helping you thrive means being there when you're struggling, helping you stay in control of your time, and guiding—not deciding—when you face personal challenges.' The company said its 'working closely' with experts, including physicians, human-computer-interaction (HCI) researchers and clinicians as well as an advisory group, to improve how 'ChatGPT responds in critical moments—for example, when someone shows signs of mental or emotional distress.' Thanks to recent 'optimization,' ChatGPT is now able to: 'Our goal to help you thrive won't change. Our approach will keep evolving as we learn from real-world use,' OpenAI said in its blog post. 'We hold ourselves to one test: if someone we love turned to ChatGPT for support, would we feel reassured? Getting to an unequivocal 'yes' is our work.'
Wall Street Journal
41 minutes ago
- Wall Street Journal
China's Z.ai and America's Self-Defeating AI Strategy
China's DeepSeek shocked the global AI community in January by building a frontier model at a fraction of Western costs. Now it has been outdone by a Chinese company subject to U.S. sanctions. It has become painfully obvious that Washington's strategy of restricting chip exports isn't working. formerly Zhipu AI, last week launched GLM-4.5, a production-level open-source model priced at 13% of DeepSeek's cost. It matches or exceeds Western standards in coding, reasoning and tool use. runs on only eight Nvidia H20 chips, which Nvidia recently gained reapproval to sell in China. That's better performance than DeepSeek with about half the hardware.
