close
close

Association-anemone

Bite-sized brilliance in every update

Meta’s next Llama AI models train on GPU cluster ‘bigger than anything else’
asane

Meta’s next Llama AI models train on GPU cluster ‘bigger than anything else’

Managing such a mammoth array of chips to develop Llama 4 is likely to present unique engineering challenges and require large amounts of energy. Meta executives sidestepped a question about the analyst on Wednesday energy access constraints in parts of the US that have hindered companies’ efforts to develop stronger AI.

Conformable an estimatea cluster of 100,000 H100 chips would require 150 megawatts of power. The largest national laboratory supercomputer in the United States, El Capitanby contrast it requires 30 megawatts of power. Meta expects to spend up to $40 billion in capital this year to provide data centers and other infrastructure, an increase of more than 42% from 2023. The company expects even more torrid growth in this spending next year future.

Meta’s total operating costs are up about 9% this year. But overall sales — mostly from advertising — rose more than 22 percent, leaving the company with higher margins and higher profits, even as it poured billions of dollars into Llama’s efforts.

Meanwhile, OpenAI, considered the current leader in cutting-edge AI development, is burning money despite charging developers for access to its models. What for now it remains a nonprofit business said it is training GPT-5, a successor to the model currently powering ChatGPT. OpenAI said GPT-5 will be larger than its predecessor, but did not say anything about the computer cluster it uses for training. OpenAI also said that in addition to scale, GPT-5 will incorporate other innovations, including a recently developed solution. approach to reasoning.

CEO Sam Altman he said that GPT-5 will be “a significant leap forward” compared to its predecessor. Last week, Altman responded to a news report that said OpenAI’s next frontier model would be released by December. by writing on X, “fake news out of control.”

On Tuesday, Google CEO Sundar Pichai said that the company’s newest version The Gemini family of generative AI models is in development.

Meta’s open approach to AI has proven controversial at times. Some AI experts fear that making significantly more powerful AI models available for free could be dangerous because they could help criminals launch cyber attacks or automate the design of chemical or biological weapons. Although Llama is fine-tuned before release to restrict inappropriate behavior, it is relatively trivial to remove these restrictions.

Zuckerberg remains optimistic about the open source strategy, even as Google and OpenAI promote proprietary systems. “It seems pretty clear to me that open source is going to be the most cost-effective, customizable, reliable, performant and user-friendly option available to developers,” he said Wednesday. “And I’m proud that Llama is leading the way in that.”

Zuckerberg added that the Llama 4’s new capabilities should be able to power a wider range of functions in Meta services. Today, the signature offering based on Llama models is the ChatGPT-like chatbot known as Meta AI, which is available in Facebook, Instagram, WhatsApp and other apps.

More than 500 million people use Meta AI every month, Zuckerberg said. Over time, Meta expects to generate advertising revenue from the feature. “There will be a broad set of queries that people use it for, and the monetization opportunities will exist over time as we get there,” Meta CFO Susan Li said on Wednesday’s call. With the potential for ad revenue, Meta might be able to subsidize Llama for everyone else.