Meta is pushing the boundaries of artificial intelligence with its latest development, Llama 4. This upcoming large language model is set to be a groundbreaking advancement in AI, as it utilizes a staggering number of over 100,000 Nvidia H100 GPUs for training. During a recent earnings call, CEO Mark Zuckerberg revealed this ambitious project, indicating that it’s the largest setup he’s aware of in the industry right now.
Llama 4 is anticipated to debut in early 2025, with initial releases of smaller versions. Although specifics about its capabilities are under wraps, Zuckerberg teased that users can expect new features, enhanced reasoning skills, and superior performance from Llama 4. Keeping in line with Meta’s tradition of democratizing AI, this model will be open-source and free to download, contrasting with competitors like OpenAI and Google, who restrict access to their models via APIs.
This initiative especially appeals to startups and researchers looking for greater flexibility in how they implement AI technologies. The scale at which Meta is operating is monumental, with energy demands reaching 150 megawatts—comparable to five times the power required by the largest US national lab supercomputer. Accordingly, Meta plans to significantly increase its infrastructure spending, projecting a 42 percent increase in 2024, amounting to $40 billion.
Despite these substantial investments, Meta has maintained a robust financial standing, reflecting a 22 percent increase in sales primarily driven by ad revenue. This financial boost has balanced the 9 percent rise in operating costs, showcasing Meta’s strategic acumen in managing its ambitious AI projects alongside its core business ventures.
