> ## Content Index
> Fetch the complete content index at: https://nordisk-ai.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# Meta's New Llama 4 Model Will Be Trained on Impressive Hardware
- URL: https://nordisk-ai.ghost.io/meta-s-new-llama-4-model-will-be-trained-on-impressive-hardware/
- Published: 2024-12-06T08:11:03.000Z
- Updated: 2026-08-27T17:20:53.000Z
- Description: Meta is building a GPU cluster "bigger than anything else" to power Llama 4, their latest AI model. Can their open AI strategy overcome the competition from OpenAI and Google?
- Author: Anders Hagen
- Tags: Hardware, AI Development, Generative AI, Google, GPU Clusters, sustainability, Llama 4, Mark Zuckerberg, Meta, OpenAI, #Migrated-1787841634512, #wp, #wp-post, #Import 2026-08-27 14:41, #Import 2026-09-07 13:41

Meta is making a major push with its Llama AI model, now building a massive data cluster to train the upcoming Llama 4\. This GPU cluster is bigger than anything built before and requires enormous amounts of energy - an estimated 150 megawatts for a cluster of 100,000 H100 chips, significantly more than, for example, the El Capitan supercomputer at the US national laboratories, which requires 30 megawatts. Meta has earmarked up to $40 billion in capital to build and maintain data centers and other infrastructure, an increase of over 42% compared to last year.

#### Meta Goes Open, While OpenAI and Google Keep Their Models Closed

Unlike OpenAI and Google, which work with closed AI models, Meta is betting on an open strategy. Zuckerberg argues open source is more cost-effective, adaptable, and delivers greater trust and performance. Despite safety concerns around open AI models and their potential for misuse, Meta claims its restrictions on Llama are sufficient, even though it's possible for some users to bypass them. But given China has started using Llama for military purposes in the form of ChatBIT, how safe is it really?

#### An Incredible Number of People Use the Llama AI Model Today

Meta AI, the chatbot based on Llama, is already used by over 500 million people every month on platforms like Facebook, Instagram, and WhatsApp. Meta intends to further develop this usage and sees potential for revenue generation through advertising within the service, which could let the company keep Llama models open and accessible to developers and users worldwide.

#### Energy Equivalent to 34 Million American Households Is Required

The powerful computing capacity Meta and other AI giants use requires enormous amounts of energy. A single modern AI GPU can draw up to 3.7 MWh per year. For a cluster of 100,000 GPUs, like the one Meta is now building, this means an annual consumption of at least 370 GWh - enough to power over 34 million average American households. This raises questions about how companies can secure such a large energy supply, especially since new power sources take time to establish. Even Mark Zuckerberg has said energy scarcity could limit AI's growth.

For example, Elon Musk needed to use several large mobile generators to power his cluster of 100,000 GPUs in Memphis. Google has fallen behind on its climate goals, increasing its emissions by 48% since 2019\. Eric Schmidt, Google's former CEO, has suggested the world could temporarily set aside climate goals to allow AI companies to grow, and then use AI to solve the climate crisis.

When an analyst asked Meta how the company plans to handle the energy needs of its massive GPU cluster, company leadership avoided the question. Meanwhile, competitors like Microsoft, Google, Oracle, and Amazon have begun investing in small modular nuclear reactors or are planning to restart older nuclear plants. Although these investments take time to realize, small nuclear reactors powering AI data centers could reduce the strain on the national power grid and contribute to more sustainable energy use in the future.

#### Competition in AI Is Ramping Up

Meanwhile, OpenAI is developing its next major AI system, GPT-5, and has signaled it will represent a significant technical leap forward. Google also continues to challenge with its Gemini series of generative AI models. The AI market is thus in an intense phase of innovation, with Meta, OpenAI, and Google all competing to lead the way.