Investing

Is Meta Platforms About to Make a $60 Billion Mistake?

Rich Duprey

Published: January 27, 2025 9:56 am

Meta Platforms (NASDAQ:META) CEO Mark Zuckerberg just declared that 2025 will be a defining year for artificial intelligence. In a Facebook post, he said his company is planning to invest between $60 billion and $65 billion this year in AI, including a data center “so large it would cover a significant part of Manhattan.” 

The facility to be built in Louisiana will power its latest Llama 4 iteration of its large language model that will serve over 1 billion people and become the leading AI assistant.

Zuckerberg promises this is not a one-off event, but will rather be a recurring theme for Meta. “We have the capital to continue investing in the years ahead,” he wrote, one which will drive Meta’s business and product development going forward.

Unfortunately, AI may not be defined in the way Zuckerberg envisions. It could be he is setting up Meta Platforms for a vast waste of resources. 

24/7 Wall St. Insights:

• Meta Platforms (META) recently revealed a massive $60 billion AI spending spree.
• It will include a new data center that is almost as large as all of Manhattan.
• However, Chinese AI lab DeepSeek just unveiled its latest LLM that outperforms all other models, and does so more efficiently, with less power, and at dramatically lower cost.
• If you’re looking for some stocks with huge potential, make sure to grab a free copy of our brand-new “The Next NVIDIA” report. It features a software stock we’re confident has 10X potential.

A new paradigm in AI

Chinese AI lab DeepSeek just released the latest iteration of its large language model (LLM), R1. It is an open-source model that allows users to modify its algorithm, fine tune it, and build on its foundation. More importantly, it outperforms all other AI models across various benchmarks, but is also dramatically cheaper and does so consuming less power.  

The prior model, called V3, was comparable to OpenAI’s GPT-4o and Anthropic’s Claude Sonnet 3.5. It also beat Meta’s Llama 3 and Alibaba‘s (NYSE:BABA) Qwen2.5. DeepSeek’s R1, however, has now surpassed ChatGPT’s own latest-generation model, o1, and is reportedly 27 times cheaper than ChatGPT. 

Being trained on just 2,048 of Nvidia‘s (NASDAQ:NVDA) H800 GPUs in only two months, it is equivalent to about 2.8 million GPU hours. In contrast, Meta’s Llama 3.1 required 16,384 H100 GPUs, or some 30.8 million GPU hours, according to Epoch AI. The cost for it was approximately $6 million.

What that means is Zuckerberg is preparing to have Meta Platforms spend 10,000 times more money than what comparable — or better — models can produce. And because DeepSeek is open-source, people will be able to tweak the algorithm to better serve their needs, and do so more cheaply. They will be saving money while Meta wastes it.

A threat to bloated AI budgets

It is a notable achievement by DeepSeek because it is using a dumbed-down version of more powerful Nvidia chips. The H800 chip was developed by the chipmaker to circumvent U.S. export restrictions on technology to China. Although the H800 chip was eventually banned as well, Nvidia had significantly reduced chip-to-chip data transfer rates, meaning the H800 is less powerful than the full-bore H100 chip. 

DeepSeek represents a significant threat to Meta and other AI stocks. It demonstrates the potential for high-performing AI models to be developed at a fraction of the cost, primarily due to its open-source nature and efficient training methods. It will be hard for Meta to justify its expensive AI development strategies and could force it to adapt to a more cost-effective approach to remain competitive. 

Meta has a history of overspending on projects. It essentially threw away $45 billion on the metaverse as Zuckerberg sought to change the social media company into a leading edge tech platform. Initially there were investor concerns when he turned his attention to AI that he would do the same thing, but the AI mania that swept the market provided cover for Meta’s ambitions in the space.

Upending the AI industry

DeepSeek R1’s efficiency could cause an immediate revaluation of Meta’s spending priorities, especially if more entities follow DeepSeek’s approach of optimizing software and algorithms to work within hardware constraints. If Meta moves forward with its $60 billion spending spree, it has the potential to significantly negatively impact its profit margins. 

Microsoft (NASDAQ:MSFT) is certainly not ignoring what’s happening. Speaking at the World Economic Forum, CEO Satya Nadella said, “We should take the developments out of China very, very seriously.”

Maybe Zuckerberg will take these developments seriously, too. Meta Platforms will need to adapt their strategies to remain competitive and not waste shareholder resources. It means Meta will have to focus on developing more efficient and cost-effective AI models to counter the competitive threat posed by DeepSeek. 

Spending $60 billion of limited shareholders resources on outdated, expensive workflows can no longer be justified.