This Week in AI: What’s the hidden truth behind DeepSeek?

Getting your Trinity Audio player ready...

Recently, DeepSeek, a generative artificial intelligence (GenAI) chatbot, disrupted the AI industry. The China-based company was founded in 2023, but until January 24, barely anyone had heard of it. Then, almost overnight, DeepSeek became the hottest topic of discussion in the AI world.

What made DeepSeek such a big talking point was the claim that the AI model was built for only $5.58 million using 2,000 Nvidia (NASDAQ: NVDA) H800 GPUs and that it performs just as well as OpenAI’s GPT-o1, a model that cost billions to make. If the DeepSeek team is telling the truth, it raises a big question for American investors and tech companies: Why are United States companies spending billions when DeepSeek supposedly pulled this off for a fraction of the cost?

Developers have also played a crucial role in making DeepSeek popular, particularly because it is cheaper to build on DeepSeek than on OpenAI.

Both of these elements have led to a variety of concerns about DeepSeek usage. Some are worried about data privacy, as reports have surfaced that DeepSeek is capturing user data from Americans and storing it on Chinese servers. On top of that, nobody knows how the data that a GenAI system captures is being used or to whom it is being sold.

Additionally, skeptics question whether the company really built this model for only $5.58 million or if they’re hiding the actual costs. If the firm did spend more to get the model built, it calls into question how they circumvented U.S. chip restrictions. Another point of controversy is a new investigation led by Microsoft (NASDAQ: MSFT) into whether DeepSeek stole OpenAI research and data to develop its model.

Each of these items has profound implications for DeepSeek. If the company somehow circumvents U.S. chip export restrictions, there will probably be legal consequences or some other kind of penalty for doing so. If DeepSeek simply copies OpenAI’s work, then it will appear to be much less innovative than it was originally suggested to be. This would likely result in DeepSeek losing some credibility, as it would be framed as a copycat rather than an innovative new entrant in the AI industry.

Microsoft, Meta, and OpenAI doubling down on AI spending

Despite concerns expressed by the tech industry and investors regarding AI tech giants in the U.S. being overvalued and overspending, it does not look like that trend will stop anytime soon. Tech giants have actually suggested that they will be doubling down rather than pulling back amidst this news.

During their latest earnings calls, Microsoft and Meta (NASDAQ: META) spent a combined $37.4 billion, mainly on AI infrastructure, in Q4 2024 alone. The money primarily went toward chips, data centers, and everything needed to keep their AI operations running and expanding.

For 2025, Microsoft has committed to spending $80 billion on AI infrastructure, and Meta plans to spend up to $65 billion. Mark Zuckerberg says all this spending is a long-term play that will be a strategic advantage in the long run.

This massive spending comes at a time when DeepSeek claims it built a GPT-o1-level model for just $5.58 million, which has caused many people to question why companies like OpenAI and Microsoft need billions. The reason is that AI operations take a significant amount of compute, data, and energy to train and deploy—an amount that can easily run into the billions for a large-scale operation. It may actually be the case that it’s not just American companies that need to raise and spend billions but that all AI companies seriously competing in the space need to spend that amount of capital. Skeptics think DeepSeek’s cost claims are far off from the actual number and that they’re either hiding their real infrastructure costs or benefiting from undisclosed support.

It’s not just Microsoft and Meta increasing their AI spend, OpenAI is reportedly looking to raise $40 billion in a new funding round that values the company at $340 billion. This shows that they are not concerned about a supposed low-cost competitor’s ability to truly change the industry’s operations.

OpenAI quietly rolls out ChatGPT Gov 

Although it was drowned out by all of the DeepSeek buzz, this week OpenAI released ChatGPT Gov. This new version of ChatGPT is designed specifically for government agencies, allowing them to run AI models on their own private cloud infrastructure using Azure Commercial or Azure Government Cloud.

Governments and large corporations are hesitant to adopt GenAI because of privacy concerns. Nobody wants their sensitive data to be used to train a chatbot, and there’s always the risk that private information entered into an AI tool could end up in a future response. That’s why many companies and government agencies have outright banned their employees from using ChatGPT for work-related tasks.

With ChatGPT Gov, OpenAI is trying to solve that problem. Although dull in comparison to the DeepSeek uproar and the fallout it has caused, easy-to-use AI products that cater to governments and corporations have been far and few in between.

By giving government agencies a way to run AI models on their own secure cloud, governments can use GenAI with the confidence that they can manage their own security, privacy, and compliance requirements to avoid the potential risks and pitfalls previously preventing them from using these systems.

In order for artificial intelligence (AI) to work right within the law and thrive in the face of growing challenges, it needs to integrate an enterprise blockchain system that ensures data input quality and ownership—allowing it to keep data safe while also guaranteeing the immutability of data. Check out CoinGeek’s coverage on this emerging tech to learn more why Enterprise blockchain will be the backbone of AI.

Watch: Transformative AI applications are coming