Read Before You Buy an AI Stock: What Is DeepSeek and Why It Matters

DeepSeek is going viral as it is now taking over AI market share after the company released its R1 model last week. Developers have been moving en-masse to this company, and it is now the most downloaded app on the App Store.

Investors should get familiar with it since it is now causing a market-wide selloff. It is a good idea to know the competition and what you’re working with before you invest in any AI stock. DeepSeek is now the place to start, as there’s a good chance DeepSeek’s revolutionary breakthrough will shape the trends in the coming months.

What Is DeepSeek?

DeepSeek is a Chinese AI startup. It is a research lab founded in May 2023. However, unlike most AI companies — in both China and the West — this company didn’t invest in expensive GPU hardware to develop its models. Instead, it relied on reinforcement learning with a budget much, much lower than what OpenAI and other AI startups have. In fact, DeepSeek-V3's $5.6 million training cost is a fraction of Sam Altman’s $20 million McLaren car. The idea that such a small investment would lead to a breakthrough this big would’ve been unfathomable just a month ago.

What Did DeepSeek Do?

DeepSeek released its R1 model on January 20, 2025. This model beats most other AI models handily. It isn’t the best model across benchmarks, but the real achievement is that DeepSeek-R1 costs $0.55 per million tokens (cache misses) and is as low as $0.14 for cache hits. OpenAI’s o1 costs $15 to $16.5 per million tokens to use. You don’t need to know the technical terms to realize how much more efficient R1 is. DeepSeek is 95% cheaper for comparable reasoning tasks. The cost to train this model is rumored at around $12 million.

GPT-4 reportedly cost hundreds of millions of dollars to make. OpenAI is reportedly spending $500 million per six-month training cycle for its GPT-5 model. And GPT-4 isn’t even a reasoning model.

How Fast Is DeepSeek Catching Up?

DeepSeek is catching up extremely fast, and there’s a good argument it is leaving most AI startups behind. DeepSeek coder was released in November 2023. It then released DeepSeek-V2 in early 2024 and it released DeepSeek-V3 in late 2024. It then released DeepSeek-R1 less than a month after DeepSeek-3. This is a rapid iteration cycle, and the fact that it is open source likely means this is going to have a huge impact as other AI companies try to replicate this efficiency.

The Potential Market Impact of DeepSeek

DeepSeek is already impacting the market significantly. In the long run, you should use caution when investing in semiconductor AI companies. This breakthrough could cause a reduction in demand for AI chips and hurt pricing power and the addressable market significantly.

It could also hurt the broader tech market and companies invested deeply in AI-related projects. Hyperscalers could see their costs shrink, and energy companies could receive much less attention if AI models don’t require as much energy as previously thought. For example, Constellation Energy (NASDAQ:CEG) and Vistra (NYSE:VST) have already seen double-digit declines.

The Bottom Line

The huge rout in tech and energy stocks could accelerate if this trend starts to change how AI startups operate. Investors are just starting to take efficiency into consideration and are calling it a “black swan event” for global markets. DeepSeek has thrown a wrench into the AI industry, so if you’re putting money into it, it will be a while before you get any concrete projections about whether or not demand will stick around.

Some believe this will just open up more room for AI startups, as DeepSeek is open source and its efficiency could be used to make AI models even more powerful, but that is yet to be seen.

Recommended