The DeepSeek Deep Dive: What Experts Are Saying

Deepseek logo

By: R. Scott Raynovich
January 30, 2025


It’s been a wild week in the AI world. Last week, we had the announcement of the Stargate AI project from the White House, allegedly to the tune of $500 billion to build a new datacenter in Texas. We did our best to debunk parts of that story. This week, we got the release of a groundbreaking AI model from Chinese AI startup DeepSeek.

At the highest level, both the Stargate and DeepSeek events demonstrate a volatile mixture of AI competition at the global business level as well as AI's implications for geopolitics, as companies and countries alike race to produce the best AI technology. It’s been framed by many as the New Cold War.

The DeepSeek launch had large ramifications for the largest technology markets, the stock market, and foreign policy. This week, I spent at least 20 hours reading analysis from the world's top AI experts about DeepSeek, as well as speaking on the phone to many sources.

Let's dig into what I found.

Why DeepSeek Freaked Out Markets

Earlier in the week, DeepSeek had Wall Street panicking, at least initially, generating the largest single day market capitalization loss in history for AI giant NVIDIA (it lost $600 billion in value on Monday).

The reason? In short, fears about hardware pricing. Many AI hardware companies—especially NVIDIA—have been enjoying high margins and glorious revenue surges, as the largest companies in the world spend hundreds of billions on an AI arms race. DeepSeek introduced the idea that maybe AI shouldn't cost that much.

DeepSeek said it had built its new R1 model for only $6 million, a cost that is surely not accurate because DeepSeek itself said that doesn't include all the intellectual property and engineering that went into the product. These details were released in a published paper.

Regardless of the total pricetag, the DeepSeek R1 model did introduce some novel engineering and certainly showed that AI pricing could be reduced with creative approaches, demonstrated because China has been limited to the most cutting-edge hardware by import restrictions.

"Lowering the cost to train will increase the ROI on AI,” Gavin Baker, managing partner with investment fund Atreides Management, wrote on X:

    Skepticism Grows and Markets Bounce Back

    On Monday, these initial fears about DeepSeek had Wall Street panicking, but later in the week things started to stabilize, as our analyst Mary Jander correctly predicted with her analysis. The share prices of the larger AI infrastructure players such as NVIDIA bounced back as markets examined more nuances to the story. A couple of these points included:

    • DeepSeek technology was clearly not nearly as cheap as stated, because it didn’t take into account many hours of engineering prior to its construction. Nor do we know exactly what type of hardware resources were used.
    • General skepticism about the R1 model, because it comes from China, which has a history of intellectual property theft and lack of transparency.
    • The acknowledgement by most experts that the largest global companies are not going to trust a Chinese AI model with their data.

    Bob Machlin, the Founder and Principal of GenAI Ignite and a longtime technology industry executive, told me that he's advising his clients to avoid DeepSeek:

    "The chat service should be off limits for corporations. The model is hosted in China putting your data at risk."

    Machlin pointed out that this mistrust will prevent widespread adoption of U.S. companies, who will stick with their tried-and-true partners. He also believes that DeepSeek heavily leveraged technology from OpenAI, which is already the subject of close inspection by Microsoft and OpenAI.

    DeepSeek Is Still a Seminal Event

    Despite the recent surge of questions and skepticism about DeepSeek, there is no question that DeepSeek was a seminal event in AI history and it has definitely changed the mindset of the industry.

    Because the AI arms race has such large ramifications, let’s frame up the view of what DeepSeek means for the AI community with the pros and cons for both sides.

    What's Known:

    • There was clever engineering. According to top engineers speaking on the public record, DeepSeek used clever engineering to reduce the cost of building its effective AI model. “Deepseek V3/R1 includes multiple innovations compared to traditional LLM architecture we know from Llama or other open Models,” wrote Hugging Face engineer Philipp Schmid on X.
    • The impact was huge. The launch of DeepSeek shook the stock market, certainly caused tech executives to call emergency meetings worldwide, and fell under the microscope of foreign governments. One key data point is that the DeepSeek Chatbot app rocketed to the top of Apple’s App Store, becoming the most downloaded app for the week.

    What’s Not Known:

    • How was it built? Western competitors are already questioning the extent to which the team leveraged existing models such as Meta’s Llama and OpenAI to help train its model, a process known as distillation. Distillation can be done legally but it can also violate terms of use. We don't know how far DeepSeek pushed the limits.
    • How will the business world adopt DeepSeek? The initial reaction from key business leaders and experts I consulted indicates that DeepSeek is considered off limits to any kind of sensitive business or personal data. Who’s actually going to use it? It may be the number one app in Apple's download store, but that could also make it a policy target like TikTok.
    • What are the foreign policy ramifications? Founded and run in China, DeepSeek will be seen as a sovereign threat to some national AI policies, especially to the United States. Given the TikTok history, it’s likely the United States will consider actions to take against DeepSeek technology, its Chatbot, and countermeasures such a tightening of AI chip export restrictions (which are already being discussed).

    Atreides' Baker concluded that DeepSeek r1 is real with important nuances:

    "Several of the factors that show it’s impact are its meteoric rise to #1 in the Apple App Store, the fact that it has produced results comparable to relevant AI models such as 01, and that it used algorithmic breakthroughs demonstrating engineering creativity.”

    How Will the AI Industry Respond?

    The threat to large, established U.S.-based tech giants did not go unnoticed. While many top technology leaders congratulated DeepSeek on its accomplishments, DeepSeek will also likely influence their development strategies, as they take a look at DeepSeek's engineering approach and how they used distilled models.

    Mark Chen, the Chief Research Officer of OpenAI, when commenting on X congratulated DeepSeek but then indicated that he didn’t see the impact as large as some:

    “Their research paper demonstrates that they’ve independently found some of the core ideas that we did on our way to o1... However, I think the external response has been somewhat overblown, especially in narratives around cost. One implication of having two paradigms (pre-training and reasoning) is that we can optimize for a capability over two axes instead of one, which leads to lower costs. But it also means we have two axes along which we can scale, and we intend to push compute aggressively into both!"

    GenAI Ignite's Machlin told me that DeepSeek's use of models by large global companies will be weighed very carefully, and in most cases will be avoided:

    "If an organization wants to experiment with the model, and presuming they have the expertise to do so, that should be done in a complete sandbox. Sitting behind a firewall is not sufficient. The model shouldn't touch the internet. For most companies, a safer option is to make the model off limits."

    The bottom line for most business is that it will be AI business as usual. The largest AI software players in the industry, such as Microsoft, OpenAI, Amazon, Google, and Anthropic, will use the DeepSeek moment to see how to improve and drive down the prices of their products. And the AI hardware industry, while likely to see short-term pressure and questions about their prices, will see plenty of growth as AI technologies are more widely adopted.

    As for DeepSeek itself? Expect it to be in the news and under the microscope for quite some time.



    Trending Articles