DeepSeek’s AI revolution: a moment of reckoning for NVIDIA?

DeepSeek is turning heads, and for good reason. Dubbed the "Chinese ChatGPT," its R1 advanced reasoning model launched on January 20, reportedly developed in under two months. The model not only rivals but in some benchmarks outperforms OpenAI's O1 model. Within days, DeepSeek became the top app in both the U.S. and China app stores. DeepSeek's rapid development, low cost, and accessibility have sent shockwaves through financial markets, raising profound questions about the future of AI innovation, scalability, and competitive advantage.

Some have likened this to the "Sputnik Moment," referencing the Soviet Union’s launch of Sputnik 1 on October 4, 1957. The satellite’s orbit sent shockwaves through American society and its military, triggering widespread panic during the early Cold War. Fast forward to today, the AI arms race is in full swing. On January 27, alarm spread through financial markets. AI chipmakers such as NVIDIA (NVDA:US) and Broadcom (AVGO:US) experienced sharp selloffs, with both stocks dropping 17% following the DeepSeek news.

Let’s unpack why this happened and what it could mean moving forward.

Why DeepSeek matters

DeepSeek matters because it appears to show that high-performance AI can be built at low cost, raising questions about current strategies of big tech companies and the future of AI. The DeepSeek R1 model delivers performance comparable or greater than OpenAI's O1 model but at just 10% of the cost. According to DeepSeek, training the model cost $5.8 million. In stark contrast, OpenAI, valued at $157 billion as of October 2024, employs over 4,500 people, while DeepSeek operates with a lean team of just 200 staff.

DeepSeek-R1 is not just a breakthrough in technology but also a testament to the growing impact of open-source AI, making advanced tools more accessible to users and businesses. The R1 model runs efficiently on modest hardware, making it freely accessible to developers. This drastically lowers barriers to entry, fuelling innovation and enabling rapid adoption. By doing so, DeepSeek directly challenges the walled garden approach of big tech giants like NVIDIA and OpenAI. DeepSeek's ‘accelerative effect’ could prove pivotal, potentially paving the way for the next phase of AI innovation.

Big tech has banked on massive capital spending as its AI strategy. The logic was simple: more investment in computing power yields stronger models, creating a competitive moat. Companies like Meta (META:US) have doubled down on this philosophy, with plans to increase spending to $65 billion this year for AI initiatives. Morgan Stanley projects that the world’s largest tech companies will collectively spend $300 billion on capital expenditures by 2025. But perhaps this strategy now needs a rethink. DeepSeek R1 flips the script. It proves that lean, agile AI innovation can rival big budgets. For big tech’s top dogs, this is a wake-up call they can’t ignore.

The impact on NVIDIA

DeepSeek’s ability to deliver high-performance AI with significantly reduced computing requirements raises meaningful concerns for NVIDIA. If the industry begins to achieve top-tier AI performance at lower costs, using simpler hardware, and open-sourcing becomes widespread, whether through DeepSeek or other competitors, the implications for NVIDIA could be significant.

NVIDIA relies heavily on its high-end AI graphics cards, such as the H100 and Blackwell, which are essential for training large language models and powering advanced AI workloads. Reports estimate that the H100 chip costs between $25,000 and $30,000 per unit. According to Visual Capitalist, NVIDIA’s Data Centre Processors for Analytics and AI segment, where the H100 plays a critical role, grew from 25% of NVIDIA’s revenue in 2019 to 78% in 2024.

This dependence has proven extremely profitable. Some analysts estimated that the H100 could have generated $50 billion in revenue in 2024, based on expected unit shipments, with profit margins approaching 1,000% per unit. With NVIDIA's total annual revenue reaching $60.9 billion in 2024, the H100 has emerged as a key contributor to the company's significant profit growth in recent years.

However, this heavy dependence now exposes vulnerabilities. If AI models shift to requiring less specialised hardware, demand for the H100 could weaken. This would impact NVIDIA’s revenue and profit margins, creating significant risks for its business. Such a shift could also have repercussions for companies heavily invested in these chips. As of October 2024, Meta reportedly owns 350,000 H100 chips, and X owns 100,000 and Tesla owns 35,000. These companies could face challenges if the hardware becomes less relevant or cost-efficient for their evolving needs.

There are several counterpoints to the concerns outlined above, however. Despite the potential risks, the overall demand for compute power seems unlikely to decline significantly with the arrival of DeepSeek or similar new entrants. The reason being that emerging applications, increasingly complex AI models, and untapped opportunities will likely continue to push the limits of existing hardware. Additionally, reduced compute requirements could spark an expansion effect, where lower costs encourage wider adoption of AI, ultimately driving greater demand for even more infrastructure, a phenomenon often referred to as Jevons paradox.

Further, DeepSeek’s advancements may not directly undermine NVIDIA but could instead complement the broader AI ecosystem. NVIDIA has the capability to incorporate breakthroughs like those pioneered by DeepSeek into its own technologies, further enhancing its offerings. NVIDIA’s track record of adaptability, innovation, and responsiveness to market shifts has consistently positioned it as a leader in the AI space, making it well-equipped to navigate and capitalise on changes in the evolving landscape.

The impact on OpenAI

OpenAI isn’t a direct concern for most investors since it’s not publicly traded (though Microsoft owns a stake), but it remains a key indicator of broader AI trends, serving as one of the main ways consumers interact with AI. With its model now underperforming compared to DeepSeek, OpenAI is under increasing scrutiny, which could lead private market investors to reassess its valuation.

OpenAI is more directly exposed to the "DeepSeek moment" than companies like Microsoft (MSFT:US), Amazon (AMZN:US), and Meta. The difference lies in their core businesses, these tech giants have vast revenues that are unrelated to AI. While AI is being integrated into their offerings, it represents only a small component of much larger ecosystems. In contrast, NVIDIA, as the "picks and shovels" provider of the AI boom, and OpenAI, as a company focused solely on AI models, are far more exposed to market shifts. The market's reaction to DeepSeek appears to reflect this distinction. Meta and Amazon surprisingly saw slight gains following the DeepSeek developments, while Microsoft fell just 2%.

OpenAI’s actions in the coming weeks and months will be critical to monitor, as they may reveal the broader implications of DeepSeek’s rise. The key question is whether OpenAI can maintain its footing—not just against DeepSeek, but in a future filled with hundreds or even thousands of competitors offering faster, cheaper, and more advanced AI.

Scepticism grows

Another angle worth considering is whether DeepSeek’s accomplishments might be overstated. Scepticism is growing within the AI sector, with some questioning the legitimacy of the lab's reported breakthroughs. Alexandr Wang, CEO of Scale AI, voiced his doubts in an interview with CNBC, though he offered no evidence to support his claims. “My understanding is that DeepSeek has about 50,000 H100s, which they can’t talk about, obviously, because it is against the export controls that the United States has put in place,” he said. Wang’s comments directly challenge DeepSeek’s claims of having trained its model on a shoestring budget. Whether these claims hold water or not, there is the possibility that the selloff in reaction to DeepSeek’s emergence has been exaggerated.

What comes next

In July 2024, I shared some thoughts with clients and readers of the ASX newsletter in an article titled Decoding the AI Landscape: 10 Considerations for Investors. Reflecting on it now, a few key factors and questions stand out as increasingly critical and are ones I will be monitoring closely in the future:

• Form factor: “Just as computers have shrunk from the size of a room to the size of our hand, AI hardware could also evolve over time.”

• Open source: “Could AI become so ubiquitous that it trends toward not being ‘owned’ by anyone, just like the internet?”

• Competition: “Competition is pivotal because, in the world of AI, one company’s software update can put another company out of business.”

That said, the future of AI remains far from settled. Its evolution, impact, and ultimate value to society and investors are still unclear. The emergence of DeepSeek highlights the need to stay adaptable and open to change in a landscape as dynamic and unpredictable as AI.

Find out more about how you can trade on the 'magnificent seven' US stocks, including NVIDA, before the main market opens.

This article provides general information only. It has been prepared without taking account of your objectives, financial situation or needs. It is not to be construed as a solicitation or an offer to buy or sell any financial instruments, or as a recommendation and/or investment advice. It does not intend to support an investment decision and it should not be relied upon by you in evaluating the merits of investing in any financial instruments. You should consider your objectives, financial situation and needs before acting on the information in this article. CMC Markets believes that the information in this article is correct, and any opinions and conclusions are reasonably held or made on information available at the time of its compilation, but no representation or warranty is made as to the accuracy, reliability or completeness of any statements made in this article. CMC Markets is under no obligation to, and does not, update or keep current the information contained in this article. Neither CMC Markets nor any of its affiliates or subsidiaries accepts liability for loss or damage arising out of the use of all or any part of this article. Any opinions or conclusions set forth in this article are subject to change without notice and may differ or be contrary to the opinions or conclusions expressed by any other members of CMC Markets.