How China’s DeepSeek R1 Model Will Disrupt the AI Industry

•
January 28, 2025    

DeepSeek’s efficiency breakthrough will have a significant but not seismic impact on the AI supply chain as lower training costs spur broader AI adoption.

See Also: Uncovering Risk With Social Due Diligence

The China-based company has proven in recent days that state-of-the-art models can be trained at dramatically lower prices, challenging the notion that only billion-dollar budgets can deliver cutting-edge AI performance. DeepSeek proved that cutting-edge hardware isn’t necessary for high performance, and their work could allow researchers and smaller organizations to build and deploy competitive models.

“To see the DeepSeek new model, it’s super impressive in terms of both how they have really effectively done an open-source model that does this inference-time compute, and is super-compute efficient,” Microsoft CEO Satya Nadella said Wednesday at the World Economic Forum in Davos, Switzerland.

Initial reaction to DeepSeek’s breakthrough included fears that demand for AI infrastructure like GPUs and data centers could decline, but many analysts argue that cheaper AI could drive more adoption and increase long-term demand. By lowering the cost of training and inference, DeepSeek could actually expand the use of AI across industries, requiring more infrastructure to support the increased demand (see: DeepSeek’s New AI Model Shakes American Tech Industry).

How Nvidia, AMD Will Fare Against DeepSeek

DeepSeek’s early success challenges Nvidia’s narrative that the most cutting-edge GPUs are required for state-of-the-art AI models, which could slow demand for the company’s most expensive hardware like the H100 or Blackwell chips. But DeepSeek’s use of Nvidia’s H800s shows Nvidia’s relevance even in export-restricted markets like China given how GPUs are used for model training and interference.

Nvidia’s stock fell 15.4% in trading Monday to $120.18 per share, which is the lowest the firm’s stock had traded since Oct. 2, 2024. The company’s stock recovered Tuesday, with shares jumping 7.65% to $127.48 per share. That pales in comparison to CrowdStrike, which saw its stock price drop more than 36% in the weeks after its July 19 outage but has since recovered and seen its stock reach record highs.

“Wall Street will view DeepSeek as a major perceived threat to US tech dominance and owning this AI Revolution,” Wedbush analyst Daniel Ives told investors “While the model is impressive and it will have a ripple impact, the reality is that Mag 7 and U.S. tech is focused on the AGI endgame with all the infrastructure and ecosystem that China and especially DeepSeek cannot come close to in our view.”

Meanwhile, AMD could capitalize on DeepSeek’s ability to deliver competitive AI performance without top-tier Nvidia GPUs given that the former’s GPUs are more affordable than Nvidia’s but were seen as less performant. But now that DeepSeek’s training architectures allow cheaper GPUs to compete with Nvidia’s high-end chips, AMD’s lower-cost GPUs now have broader user cases.

AMD’s stock fared better than Nvidia’s Monday, dropping just 5.8% to $115.75 per share, though that represents the company’s lowest stock price has traded since Nov. 3, 2023.

“We think model developers will look to incorporate some of [the DeepSeek] R1’s novel techniques into their own models which would help to improve efficiency,” UBS Analyst Timothy Arcuri wrote in an investor note Monday. “Even though this may read negatively for compute demand, the fact remains that compute continues to drive model performance even as models become more efficient.”

What DeepSeek’s Entry Means for Microsoft, Meta, OpenAI

Microsoft Azure can support both OpenAI’s proprietary advancement as well as DeepSeek’s open-source developments, allowing Azure to dominate in AI infrastructure regardless of which models gain traction. The company’s AI products like Office 365 and GitHub Copilot rely on large, generalized models, which remain unthreatened by DeekSeek’s narrow reasoning capabilities.

“We expect Microsoft to begin demonstrating more data center scale from here,” wrote Bank of America analyst Brad Sills. He added that he expects Azure to accelerate in the second-half of fiscal 2025 and offer up a better margin outlook, “largely from datacenter efficiencies.”

DeepSeek’s open-source model poses a direct challenge to OpenAI’s closed-source models and Meta’s Llama, threatening their dominance in developer-focused AI ecosystems since developers now have an alternative that costs significant less to use. Meta’s Llama models were previously seen as the leading open-source alternative, but DeepSeek’s efficiency and performance are beginning to erode its lead.

“To people who see the performance of DeepSeek and think: ‘China is surpassing the U.S. in AI, you are reading this wrong,” Meta Chief AI Scientist Yann LeCun wrote on LinkedIn. “The correct reading is: ‘Open source models are surpassing proprietary ones.'”

DeepSeek’s open-source model licensed under MIT has implications for democratizing AI, allowing researchers, universities and smaller companies to study, modify and expand the company’s R1 model. Open-source competitors like R1 may erode market share for proprietary platforms like OpenAI’s GPT-4, particularly in the developer and research communities.

“DeepSeek R1 is AI’s Sputnik moment,” Venture Capitalist Marc Andreessen wrote on X Sunday.