DeepSeek: A Flash or a Major AI Tipping Point?

KResearch Analysis Spotlight: “DeepSeek”: A Temporary Trend or a Major Turning Point in AI? Who Benefits or Is It Affected? What Will the Future Hold?
Dr. Krit Seetathani, Assistant Managing Director of Kasikorn Research (KResearch), revealed an analysis of DeepSeek, titled "A Fleeting Trend or a Major Turning Point in AI." He began by explaining DeepSeek as the latest AI technology company founded by the Chinese fund High-Flyer. It garnered significant attention after the launch of its new AI, DeepSeek-R1, resulting in a 16.91% drop in Nvidia's stock price in a single day (US$6 billion, equivalent to Thailand's GDP).
The reason DeepSeek has garnered so much attention stems from the vast cost difference between DeepSeek's AI development and OpenAI's. OpenAI spent $7 billion USD in 2567 to develop its latest AI models, while DeepSeek spent only $5.6 million USD. However, DeepSeek's AI can perform nearly as well as OpenAI's at a significantly lower cost.
DeepSeek's total cost is expected to be higher than previously announced
Kasikorn Research Center forecasts that DeepSeek's true total cost is higher than the frequently reported figure of US$5.8 million, as this figure does not include costs associated with prior research, algorithm development, or data. However, the actual cost remains lower than the multi-billion dollar annual investment by AI companies in the United States.
DeepSeek also reported that it only used 800 Nvidia H2,000 chips, which is likely an understatement, as DeepSeek’s CEO previously announced that it had purchased between 100 and 800 Nvidia A10,000 and H50,000 chips, making DeepSeek one of the world’s largest Nvidia chipholders.
Stock market response
US AI technology companies were less directly affected, while AI infrastructure companies such as data center builders and microchip makers saw their stocks fall significantly. Although Nvidia shares fell 16.91% on concerns about AI infrastructure, shares of companies focused on AI software such as Apple, Amazon and Meta rose, reflecting the market’s continued belief in software over infrastructure. The hardest hit companies were energy companies related to data centers, down 18.8-21.0%, and data center service providers, down 7.8-12.0%.

However, the stock market remains bullish on the US AI capabilities, but the market correction comes as the realization that billions of dollars in investment in microchips and power for AI clusters may no longer be necessary.
The Future of DeepSeek
Many view the launch of DeepSeek-R1 as a "Sputnik Moment" for the US AI industry, or even the end of the US's AI dominance. However, Kasikorn Research Center predicts that while DeepSeek may not yet surpass OpenAI in terms of capability, it will serve as a wake-up call for the US.

DeepSeek is likely to lead to the emergence of more small-scale AI, as its open-source model and ability to work with lower-cost mid-range microchips will be a concern for the US.
However, there is a key question to watch: does DeepSeek have a chance to capture market share? No Chinese AI model has captured significant market share at the moment, which will be the real test of DeepSeek’s success, not just benchmark results or low cost claims.
But while DeepSeek-R1 may not immediately compete with the highest-end models, its reduced cost could make AI more accessible to smaller companies, intensifying competition across many industries. AI that is “good enough” can be as disruptive to the market as the most advanced AI, especially if there are AI-infused products like smartphones or military drones, where the total cost of ownership could be significantly reduced.
Regarding those who will benefit and those who will be negatively impacted by DeepSeek-R1, Kasikorn Research Center predicts that if DeepSeek's investment proceeds as announced, it will affect the entire supply chain, primarily benefiting those downstream and negatively impacting those upstream of the AI supply chain.
-
- Winners
The beneficiaries of this change are downstream companies that have started to use AI in their business, such as e-commerce and marketing, where the share of AI usage is expected to increase from 65% to 70-80% within 2-3 years.
In addition, Chinese tech companies such as DeepSeek, Alibaba, and ByteDance are expected to increase their share of the AI service market from less than 10% to 25% within 1-3 years, as lower AI development costs allow them to compete aggressively in the global market and reduce their reliance on US technology.
Another group that will benefit is organizations and developers who focus on open-source AI development. Smaller organizations such as startups will have access to higher-potential AI, significantly intensifying competition in industries such as manufacturing, logistics, finance, and health technology.

-
- Negatively Affected Group (Losers)
The sectors that may face the biggest problems are manufacturing and data center service providers, especially data center energy providers whose stock prices fell 18.8-21.0% in a single day after DeepSeek-R1 was released. This is because of long-term investment businesses such as Talen, which invests in nuclear power plants for data centers.
Additionally, companies building Hyperscale Data Centers or AI Clusters 10 face challenges as customers may turn to lower-cost solutions such as On-Premise AI or General Cloud Solutions instead.
Another industry that will be impacted is high-end chipmakers like Nvidia, ASML, and TSMC, as the AI market shifts from a competition for processing power and speed to one of resource efficiency, making 2nm chip fabrication unprofitable for the $3 billion invested.
International political implications
The China-US tech war is likely to intensify, leading to changes in AI policies in both countries.
-
- China will move towards becoming an AI leader
China is expected to use this opportunity to push forward its Next Generation AI Development Plan, which aims to make the country a global AI innovation hub by 2030. Leading companies like Alibaba are already launching new AI models (such as Qwen 2.5), although these products may not yet be able to match DeepSeek’s cost advantages.
It remains to be seen whether DeepSeek can truly rely on low-cost microchips, especially given concerns about the potentially higher-than-expected cost of training AI and China’s restrictions on access to top-tier microchips.
Another point is that China’s data center energy consumption will continue to decline. Data center energy consumption in China is expected to grow by 6.56% between 2025 and 2030, reaching 4.8 gigawatts (GW) by 2030. However, if DeepSeek’s AI models can truly scale, this number could be significantly reduced.
-
- US policy and export controls
President Trump is expected to take action to maintain his leadership in AI, focusing on increasing investment in AI research and development rather than increasing regulation.
While additional export controls may be announced, the actual impact may be limited, as many of Nvidia’s high-end microchips are already subject to CHIPS Act and existing export controls.
If DeepSeek’s approach can truly reduce reliance on cutting-edge hardware, additional controls may not have much of an impact on China’s progress.
Overall, the debate over DeepSeek’s true cost and ability to compete with top-tier models could shape the geopolitical landscape of AI in the coming years. While DeepSeek-R1 may not live up to all expectations, the mere possibility of building high-performance AI with fewer resources could spark competition, encourage open-source development, and help expand AI adoption beyond the traditional framework of American technology companies.































