DeepSeek Raises Alarms for US Tech Giants: What’s Next?

TradingKey - The recent release of DeepSeek’s AI model has triggered intense discussions and market volatility. On one hand, US investors and tech executives are beginning to question whether the massive capital expenditures on AI by major US tech firms is truly justified. On the other hand, Chinese tech companies are regaining attention for their innovation capabilities.  

"Spending on anything AI has got the market thinking: ‘Geez, how much more has to be done before we see capital expenditures reduced, instead of increasing by tremendous amounts?’” said Robert Pavlik, senior portfolio manager at Dakota Wealth Management. “When is enough, enough?”  

The market has reacted accordingly. Year-to-date, NVIDIA (NASDAQ: NVDA) has dropped by 8%, while Alibaba (NYSE: BABA) has surged by more than 31%.  

Source: TradingView; NVDA,BABA,PDD,JD and BIDU YTD Performance

 What Makes DeepSeek Stand Out?  

DeepSeek’s latest model has delivered impressive performance on multiple industry benchmarks, even surpassing OpenAI in certain areas. More importantly, DeepSeek V3 achieves this level of performance using H800 GPUs, which are not Nvidia's most advanced GPUs currently available.

According to TradingKey analyst Viga Liu, DeepSeek has introduced several key innovations, including Mixture of Experts (MoE), Matrix Low-Rank Approximation (MLA), Memory-Tensor Parallelism (MTP), and Reinforcement Learning (RL), which have redefined AI model development.  

MoE significantly improves efficiency by activating only a subset of parameters instead of processing all information like traditional large models. DeepSeek has also developed MLA, which uses low-rank decomposition to compress storage matrices, significantly reducing cache size and computational complexity.

In addition, DeepSeek introduces a novel post-training approach—reinforcement learning. This method challenges traditional scaling laws, demonstrating that innovative training techniques can achieve significant performance gains at a given scale, improving resource efficiency.

DeepSeek’s innovations drive breakthroughs and cost-effective AI strategies, appealing to investors and professionals.  

DeepSeek Challenges US AI Strategies  

The U.S. tech industry has invested heavily in artificial intelligence, but many observers have questioned the direction of this investment.  

For years, US tech giants have been focused on scaling AI models, believing that increasing data input and computing power is the key to better AI performance. Some have even argued that “scaling is all you need.”  

While DeepSeek follows a similar approach, it places greater emphasis on reinforcement learning, MoE (which rely on multiple smaller, more efficient models), model distillation, and refined reasoning techniques. This contrast highlights the groupthink within the U.S. AI industry, raising concerns among experts about whether there are even more significant blind spots in America's approach to artificial intelligence.

What Does This Mean for NVIDIA?  

DeepSeek’s efficiency has sparked fears about the long-term demand for high-end AI chips, leading to a decline in NVIDIA’s stock price. If AI companies shift their focus toward algorithmic optimization rather than sheer computing power, demand for NVIDIA’s GPUs could slow down.  

Morgan Stanley recently lowered its shipment forecast for NVIDIA’s GB200 chips from 30,000-35,000 units to 20,000-25,000 units, which could wipe out $30-35 billion in market value.  

Furthermore, Liu pointed out that DeepSeek may reduce reliance on NVIDIA’s CUDA ecosystem. By using custom PTX (Parallel Thread Execution) instructions and optimizing communication block sizes, DeepSeek has found ways to interact directly with GPU hardware, bypassing some of CUDA’s high-level interfaces. Since CUDA has long been considered a key part of NVIDIA’s competitive advantage, any ability to work around it could pose risks to the company’s long-term GPU demand.  

Boost for Chinese Internet Stocks  

DeepSeek’s rise has also shifted investor sentiment toward Chinese tech stocks. Many now see China’s AI sector as a serious competitor in the global AI race, leading to renewed interest in companies like Alibaba, Baidu (NASDAQ: BIDU), Pinduoduo (NASDAQ: PDD), and JD.com (NASDAQ: JD).  

Jian Shi Cortesi, portfolio manager at GAM Investment Management, commented:  "This renewed attention could shift investor focus from ‘thinking about risks’ to ‘thinking about potential’ when evaluating large Chinese internet companies. This is positive, especially given the big valuation discount of these names compared to US peers.” 

 AI Stocks to Watch

For a long time, investors have primarily focused on AI infrastructure—chips, cloud computing, and data centers—while remaining skeptical about the commercial potential of AI applications. However, DeepSeek’s advances may accelerate AI adoption.  

While DeepSeek’s efficiency improvements could challenge AI infrastructure providers, the overall demand for AI remains uncertain. If AI applications continue to expand and drive broader adoption, computing demand could still increase despite improvements in model efficiency.  

Here are stocks related to AI applications.

AI Content & Advertising: Meta (META), Snapchat (SNAP), AppLovin (APP)

AI Enterprise Service: Palantir (PLTR), C3 AI (AI)

AI Productivity: Adobe (ADBE), Salesforce (CRM), Autodesk (ADSK)

AI-Driven Big Data: MongoDB (MDB)

AI-Powered Entertainment: Netflix (NFLX), Disney (DIS), Spotify (SPOT), Roblox (RBLX)

AI in Healthcare: Intuitive Surgical (ISRG), Bullfrog AI (BFRG)

AI in Autonomous Vehicles: Tesla (TSLA), Uber (UBER)

AI in Finance & Payments: Intuit (INTU), PayPal (PYPL)

AI in Smart Home & IoT: Amazon (AMZN), Apple (AAPL), iRobot (IRBT)