Five Key Questions: The Impact of DeepSeek’s Rise on the AI Industry

As DeepSeek rises rapidly, this Chinese AI startup has sparked intense debate over cost, technological innovation, and market dynamics. We spoke with several AI industry experts in China to explore how DeepSeek is shaping competition and the development of AI technology. Here are the five most pressing questions:

1. Does DeepSeek’s rise signal an “AI Sputnik Moment”? How is the market evaluating its real cost versus official cost?

Labeling DeepSeek’s rise as an “AI Sputnik Moment” is likely an exaggeration from media and an overreaction from the market. While DeepSeek has indeed made breakthroughs in training and inference costs, it has not achieved a disruptive expansion of capability boundaries nor altered the trajectory of AI development. Compared to OpenAI, DeepSeek’s real impact is lowering AI entry barriers rather than redefining AI capabilities.

DeepSeek’s innovations are not entirely original but rather refinements of existing technologies. For instance, its key breakthroughs—Mixture of Experts (MoE) architecture and MLA mechanisms—have been used in other models before. DeepSeek optimized these technologies to improve computing efficiency, making them more practical. Unlike China’s market, where computational resources are constrained, U.S. AI companies face fewer such limitations, making DeepSeek’s efficiency-focused approach less urgent outside China.

Despite this, DeepSeek’s success represents an important evolution in AI: refining compute efficiency rather than simply scaling up resources. Historically, AI progress has been driven by the “Scaling Law”—increasing data and computational power to improve model performance. However, AI firms, including OpenAI, are now prioritizing cost control over infinite model expansion.

DeepSeek officially claims its training cost is only $5.57 million, but industry experts point out that this figure only accounts for the final training phase. It excludes R&D, data labeling, testing, and experimentation. When these costs are included, the real development cost is estimated at around $100 million—far higher than the official number. However, even at this higher estimate, DeepSeek’s cost efficiency is evident. OpenAI’s single training run costs about $60 million, while DeepSeek’s is roughly one-tenth of that. OpenAI has spent about $500 million on training overall, while DeepSeek’s spending remains a fraction of that, highlighting its cost advantage.

Experts believe that Google and Meta will continue to lead the frontier development of general-purpose large models, while DeepSeek is more likely to excel in specific vertical domains. These fields often require more than just computing power to achieve breakthroughs—they depend on high-quality, domain-specific training data and well-defined application scenarios.

2. Can DeepSeek survive long-term and achieve profitability in a competitive market?

Although DeepSeek announced during “Open Source Week” that its theoretical cost-profit margin is around 545%, the actual profit level in real-world applications falls far short of this figure and has yet to create a significant profit margin. Since the launch of DeepSeek V2, China’s AI industry has seen an intensified price war, with competitors slashing prices in response—even impacting OpenAI’s pricing strategy. Since DeepSeek itself does not rely on inference-side profits and primarily aims to expand its market influence, its API price war appears to be more about building market appeal rather than generating immediate revenue.

DeepSeek’s current strategy is to first expand its user base and build market influence through an open-source approach before seeking specific monetization and commercialization models in the future. This has intensified the price war in China’s AI services market.

While this strategy can boost market share in the short term, DeepSeek will ultimately need to establish a viable monetization path or a more sustainable revenue model in the long run. Potential paths include:

• Premium enterprise services such as tailored AI solutions for finance, healthcare, and e-commerce.
• Collaborations with large companies for co-developing AI applications.
• Launching its own cloud computing services to reduce dependence on third-party providers.

Chinese tech giants and startups integrating DeepSeek have seen a positive impact on stock prices and investor sentiment. However, experts point out that for some of these companies, the integration is merely a marketing gimmick, and its actual effectiveness in driving business growth remains uncertain.

3. Who are the biggest winners and losers from DeepSeek’s rise?

Winners:

• AI application developers & startups: DeepSeek’s open-source model allows businesses to deploy AI locally at lower costs, reducing reliance on expensive proprietary APIs.
• Chinese GPU manufacturers & AI hardware suppliers: DeepSeek’s efficient inference capabilities make non-NVIDIA AI hardware more viable, attracting customers to domestic alternatives like Huawei GPUs.
• AI compute leasing providers: Companies offering on-premise AI computing benefit as firms choose localized DeepSeek deployments over cloud-based solutions.
• Cloud service providers (short-term): Alibaba Cloud, Tencent Cloud, and others benefit from increased cloud-based DeepSeek deployments. However, in the long run, DeepSeek’s open-source nature may erode their market dominance.
• Open-source AI community: DeepSeek strengthens the position of open-source AI, accelerating industry-wide adoption and challenging closed AI models from companies like OpenAI and Anthropic.

Losers:

• Proprietary AI model providers (e.g., OpenAI, Anthropic): Businesses may move away from expensive GPT APIs in favor of open-source alternatives.
• Chinese AI firms developing proprietary models (e.g., Baidu, ByteDance): DeepSeek’s competition may reduce their market share.After integrating DeepSeek, theythey might even have to revert to the old mobile internet era, competing on data, user scenarios, and marketing.
• High-cost AI API providers (e.g., Google Cloud, Azure, AWS): As businesses adopt local AI models, demand for costly cloud-based AI services may decline.
• Chinese AI Startups: The rise of DeepSeek’s models has led to decreased market attention for leading AI startups, such as the “Six Little Tigers” of AI. This could result in lower demand for their model API integration, reduced token usage, and fewer custom AI projects.
• Foreign cloud providers: DeepSeek’s cost-effective AI computing weakens the competitive edge of Western cloud firms that rely on high-end AI infrastructure.

4. How will DeepSeek’s rise impact NVIDIA’s dominance in the AI hardware market?

DeepSeek’s latest technological breakthroughs have sparked discussions about shifts in AI hardware demand, particularly regarding NVIDIA’s long-term competitive advantage. However, experts believe that while DeepSeek has improved inference efficiency, high-performance AI applications—such as reinforcement learning for inference and large-scale coding models—are driving increased demand for computing power. This growth in demand could offset the potential cost-saving effects brought by DeepSeek and further boost the need for high-performance GPUs.

Industry experts argue that the truly commercially valuable models remain the 671B full-scale versions of V3 and R1. Deploying such expert-level models typically requires high-end hardware, such as H800, H100, and H200, and relies on NVLink protocols, further expanding the demand for computing power. In particular, as consumer-facing (to-C) applications grow, the resulting increase in computational consumption far exceeds the reduction achieved through algorithmic optimization. Experts further emphasize that to retrain and optimize DeepSeek’s models, the minimum technical threshold has now risen to between 512 and 1,024 GPUs, reaching the scale of thousands to tens of thousands of computing cards. Consequently, setups with fewer than 64 or even 32 GPUs are no longer sufficient to meet these technical requirements.

In fact, NVIDIA’s recent financial performance indicates that its hyperscaler business remains strong. Q4 sales of Blackwell reached $11 billion, accounting for nearly one-third of total data center revenue, signaling a significant easing of supply chain constraints. Over the next few quarters, as the Blackwell architecture continues to advance, experts expect NVIDIA’s data center business to maintain double-digit quarter-over-quarter growth.

However, as the AI market shifts toward inference, NVIDIA is facing increasing competitive pressure in this space. While NVIDIA still holds approximately two-thirds of the AI inference market, hyperscaler customers are gradually increasing their adoption of custom-designed ASIC chips. Over the next 2–3 years, NVIDIA’s market share in this segment could decline to 50%.

Global investment in AI infrastructure is expected to continue growing strongly. Major cloud service providers, including Google, Microsoft, and Meta, are planning to increase their AI-related capital expenditures, a trend projected to last until at least 2026. Meanwhile, the rising participation of small and medium-sized enterprises (SMEs) is driving demand for mid-range and entry-level GPUs, with some companies opting to rent cloud-based computing power, further accelerating the growth of the general-purpose GPU (GPGPU) market.

5. Can DeepSeek transform from a disruptor into an AI giant, or is it just a fleeting phenomenon?

DeepSeek’s success in optimizing compute efficiency and lowering inference costs has made large AI models more accessible. However, it has yet to push the boundaries of AI capabilities, and its future faces significant challenges:

• Policy risks: As a Chinese AI company, DeepSeek faces significant policy risks. The U.S. has imposed strict export controls on advanced AI chips to China, limiting DeepSeek’s access to high-performance GPUs and thereby restricting its AI training and deployment capabilities. Additionally, due to data privacy and security concerns, several countries have banned the use of DeepSeek’s AI technology, posing challenges to its global business expansion.
• Market shifts: As AI hype subsides, sustaining market interest will require clear application scenarios and a profitable business model. Competition from more powerful models and lower-cost specialized AI solutions could pose a threat.
• Technological innovation & competition: Tech giants like Google and Meta continue to invest heavily in AI research. DeepSeek’s ability to keep up with technological advancements remains uncertain. Whether it can transition from aggressive pricing tactics to true technological innovation and real-world applications will determine its long-term fate.