From Sputnik moment to price war

When DeepSeek released its R1 model in early 2025, Western tech giants experienced a shock. The Chinese startup managed to train a model comparable to GPT-4 for less than $6 million — while OpenAI invested billions into its models. According to Bloomberg, DeepSeek "awakened a long-dormant Chinese tech industry."

The effect was immediate. Within months of the R1 launch, companies like Alibaba (Qwen series), Baidu (Ernie), ByteDance, Tencent, and dozens of others began releasing their own models. Each of them bet on the same strategy: an open model at a price that makes no sense for Western competitors.

Numbers that hurt: DeepSeek versus OpenAI

A comparison of current API prices (May 2026) speaks clearly:

DeepSeek V4 Flash: $0.14 per million input tokens, $0.28 per million output tokens. 1 million token context window. Open-source.

OpenAI GPT-5.4 mini (OpenAI's cheapest model): $0.75 per million input tokens, $4.50 per million output tokens.

OpenAI GPT-5.5 (flagship): $5.00 per million input tokens, $30.00 per million output tokens.

Simple math: output tokens from DeepSeek V4 Flash are 16× cheaper than GPT-5.4 mini and 107× cheaper than GPT-5.5. For developers who consume billions of tokens per month, this means the difference between a bill of hundreds of dollars and a bill of tens of thousands.

And that's not all. The DeepSeek V4 Pro version — currently at a 75% discount — offers performance comparable to the best OpenAI models at a price of $0.87 per million output tokens. Even after the promotional pricing ends (May 31, 2026), it will cost $3.48 — still nearly 9× less than GPT-5.5.

Xiaomi enters the ring

Xiaomi, best known for manufacturing smartphones, surprised the market in April 2026 by launching its model MiMo. It's a massive trillion-parameter model that Xiaomi offers at prices even more aggressive than DeepSeek. According to available information from DigiTimes, MiMo's deployment "puts it on a collision course with OpenAI."

For Xiaomi, entering the LLM market makes strategic sense: the company wants to integrate AI into its ecosystem of smart devices, from phones to electric vehicles to home appliances. Having its own model means independence from foreign suppliers and zero licensing costs.

Why are Chinese models so cheap?

Several key factors are at play. First, technical efficiency — DeepSeek proved that models can be trained with significantly less computing power using the Mixture of Experts (MoE) architecture, which activates only part of the neural network for each query.

Second, the open approach. Most Chinese models are open-source — anyone can download and run them on their own hardware. This avoids ongoing API fees.

Third, state support and a massive domestic market. The Chinese government is investing hundreds of billions of yuan into AI infrastructure. Companies like Alibaba and Tencent operate cloud services where AI models are a natural part of the ecosystem.

And finally: Chinese companies face no pressure for immediate profitability in AI. While OpenAI must generate revenue for investors, Chinese players can afford to keep prices at minimum levels longer — the goal is to gain market share.

Impact on Nvidia and the chip market

Paradoxically, US export restrictions on advanced chips have actually fueled Chinese innovation. When China had no access to the latest Nvidia GPUs, it had to learn to optimize. DeepSeek V4 runs not only on Nvidia H800 chips but also on domestic Huawei Ascend processors.

For Nvidia, this poses a dual challenge: Chinese customers make up a significant portion of revenue, but at the same time, Chinese companies are increasingly turning to domestic alternatives. According to Bloomberg, the fact that Chinese models require fewer chips also means lower demand for hardware overall.

What does this mean for the Czech Republic and Europe?

For Czech companies and developers, the AI price war is unequivocally good news. The costs of integrating language models are falling, and competition is driving price reductions across the entire market.

The European Union, however, faces a dilemma. On one hand, the EU AI Act requires strict regulation of AI systems, including transparency of training data. Chinese models — which often do not meet European data protection requirements — may thus not find their way into the corporate sector. On the other hand, if European companies remain dependent on more expensive American models, they will lose competitiveness.

DeepSeek V4 is also available in Czech — the model handles dozens of languages, including Czech, at a very decent level. For the average Czech user, this means that a quality AI assistant is now free through the DeepSeek chat interface, while ChatGPT in its free version offers a more limited model.

Where is the market heading?

The price war is far from over. While OpenAI recently reduced the price of GPT-5.5 Instant as the default model for ChatGPT, DeepSeek counters by extending the 75% discount on V4 Pro. Xiaomi's MiMo pushes prices even lower.

For developers and companies, the key takeaway is that API price is no longer the primary decision criterion — instead, it becomes quality, speed, language support, and regulatory compliance. And in this environment, space opens up for European players who can offer models fully compatible with the EU AI Act.