The meteoric rise of the Chinese AI start-up DeepSeek has sparked global conversations around the future of artificial intelligence and its geopolitical stakes. While DeepSeek has disrupted the landscape and challenged the U.S.’s dominance in AI, it’s far from signaling Silicon Valley’s demise.
DeepSeek’s emergence is not a fluke — it’s a natural consequence of a globally connected, yet increasingly fragmented, AI environment. Its recent achievements have led to three major misconceptions that need addressing.
First is the idea that U.S. semiconductor export controls have failed. This belief surfaced after DeepSeek-R1 — a reasoning-based AI model — was released. However, those controls were never intended to halt China’s progress outright. Their goal was to set a technological ceiling — a limit that has now become more visible.
Although investigations are ongoing to determine whether DeepSeek obtained restricted chips through illicit means, the bulk of its training was done using Nvidia H800s. These chips were made specifically for China following the 2022 restrictions, though they too are now under tighter regulation. DeepSeek’s ability to perform well with limited compute might warrant policy revisions, but the current controls are beginning to take effect and could soon slow future progress, especially as China races to scale up its chip manufacturing capabilities.
The second myth is that restricting model exports will halt China’s momentum in AI. Following DeepSeek’s breakthrough, U.S. Senator Josh Hawley proposed legislation to ban AI model sharing with China entirely. However, enforcing such measures would require strict identity checks and compliance protocols, which would not only stifle U.S. innovation but also prove ineffective against digital leakage.
Unlike hardware, model weights and data can cross borders effortlessly — via downloads or leaks — with no need for physical smuggling. As models like DeepSeek and the Alibaba-supported Qwen become more competitive, China’s reliance on U.S. open-source models will diminish. Overregulating exports could push global developers toward Chinese alternatives. While DeepSeek faces criticism for censorship, many developers prioritize performance and cost over national origin.
To maintain leadership, the U.S. should instead promote responsible openness. Encouraging safe deployment strategies — such as tiered releases and external audits — can allow innovation to flourish while minimizing risk.
The third flawed belief is that Silicon Valley’s once-imposing lead, backed by investments in chips, talent, and energy infrastructure, has eroded. DeepSeek’s reported training costs of just $6 million — about 1% of the budget for Meta’s Llama model — initially alarmed industry watchers.
But that figure only accounts for GPU costs for a single pre-training run. It doesn’t reflect the full model development expense. More importantly, DeepSeek’s innovations are deeply rooted in Silicon Valley’s own breakthroughs. Google’s mixture-of-experts model and OpenAI’s test-time compute techniques laid the groundwork. DeepSeek also benefited from using ChatGPT outputs to train its own systems, a process known as distillation.
Though DeepSeek laid a plank over Silicon Valley’s moat, Silicon Valley is already digging that moat wider. American tech giants will likely adopt and enhance DeepSeek’s open-source innovations, leveraging their superior compute resources. European players may follow suit, potentially reshaping the global AI landscape.
DeepSeek’s success wasn’t unexpected. It reflects a dynamic, competitive, and global knowledge economy. China’s heavy investment in domestic talent — over 2,300 AI undergraduate programs since 2018 — is showing real results. With fewer Chinese researchers moving to the U.S., China’s internal AI capacity will only grow.
Now, China faces a new challenge: effective implementation. While the country leads in open-source development, the commercial deployment of models often suffers from hype-driven initiatives and misaligned incentives. Some DeepSeek applications have been dismissed as publicity stunts. How China closes this implementation gap will influence the future of its AI trajectory.
Despite rising efforts to decouple, the U.S. and Chinese open-source ecosystems remain deeply connected. Even if each country focuses on its own tools, innovation will continue to flow across borders.
Compute power isn’t the only factor, but the United States still holds the bigger shovel — and the capacity to dig faster and deeper.
The real risk isn’t competition from China — it’s complacency. The U.S. must act with both urgency and responsibility to retain its leadership in this fast-evolving field.
Source: eastasiaforum