In a move that signals a tectonic shift in the global AI supply chain, Hangzhou-based startup DeepSeek launched a preview of its highly anticipated V4 model on Friday, April 24, 2026.
The launch is a landmark for two reasons: it marks the company’s formal departure from its past reliance on Nvidia hardware, and it serves as a powerful demonstration of Huawei’s growing capability to power frontier AI. This transition follows months of geopolitical tension and accusations from the White House regarding the “industrial-scale” theft of US intellectual property—claims that DeepSeek and the Chinese government have flatly denied.
Technical Performance: Huawei Silicon Meets Open-Source Prowess
DeepSeek, owned by High-Flyer Capital Management, shocked the world in 2025 with its hyper-efficient V3 model. The V4 appears to double down on that efficiency, specifically optimized for Huawei Ascend AI processors.
The Benchmarks
DeepSeek claims the “Pro” version of V4 is now the world-leading open-source model in world-knowledge benchmarks.
- The Competition: V4 currently trails only Google’s Gemini 3.1 Pro—a closed-source model—in global reasoning tests.
- The Flash Version: Alongside the Pro model, DeepSeek released a “Flash” version designed for high-speed, low-cost API calls, targeted at developers looking to scale without the “Nvidia Tax.”
Huawei Integration
Huawei confirmed that its entire Ascend supernode product line now fully supports the V4 series. This deep integration suggests that Chinese domestic hardware has reached a level of maturity where it can not only run but also train frontier-level Large Language Models (LLMs) effectively.
Geopolitical Friction: IP Theft and Export Controls
The V4 launch occurred just 24 hours after the White House accused China of violating export controls and “distilling” proprietary models from OpenAI and Anthropic.
| Accusation | Source | DeepSeek Response |
|---|---|---|
| IP Theft | White House / US Labs | Claims are “baseless”; focus is on web-crawled data. |
| Model Distillation | OpenAI / Anthropic | No intentional use of synthetic data from US rivals. |
| Export Violations | Washington | Acknowledges Nvidia use; silent on ban status of specific chips. |
Export to Sheets
This “chip independence” push is a direct result of US restrictions started in 2022. By moving to Huawei, DeepSeek is attempting to insulate itself from future US sanctions while proving that Chinese AI can thrive in a decoupled ecosystem.
Market Impact: The $20 Billion Valuation
The release of V4 sent shockwaves through the Chinese tech market. While DeepSeek soared, domestic rivals saw their shares tumble as investors pivoted to the new Huawei-backed leader:
- Zhipu AI: Shares fell 9%
- MiniMax: Shares slid 7%
DeepSeek is currently seeking fresh funding at a valuation exceeding $20 billion, with tech giants Alibaba and Tencent reportedly in discussions to take significant stakes. This investment would solidify DeepSeek’s position as the primary challenger to Western AI dominance.
Cybersecurity Perspective: Is Huawei-Native AI More Secure?
For organizations evaluating the V4 model, the shift to Huawei hardware introduces a new set of security considerations:
- Supply Chain Resilience: For firms operating within the Chinese ecosystem, a Huawei-native model provides immunity against US-based hardware “kill switches” or supply chain interruptions.
- Detection of Distillation: The controversy surrounding “model distillation” highlights a new security frontier. If V4 is indeed built on distilled data from US labs, it may inherit the same adversarial vulnerabilities found in GPT-4 or Claude 3.
- Hardware-Level Audit: Using the Ascend supernode architecture allows for a “Sovereign AI” stack where the hardware, firmware, and model are all produced within a single regulatory jurisdiction, theoretically reducing the risk of foreign backdoors—or increasing them, depending on your perspective.
FAQs
Q: Can DeepSeek V4 run on Nvidia chips? A: While optimized for Huawei Ascend, as an open-source model, V4 can likely be adapted for Nvidia hardware, though it may not reach peak efficiency without Huawei’s specific supernode optimizations.
Q: What is “Model Distillation”? A: It is the process of using a larger, more powerful model (like GPT-4) to generate data to train a smaller model. US labs claim DeepSeek used their models to “short-cut” the training process.
Q: Why did DeepSeek’s rivals’ stocks drop? A: Investors see the Huawei-DeepSeek partnership as the “winning” domestic alliance for China’s AI future, leaving companies that still rely on restricted Nvidia chips in a precarious position.
Conclusion: The End of Nvidia’s Monopoly in China?
DeepSeek V4 is more than just a software update; it is a declaration of technological independence. By proving that a world-class model can be built on Huawei silicon, DeepSeek has provided the blueprint for a post-Nvidia future in the East.
Is your AI strategy geographically hedged? As the world splits into two distinct AI stacks—Western-Nvidia and Eastern-Huawei—the ability to run models across both will be the ultimate competitive advantage.
