What is DeepSeek V4 on Huawei chips? DeepSeek V4 on Huawei chips refers to China's most advanced AI lab, DeepSeek, building its next-generation trillion-parameter model to run exclusively on domestically produced Huawei Ascend 950PR processors rather than Nvidia GPUs. Reported by The Information on April 3, 2026, this decision, combined with bulk chip orders from Alibaba, ByteDance, and Tencent, signals the emergence of a fully bifurcated global AI hardware supply chain with direct consequences for CRE data center investors worldwide. For a comprehensive comparison of how different AI models serve CRE professionals, see our AI model comparison guide for CRE investors.
Key Takeaways
- DeepSeek V4 will use a nearly 1 trillion parameter architecture running entirely on Huawei Ascend 950PR chips, with 1.8x faster inference speeds and a 1 million token context window.
- Alibaba, ByteDance, and Tencent have placed bulk orders for hundreds of thousands of Huawei AI chips, signaling domestic supply chain maturation in China.
- DeepSeek gave Huawei and Cambricon early model access while excluding U.S. chipmakers, breaking from standard industry practice and confirming strategic decoupling.
- CRE data center investors face a bifurcated infrastructure market where Chinese and Western AI ecosystems require entirely different hardware, power, and cooling specifications.
- U.S. data center markets benefit as Western enterprises seeking AI capabilities cannot rely on Chinese models running on sanctioned hardware, reinforcing domestic infrastructure demand.
The DeepSeek V4 Hardware Shift Explained
According to Reuters, DeepSeek has spent months collaborating directly with Huawei and Chinese chip designer Cambricon Technologies to rewrite portions of the model's underlying code specifically for Huawei's Ascend architecture. The company granted early access to these domestic chip suppliers while deliberately excluding U.S. chipmakers from pre-launch testing, a break from standard practice that would have been unthinkable even a year ago.
The V4 model architecture is expected to use a nearly 1 trillion parameter design, requiring hundreds of thousands of Ascend 950PR chips. The 950PR is Huawei's answer to Nvidia's Blackwell architecture and was designed specifically for AI training and inference workloads. When DeepSeek released its V3 and R1 models in late 2025, the market impact was dramatic: the demonstration that competitive AI models could be built at a fraction of the cost of Western alternatives triggered a global tech stock selloff and forced investors to question whether U.S. hyperscalers needed to spend hundreds of billions on Nvidia hardware.
DeepSeek V4 takes this further by proving that competitive AI can be developed entirely outside the Western semiconductor ecosystem. The company is also building two additional V4 variants optimized for different capabilities, all designed to run on Chinese-made chips. For background on the Huawei chip that makes this possible, see our earlier analysis of Huawei's CUDA-compatible 950PR AI chip.
Bulk Orders Signal Domestic Supply Chain Maturation
Perhaps more significant than DeepSeek's own hardware choice is the cascade of bulk orders from China's largest technology companies. Five sources with direct knowledge confirmed to The Information that Alibaba Group, ByteDance, and Tencent Holdings have placed orders for hundreds of thousands of Huawei AI chip units. This is not a pilot program or a proof of concept. It is production-scale procurement for deployment across China's most demanding AI workloads.
The collective purchasing power of these companies represents a validation event for Huawei's AI chip ecosystem. When a startup chooses a domestic chip, it signals ambition. When Alibaba, ByteDance, and Tencent collectively commit hundreds of thousands of units, it signals that the Chinese AI chip supply chain has reached the minimum viable scale required for enterprise deployment. This has three immediate CRE implications:
- Chinese data center construction will accelerate on domestic specifications. Data centers optimized for Huawei Ascend chips will use different power, cooling, and rack configurations than facilities designed for Nvidia GPUs. CRE developers building for the Chinese market must now design to domestic hardware specifications rather than adapting Western blueprints.
- Nvidia-dependent Western data centers gain relative scarcity value. As Chinese demand partially shifts away from Nvidia hardware, the premium on Western data centers with guaranteed access to Nvidia Blackwell and Vera Rubin chips increases. Tenants seeking these chips will pay higher lease rates for facilities that can guarantee power, cooling, and connectivity for Nvidia-based deployments.
- The global data center supply chain permanently bifurcates. Equipment manufacturers, cooling system providers, and power infrastructure companies will increasingly need to support two distinct technology stacks. CRE investors must evaluate which stack their assets are designed to serve and whether switching costs create tenant lock-in advantages.
CRE Data Center Investment in a Bifurcated World
The DeepSeek V4 decision accelerates a trend that CRE data center investors must now actively price into their underwriting. The global AI infrastructure market is splitting into two parallel ecosystems: a Western stack built primarily on Nvidia, AMD, and Intel silicon, and a Chinese stack built on Huawei, Cambricon, and other domestic processors. Each ecosystem will generate its own data center demand, its own construction specifications, and its own investment dynamics. CRE investors looking for hands-on AI implementation support can reach out to Avi Hacker, J.D. at The AI Consulting Network for guidance on navigating this bifurcation.
For U.S.-focused CRE investors, the bifurcation is largely positive. Western enterprises and government agencies cannot use AI models running on sanctioned Chinese hardware. This creates a captive demand pool for U.S.-based data centers that will only grow as AI adoption increases. According to JLL, 92% of corporate occupiers have initiated AI programs, but only 5% report achieving most of their AI goals. As these programs scale, demand for compliant, Western-stack data center capacity will intensify.
The risk dimension is more nuanced. If Chinese AI models running on domestic chips achieve performance parity with Western alternatives at significantly lower cost, it could pressure enterprise AI pricing globally. Lower AI service costs could reduce the revenue per megawatt that data center tenants generate, potentially affecting lease economics. CRE investors should monitor the price-performance gap between DeepSeek V4 (on Huawei) and comparable Western models (GPT-5.4, Claude Mythos 5, Gemini 3.1 Ultra) as a leading indicator of potential margin compression. For context on how open-source models from China and elsewhere are already changing the competitive landscape, see our guide on open-source AI models for CRE.
Geopolitical Supply Chain Risks for CRE Portfolios
The DeepSeek V4 decision must be understood within the broader context of U.S.-China technology competition. U.S. export controls have restricted Nvidia's ability to sell advanced AI chips to China, pushing Chinese companies toward domestic alternatives. China has responded with massive investment in domestic semiconductor capability, with Huawei's Ascend line now the centerpiece of a national strategy to achieve AI hardware independence.
For CRE investors, the geopolitical risks cut both ways:
- Escalation risk: Further U.S. restrictions on technology exports to China could accelerate the bifurcation, creating short-term disruption but long-term demand growth for Western data centers. Conversely, any relaxation of export controls could reintroduce Chinese demand for Nvidia chips, affecting the supply-demand balance in Western chip markets.
- Equipment supply chain: As our analysis of US data center builds delayed by Chinese equipment documented, half of U.S. data center construction projects face delays due to shortages of Chinese-manufactured transformers, switchgear, and batteries. The hardware bifurcation extends beyond chips into the full infrastructure stack, creating supply chain chokepoints that CRE developers must navigate.
- Tenant screening: CRE data center operators must increasingly evaluate tenant compliance with export control regulations. Hosting AI workloads for entities with Chinese government connections on U.S. soil carries regulatory risk. Thorough tenant due diligence becomes a competitive differentiator for data center owners serving security-sensitive customers.
What CRE Investors Should Do Now
The emergence of a fully independent Chinese AI hardware ecosystem changes the calculus for CRE data center investment in five specific ways:
- Underwrite for hardware-specific tenant demand: Evaluate whether your data center assets are designed for Nvidia-based deployments (the dominant Western configuration) and whether the power and cooling specifications support next-generation chips like Vera Rubin NVL72 at 190 to 230 kW per rack.
- Monitor the DeepSeek V4 launch closely: When V4 releases in the coming weeks, benchmark its performance against GPT-5.4 and Claude Mythos 5. If performance is comparable at lower cost, expect renewed pressure on AI pricing that could affect tenant unit economics in Western data centers.
- Evaluate export control compliance programs: Ensure your data center operations and tenant screening processes comply with current BIS (Bureau of Industry and Security) export control regulations. Non-compliant facilities face enforcement risk that could impair property operations and valuations.
- Assess supply chain dependencies: Identify which components in your data center construction and operations pipeline originate from China and develop alternative sourcing strategies. The bifurcation of the technology ecosystem will eventually extend to the full infrastructure supply chain.
- Position for sustained Western AI demand: The most certain CRE outcome of the bifurcation is that Western enterprises and government agencies will need more domestic data center capacity, not less. Markets with available power, permitting efficiency, and connectivity, particularly Northern Virginia, Dallas, Phoenix, and Columbus, remain the highest-conviction data center investment targets.
For personalized guidance on evaluating how the U.S.-China AI hardware split affects your specific data center portfolio, connect with The AI Consulting Network. The bifurcation of global AI infrastructure is creating both risks and opportunities that require informed, data-driven investment decisions. CRE sales volume is forecast to increase 15 to 20% in 2026 (Source: Cushman and Wakefield), and data centers positioned on the right side of the hardware divide will capture a disproportionate share of that growth.
Frequently Asked Questions
Q: Will DeepSeek V4 compete with GPT-5.4 and Claude for CRE applications?
A: DeepSeek V4 will likely be competitive on reasoning and analysis benchmarks based on its predecessor's performance. However, for U.S.-based CRE firms, regulatory compliance and data residency requirements make it impractical to run production workloads on a Chinese model hosted on sanctioned hardware. Western CRE professionals will continue to rely on GPT-5.4, Claude, and Gemini for sensitive underwriting, due diligence, and tenant analysis.
Q: How does the Huawei chip decision affect U.S. data center construction?
A: It reinforces domestic demand. As Chinese AI companies build their own infrastructure on domestic chips, U.S. enterprises and government agencies must invest more heavily in Western-stack data centers. Goldman Sachs projects hyperscaler capex will exceed $650 billion over the next 12 months, and the bifurcation with China adds urgency to domestic buildout plans.
Q: Should CRE investors avoid exposure to Chinese data center markets?
A: Not categorically, but the risk profile has changed. Chinese data centers will increasingly use domestic hardware with different specifications, regulatory frameworks, and tenant dynamics than Western facilities. U.S.-based investors face additional complications including sanctions compliance, repatriation risk, and limited legal protections. Most institutional CRE investors are better served focusing on domestic data center markets where regulatory clarity and demand fundamentals are strongest.
Q: What is the timeline for full Chinese AI chip independence?
A: The DeepSeek V4 launch, expected within weeks, represents a significant milestone. However, full independence requires not just competitive chips but also a mature ecosystem of software tools, development frameworks, and trained engineering talent. Analysts estimate China is 2 to 3 years behind Western chip performance at the frontier, but the gap is narrowing rapidly. For CRE investors, the relevant timeline is not full parity but rather "good enough" performance for most enterprise workloads, which is arguably being achieved now.