DeepSeek V4 Launches on Huawei Chips, Shaking AI Race

DeepSeek Launches V4 AI Models, Running on Chinese Chips and Priced Far Below US Rivals

Chinese AI startup DeepSeek released preview versions of its long-awaited V4 large language model on Friday, April 24, 2026, in what analysts and industry observers are calling the company's most significant launch since its R1 model upended global markets over a year ago. The release introduces two distinct models — DeepSeek-V4-Pro and DeepSeek-V4-Flash — and marks a pivotal shift: for the first time, DeepSeek has built a flagship model optimized to run on domestic Chinese hardware, specifically Huawei's Ascend chips, rather than Nvidia's industry-dominant GPUs.

The launch immediately moved markets. Shares of Chinese contract chip manufacturers surged in Hong Kong trading, with SMIC rising 10% and Hua Hong Semiconductor climbing 15%. Competing Chinese AI firms were not as fortunate — MiniMax and Zhipu (Knowledge Atlas Technology) each fell around 8%, while Manycore Tech plunged 9%, reflecting investor concerns about intensifying domestic competition from DeepSeek.

What's Inside DeepSeek V4: Scale, Efficiency, and a Million-Token Context Window

According to DeepSeek's official technical documentation published on Hugging Face, DeepSeek-V4-Pro is a massive model with 1.6 trillion total parameters — though only 49 billion are activated at any one time. DeepSeek-V4-Flash, the lighter variant, has 284 billion total parameters with 13 billion activated. Both models were pre-trained on more than 32 trillion diverse, high-quality tokens.

Perhaps the most technically notable feature of both V4 models is their context window: one million tokens, up from 128,000 tokens in prior DeepSeek models. That represents nearly a tenfold increase in the amount of information the model can process in a single pass — a capability with significant implications for enterprise applications involving long documents, codebases, or extended conversations.

Efficiency gains are also substantial. According to DeepSeek's own technical report, V4-Pro requires only 27% of the single-token inference FLOPs and just 10% of the KV cache compared with DeepSeek-V3.2 in a one-million-token context setting — meaning the model can handle far longer inputs without the proportional cost and latency penalties that have historically made long-context processing impractical at scale.

DeepSeek's tech report also states that V4 "falls marginally short of GPT-5.4 and Gemini 3.1 Pro, suggesting a developmental trajectory that trails state-of-the-art frontier models by approximately three to six months" — a candid self-assessment that nevertheless positions DeepSeek as firmly competitive with the leading US models, particularly given its dramatically lower pricing.

Pricing That Undercuts the Competition — By a Wide Margin

Pricing is where the V4 launch could have its most immediate commercial impact. DeepSeek-V4-Pro is priced at $3.48 per one million output tokens, compared to $30 for OpenAI and $25 for Anthropic for equivalent output volume. DeepSeek-V4-Flash goes further still, priced at just $0.28 per million tokens — making it one of the most affordable large language model APIs available anywhere in the world.

For developers, enterprises, and researchers making decisions about which AI infrastructure to build on, this pricing differential is difficult to ignore. The gap between DeepSeek's V4-Pro and OpenAI's equivalent is nearly ninefold. Combined with the open-source nature of the models and the expanded context window, DeepSeek appears to be positioning V4 for rapid adoption — particularly in markets where cost and hardware accessibility are primary concerns.

The Huawei Partnership: A Milestone for Chinese AI Hardware

The most geopolitically significant aspect of the V4 launch is its hardware story. According to MIT Technology Review, V4 is DeepSeek's first model optimized for domestic Chinese chips — specifically Huawei's Ascend series. Huawei confirmed it partnered with DeepSeek on the release, stating that its "Supernode" technology, which combines large clusters of its Ascend 950 chips, would support DeepSeek's V4 model. AI chipmaker Cambricon Technologies also announced compatibility with the new V4 models shortly after launch.

In a move that MIT Technology Review described as a reversal of common industry practice, DeepSeek did not give American chipmakers like Nvidia or AMD early access to V4 ahead of its release. Instead, early access was reportedly granted exclusively to Chinese chipmakers — a deliberate signal about where DeepSeek's hardware allegiances now lie.

The significance of this shift is hard to overstate. Previous DeepSeek models, including the R1 that shocked global markets in early 2025, were built on Nvidia hardware. V4's optimization for Huawei's Ascend chips demonstrates that China's domestic AI chip ecosystem has matured to the point where it can support frontier-level model development and deployment — a development with long-term implications for US export controls and the global semiconductor landscape.

The Broader Context: US-China AI Tensions and DeepSeek's Trajectory

DeepSeek was founded in 2023 and is owned by High-Flyer Capital Management, a Chinese hedge fund. The company first gained international attention in late 2024 with its open-source V3 model, which it claimed was built using less powerful chips at a fraction of the cost of comparable US models. The real inflection point came with its R1 reasoning model, which reportedly cost less than $6 million to build over approximately two months using lower-capacity Nvidia chips — a disclosure that triggered a significant market sell-off and prompted widespread reassessment of assumptions about the cost of frontier AI development.

Multiple governments moved quickly to restrict DeepSeek's earlier models. Multiple US states, Australia, Taiwan, South Korea, Denmark, and Italy introduced bans or restrictions on DeepSeek-R1, citing privacy and national security concerns. That geopolitical backdrop has only intensified ahead of the V4 launch.

On Thursday, April 23 — the day before the V4 release — Michael Kratsios, White House Director of the Office of Science and Technology Policy, posted on X: "The US has evidence that foreign entities, primarily in China, are running industrial-scale distillation campaigns to steal American AI." Both Anthropic and OpenAI have separately raised concerns that DeepSeek improperly built on their models, though those allegations remain contested.

DeepSeek is also reportedly in discussions to raise funds at a valuation exceeding $20 billion, with Alibaba and Tencent said to be in talks to take stakes in the company. The Stanford AI Index 2026, cited by Al Jazeera, concluded that Chinese companies have "effectively closed" the AI performance gap with their US rivals — a framing that gives added weight to the V4 release as both a technical and strategic milestone.

Expert Reactions: What Industry Analysts Are Saying

Analysts tracking both the semiconductor and AI sectors reacted quickly to the V4 news, with many focusing on the Huawei hardware angle as the launch's most consequential element.

He Hui, director of semiconductor research at consultancy Omdia, called the development significant for China's broader AI ambitions: "This is a big deal for China's AI industry. Huawei's Ascend chips are the country's best homegrown alternative to Nvidia, and supporting DeepSeek V4 shows that top Chinese AI models can now run on Chinese hardware."

Wei Sun, principal analyst at Counterpoint Research, argued the implications could ultimately exceed those of the R1 launch: "It allows AI systems to be built and deployed without relying solely on Nvidia, which is why V4 could ultimately have an even bigger impact than R1 — accelerating adoption domestically and contributing to faster global AI development overall."

Ivan Su, senior equity analyst at Morningstar, highlighted the intensifying domestic competitive dynamics the V4 launch reflects: "This is a framing that didn't exist with R1, and that alone tells you how much domestic competition has intensified."

Zhang Yi, founder of tech research firm iiMedia, pointed to the expanded context window as the headline technical achievement: "This addresses the long-standing issues of slower performance and higher costs associated with long context lengths, marking a genuine inflection point for the industry."

Max Liu, a veteran AI industry analyst, offered a broadly optimistic view of the launch's impact on the Chinese AI ecosystem: "It's a good thing for the entire domestic AI industry. It can provide better models for domestic users and we can now expect a lot more things — more products (and a) more competitive market."

Not everyone is sanguine. Nvidia CEO Jensen Huang, speaking on the Dwarkesh Podcast, had previously flagged the scenario now unfolding as a worst-case outcome: "The day that DeepSeek comes out on Huawei first, that is a horrible outcome for our nation." That day has now arrived.

What Comes Next: Adoption, Regulation, and Hardware Rivalry

The V4 release is currently in preview, with users able to test capabilities ahead of a broader rollout. Given DeepSeek's track record with V3 and R1, the open-source release of V4 is widely expected to follow, which would make its models freely available for download and deployment on compatible hardware — including Huawei's Ascend chips — by developers and enterprises worldwide.

The pricing structure of V4's API, particularly the $0.28 per million token cost of V4-Flash, suggests DeepSeek is targeting high-volume, cost-sensitive use cases — areas where US model providers have struggled to compete on economics alone. For enterprise buyers in markets that have access to Huawei infrastructure and are less constrained by US-aligned regulatory frameworks, V4 represents a compelling alternative to OpenAI and Anthropic offerings.

Regulatory responses are harder to predict. Given the bans and restrictions imposed on R1 across multiple jurisdictions, governments in the US and allied nations will likely scrutinize V4 — particularly given its explicit ties to Huawei, a company already subject to extensive US export controls and sanctions. The White House's pre-launch accusations of "industrial-scale distillation campaigns" suggest the US government is closely monitoring developments, though no specific V4-related regulatory actions have been announced as of April 25, 2026.

For the global AI ecosystem, V4's most lasting significance may be its demonstration that frontier AI development no longer requires Nvidia hardware. Whether that translates into lasting competitive parity — or whether the "three to six month" gap DeepSeek acknowledges relative to GPT-5.4 and Gemini 3.1 Pro widens or narrows — will be one of the defining storylines of the AI industry through the remainder of 2026.

For more tech news, visit our news section.

Why This Matters for Your Productivity and Digital Life

The rapid evolution of AI models — from cost reductions to expanded context windows to new hardware ecosystems — is reshaping what tools are available to individuals and teams trying to work smarter. As capable AI becomes cheaper and more accessible, the question shifts from whether to use it to how to integrate it effectively into daily routines, health management, and professional workflows. At Moccet, we track these developments so you don't have to. Join the Moccet waitlist to stay ahead of the curve.