Why DeepSeek-V4 Didn’t Trigger Another Global AI Stock Shock

DeepSeek-V4’s April 24, 2026 preview did not recreate the shock associated with DeepSeek-V3 and R1 because the market had changed. Low-cost, capable Chinese AI was no longer an unexpected challenge to U.S. technology companies; it had become an established part of the competitive landscape. Reuters described the initial market response as subdued, with V4 improving on earlier DeepSeek models but facing strong competition from Kimi and Qwen. 6

The most important reveal was therefore not a sudden collapse in confidence in AI infrastructure. It was a change in what investors considered surprising—and in how they assessed the strategic importance of an AI model optimized for Huawei hardware.

The first DeepSeek shock was about economics, not just benchmarks

The earlier V3 and R1 releases drew attention because they challenged assumptions behind high U.S. AI-related valuations. Investors had been asking whether frontier-level performance necessarily required enormous capital spending, scarce advanced chips and sustained pricing power for model developers and hardware suppliers. Reporting on the earlier episode linked DeepSeek’s low-cost approach to pressure on global technology stocks and renewed questions about the economics of AI infrastructure.

By April 2026, however, the basic proposition—that Chinese researchers could build highly capable models under computing and chip constraints—was already familiar. One analyst quoted by CNN described R1 as surprising because of its unexpected performance, while characterizing V4 as a continuation of an existing trend that markets had already priced in. 8

That distinction matters. The V4 preview could reinforce concern about margins, hardware demand or competitive durability without creating the same immediate valuation shock. Investors could believe both that Chinese AI was becoming more formidable and that one new preview did not, by itself, invalidate the global AI investment case.

V4-Pro was strong, but not an uncontested open-weight leader

DeepSeek released V4 in two versions: the more capable and expensive Pro model and the lighter Flash variant. Both were presented as open models, with V4 adapted for Huawei chip technology. 5 9

Reuters reported that V4-Pro showed improvement over previous DeepSeek models but faced strong competition from China’s Kimi and Alibaba’s Qwen. The model was therefore viewed as one of several leading open-weight systems, rather than as a release that had clearly left its rivals behind. 6

DeepSeek said V4-Pro outperformed other open-source models on world-knowledge benchmarks, trailing only Google’s closed-source Gemini-Pro-3.1 in the comparison cited by Reuters. That claim indicates meaningful capability, but it is not the same as independent evidence that V4 had decisively surpassed every open-weight competitor. 7

The later commercial positioning reinforced this more nuanced picture. In August, DeepSeek formally released V4 Pro at prices substantially above V4 Flash, presenting stronger benchmark performance as a premium offering. Reuters reported that the released V4-Pro-0813 version was priced at nine times V4 Flash’s input rate and 14 times its output rate, according to Artificial Analysis. 17

That pricing strategy suggests a market with differentiated products and performance tiers—not a single model instantly commoditizing the entire AI sector.

China’s AI market had become a multi-player race

The muted reaction also reflected competitive normalization inside China. DeepSeek was no longer operating against a backdrop in which a single low-cost model could define the country’s AI capabilities for global investors. Kimi, Qwen and other domestic systems were advancing rapidly, making each release easier to compare and harder to treat as an isolated event. 6

This creates two opposing forces for markets:

• More competition can pressure prices and margins. Developers have stronger incentives to cut API costs and improve performance quickly.
• More competition reduces the shock from any single launch. If investors expect another capable model to appear soon, a preview is less likely to reset valuations across the global technology sector.

DeepSeek’s early V4 pricing moves illustrated that competitive pressure. The company announced a 75% temporary discount for V4-Pro and reduced prices for input-cache hits across its API lineup to one-tenth of the previous level. 3

The result is a more demanding market in which capability, inference cost, availability and ecosystem support matter together. A benchmark improvement alone is less likely to produce a broad sell-off unless it also demonstrates a much larger and more durable change in the economics of AI deployment.

Huawei optimization was the bigger strategic signal

V4’s most consequential feature may have been its adaptation to Huawei’s chips rather than its position on a model leaderboard. DeepSeek said the new model was adapted to run on Huawei hardware, and Huawei said its Ascend 950-based supernode infrastructure fully supported V4. 5

That collaboration points to a broader shift in China’s AI strategy: building a domestic stack in which chips, models, software and deployment reinforce one another. Reuters reported that demand for Huawei’s Ascend 950 chips surged after the V4 launch, with major Chinese internet companies seeking orders. 2

The significance is cumulative. A model optimized for domestic accelerators can provide:

• a high-profile workload for Chinese chips;
• software and deployment experience outside Nvidia’s ecosystem; and
• a commercial reason for Chinese cloud and internet companies to adopt local infrastructure.

But the evidence does not establish that Huawei hardware had matched Nvidia’s full performance, supply scale, software maturity or global ecosystem. Model compatibility is strategically important, but it is not proof of complete technological self-sufficiency.

The U.S.–China rivalry is becoming a systems contest

V4 suggests that chip restrictions may have two effects at once. They can limit China’s access to leading foreign hardware, while also increasing the incentive to optimize models and software around domestic alternatives. Reuters described the Huawei-focused release as part of China’s effort to reduce reliance on Nvidia and build a more self-sufficient AI ecosystem. 4

That makes the long-term rivalry broader than a contest over which laboratory produces the best model. It involves:

1. advanced and domestic chip supply;
2. model efficiency and open-weight distribution;
3. compilers, software and developer tools;
4. cloud capacity and inference economics; and
5. the size and willingness of each market to adopt its own technology stack.

The plausible risk for U.S. companies is gradual erosion of their position in China and other price-sensitive markets if domestic chips and open-weight models become capable and deployable enough. The available evidence supports momentum in that direction, but not a definitive conclusion that China has achieved full AI or semiconductor independence. 2 4

Why South Korea and Taiwan did not trade as if AI had been disproved

The absence of a V4-driven global sell-off also fit the market’s continued focus on AI hardware demand. Reuters reported that South Korean and Taiwanese technology stocks remained central beneficiaries of the AI rally, with Goldman Sachs upgrading Taiwan to overweight and raising its target for South Korean shares.

A later Reuters report said South Korea’s KOSPI had doubled in little more than six months and that chip revenues at Samsung had surged, showing how strongly investors continued to associate AI growth with hardware demand.

That optimism was not permanent or risk-free. Foreign investors sold Asian equities at the fastest pace in at least 16 years during the first half of 2026 while trimming crowded AI winners in South Korea and Taiwan. Reuters later reported that the KOSPI and Taiwan’s market suffered sharp pullbacks after their earlier gains.

The lesson is not that Asian AI markets were immune to a DeepSeek effect. It is that investors were differentiating among types of exposure. A cheaper Chinese model could threaten the economics of some U.S. software and hardware assumptions while simultaneously supporting demand for memory, foundry capacity and domestic Chinese accelerators.

What the muted reaction really means

DeepSeek-V4 did not fail to matter; it failed to surprise markets in the same way. The V3 and R1 episode forced investors to reconsider whether expensive AI infrastructure was the only route to competitive performance. By April 2026, that question had already entered market expectations. 8

V4’s more durable message was that China’s AI progress was becoming normalized, its model market was becoming more crowded, and its developers were linking model capability to domestic hardware. That is a slower-moving but potentially broader challenge than a single overnight stock shock.

For investors, the key test is no longer whether DeepSeek can produce one impressive preview. It is whether China can turn capable open-weight models, Huawei-based deployment and sustained price competition into a scalable domestic AI ecosystem. The April reaction suggests markets were prepared to watch that process unfold rather than reprice the entire global technology sector after one release. 2 4