The dark side of the moon ignites an AI trading upheaval, reviewing the winners and losers in the new landscape.

The release of the Kimi K3 model by Moonshot AI has triggered the "DeepSeek Effect," sparking a new wave of repricing in global tech stocks and forcing investors to re-examine the core assumptions supporting AI trading logic.

Moonshot AI unveiled its new generation model, Kimi K3, last Friday, claiming its performance rivals the top products from OpenAI and Anthropic but at just a fraction of their cost. The market began comparing this event to last year's "DeepSeek Moment." IG market analyst Tony Sycamore said the shock caused a combined evaporation of $314 billion in expected valuations for the two unlisted U.S. companies, OpenAI and Anthropic.

Commentators believe Kimi K3’s emergence shakes the current market logic from two angles: first, the judgment that U.S. restrictions on advanced chip exports will hinder AI progress in other countries now faces a challenge; second, with more powerful and cheaper models accelerating global AI application adoption, investors are now concerned about potential overcapacity in AI infrastructure.

However, the impact is not entirely negative. Judging from market performance, players in different sectors are experiencing sharply contrasting fortunes. Here is a rundown of potential winners and losers.

Winner 1: China’s Semiconductor Ecosystem

The specific chips used in the development of K3 by Moonshot AI are still unknown, but the release of the new model has directly boosted the Chinese semiconductor sector. In Monday’s trading, SMIC's A-shares surged over 5% at one point, as the market expects intensified competition among AI model developers will drive continuous increases in related infrastructure spending.

Gary Tan, portfolio manager at Allspring Global Investments, said: "We expect the biggest winners will still be at the AI infrastructure level. With strong government support, China’s push for open-source AI may accelerate deployment of advanced domestic AI models, which will require more computing resources and drive demand for underlying hardware, especially network equipment and memory chips."

Last weekend, Alibaba released a preview version of its flagship model Qwen3.8 Max, which features 24 trillion parameters and reportedly rivals the world’s leading frontier models in performance. The consecutive releases of these two products, coupled with Chinese AI developers’ growing preference to optimize models for domestic chips, jointly drove the Star Market Composite Index up as much as 3.6% on Monday.

Winner 2: Memory Chip Manufacturers

Kimi K3, with its 2.8 trillion parameters, is China’s largest AI model to date, and its sparsity ratio—a critical indicator of computational efficiency—has reached a historic high. A higher sparsity ratio means fewer parameters are activated per inference task, thus reducing compute consumption.

However, all 2.8 trillion parameters still need to reside in memory. Even when compressed into a low-precision format, Kimi K3 occupies about 1.4 terabytes of memory space. This means deploying the model requires AI processor clusters with large-capacity memory, such as Nvidia’s Blackwell GB300 system, continuing the demand for memory chip suppliers like SK Hynix and Samsung Electronics.

Stanley Tang, senior portfolio manager at Sumitomo Mitsui DS Asset Management, said: “Due to the limited number of suppliers, memory chip manufacturers remain among the most advantaged businesses in the semiconductor sector.” He further pointed out that if models like Kimi accelerate in adoption, overall demand is unlikely to fall, and may instead drive faster penetration of AI agents, further boosting memory demand.

Winner 3: AI Agents and Software Developers

The release of Kimi K3 has also boosted confidence in Chinese software companies and AI agent developers—even as access to the most advanced global chips is restricted, these companies continue to improve model performance and efficiency.

“What’s certain is that an open-source model with outstanding performance and lower cost will reduce application development costs and may accelerate adoption across programming, customer service, and industrial processes,” said Charu Chanana, chief investment strategist at Saxo Markets.

However, Morningstar analyst Malik Ahmed Khan takes a more cautious view of the competitive risks faced by software developers.

Khan believes attributing last Friday’s sell-off of Alphabet, Amazon, and Microsoft to market concerns that more companies will turn to open-source models inspired by K3 is “untenable.” He said, U.S. companies, government agencies, and those cooperating with the U.S. government will not adopt foreign open-weight models just to lower inference costs.

Loser 1: AI Model Developers

If K3 has created new winners, the competitive landscape for AI model developers has clearly worsened.

Zhihu, previously seen as China’s best open-source model developer prior to K3’s release, saw its Hong Kong stock price plunge nearly 40% over the past two trading days, reflecting market concerns about intense competition in the model layer.

The competition won’t subside soon. Alibaba’s stock rose Monday following the release of Qwen3.8 Max, which the company positions as a product whose capability is second only to Anthropic Fable 5. MiniMax Group Inc. reportedly plans to launch an upgraded version of its M3 model. Moonshot AI itself announced it would publicly release Kimi K3’s model weights on July 27, enabling businesses to run the model independently without relying on its cloud services.

Sumitomo’s Tang said: "We are still at an early stage, and better models will keep emerging. Some models may become worthless, and I believe the market has not yet fully priced in this risk."

Loser 2: High-End Chip Manufacturers

The release also made the market re-examine whether core beneficiaries of the AI boom, such as Nvidia and AMD, can continue to enjoy scarcity premiums. Unlike the DeepSeek shock, K3’s breakthrough is not in training efficiency, but in achieving top-tier inference performance at lower cost—a fundamentally different reliance on high-end computing power.

Mark Malek, chief investment officer at Siebert Financial, said: "On that day, investors in Tokyo and Taipei priced in doubts about whether capital expenditures on AI will ultimately pay off. Whenever a lab anywhere in the world—no matter which country—releases a free, frontier-level model, these doubts will resurface."

From a broader perspective, the consecutive releases of Kimi K3 and Qwen3.8 Max indicate that frontier AI model competition is expanding far beyond OpenAI, Anthropic, and other leading U.S. labs. According to Bloomberg, this trend could pressure the pricing power of U.S. model providers and continue to drive global investment in infrastructure needed to run stronger AI systems. The winner-and-loser landscape on the AI track may be undergoing a profound restructuring.

Risk Warning and DisclaimerThe market has risks; investment requires caution. This article does not constitute personal investment advice, nor does it take into account the individual investment objectives, financial situations, or needs of specific users. Users should consider whether any opinions, viewpoints, or conclusions in this article fit their particular situations. Investing based on this article is at your own risk.