China's Low-Cost AI Emerges: Why It Could Be a Boon for Nvidia, Micron, and Other Chip Stocks

China's launch of low-cost artificial intelligence (AI) models has reignited debate over corporate investment in massive memory and computing costs. However, some analysts argue this may not be detrimental to U.S. semiconductor leaders like Nvidia (NVDA-US) and Micron (MU-US), whose stock prices have surged due to strong AI hardware demand. As AI usage costs decline, chipmakers may actually emerge as key beneficiaries.

Chinese AI startup Moonshot AI unveiled its open-weight model Kimi K3 last week, which has demonstrated performance competitive with top-tier U.S. models in certain benchmark tests. With 2.8 trillion parameters, Kimi K3 is set to become the world's largest open-source model once Moonshot AI releases its model weights by the end of this month.

According to the Arena Frontend Code leaderboard, Kimi K3 surpasses Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol in coding capabilities—both of which are closed models.

The development of a model approaching world-leading capabilities by Chinese AI labs—despite significantly less funding than their U.S. counterparts and under U.S. export restrictions on Nvidia's high-end AI chips—has shocked the market. The release of Kimi K3 also accelerated selling pressure on U.S. chip stocks, with the Philadelphia Semiconductor Index officially entering bear market territory last Friday.

However, the surge in demand for Kimi K3 has revealed that Chinese AI models, like their Western counterparts, remain highly dependent on computing power (compute). Over the weekend, Moonshot AI announced on social media platform X that demand for Kimi K3 has far exceeded expectations, pushing computing resources to their limits. As a result, the company has temporarily paused new user registrations to prioritize compute resources for existing members.

Anni Sen, Managing Partner at BluBird Capital, stated this is actually "a major positive for the memory industry."

According to data from open-source platform Hugging Face, while Kimi K3 has 2.8 trillion parameters, only about 50 billion are activated simultaneously. Parameters are variables used by AI models to learn and recognize patterns during training, essentially serving as the model's "memory."

Sen noted that while this design improves computational efficiency, enterprises deploying Kimi K3 on-premise must still store all 2.8 trillion parameters in memory, meaning demand for high-performance memory remains substantial.

Currently, Kimi K3 charges $15 per million tokens of output, lower than OpenAI's GPT-5.6 Sol at $30 and Anthropic's Claude Fable 5 at $50.

Sen believes this aligns with the economic principle of the "Jevons Paradox": increased efficiency and lower costs may actually drive overall demand higher.

She stated that Kimi will stimulate demand for more AI applications, prompting developers to leverage its cost advantage to launch new AI services, thereby increasing AI inference demand. In the future, companies may shift large volumes of mid-tier AI tasks to lower-cost models, reserving only complex, multi-step AI agent tasks for the most advanced models.

Nathan Lambert, an AI researcher and founder of the Interconnects AI blog, noted that open-weight models are accelerating AI's penetration across the broader economy, as the cost of achieving a certain level of AI capability has significantly decreased.

He also pointed out that open models offer greater customization, making them more practically valuable for enterprises.

Wedbush analyst Matt Bryson stated that as AI model sizes continue to grow, memory demand to support more parameters will also rise. Two scenarios may emerge: either more memory is integrated per AI chip, or AI computing clusters expand in scale to accommodate more model weights.

Therefore, Bryson said, if Chinese AI models continue to gain traction, "it's actually a positive for memory suppliers." If cluster sizes keep expanding, it will also boost demand for high-speed networking equipment suppliers.

Globally, high-bandwidth memory (HBM) is primarily supplied by Micron, SK Hynix, and Samsung Electronics. With HBM supply remaining tight, these three companies have maintained strong pricing power.

Gil Luria, Executive Director at D.A. Davidson, stated that as AI model operating costs decline, overall AI demand may continue to grow, and chip demand will rise accordingly—especially for memory chips, whose supply remains extremely tight.

Joseph DeYonker, CEO of PurePlay ETFs, also believes that the cheaper and more accessible AI models become, the faster they will test infrastructure limits. Kimi K3 is a prime example, proving that the semiconductor industry's long-term growth trend remains solid.

DeYonker said that running advanced frontier AI models and ultra-long context windows will place immense pressure on memory and advanced packaging supply chains, meaning HBM suppliers and advanced packaging firms are likely to maintain strong pricing power.

On the other hand, he expects GPU and custom AI chip design companies like Nvidia and Broadcom (AVGO-US) to maintain strong order volumes, as enterprises and large cloud providers (hyperscalers) continue to race to acquire AI hardware.

DeYonker believes Kimi K3's temporary suspension of new user registrations due to compute shortages "is not a warning sign of weakening semiconductor cycles, but proof of persistently strong structural demand in the semiconductor industry."

He stated that the unresolved bottlenecks in memory and compute indicate that the semiconductor industry's long-term profit growth momentum remains intact.

Futurum's Chief Market Strategist Shay Boloor pointed out that Kimi K3 has already completed model training, and the real challenge now is not training but how to deliver services at scale. This means AI inference demand has become the primary bottleneck constraining AI industry development.