Chinese open-source models are racing ahead, says QGQ Partners' Kersmanc

Key Concepts

• AI Chip Partnerships: Competition and collaboration between major tech companies for AI hardware.
• Large Language Models (LLMs): Advanced AI models capable of understanding and generating human-like text.
• Barriers to Entry: The level of difficulty and resources required to enter a specific market or technology space.
• Open Source Models: AI models whose source code is publicly available, allowing for wider development and adoption.
• Compute Power: The processing capability of hardware, crucial for training and running AI models.
• GPUs (Graphics Processing Units): Specialized processors widely used for AI and machine learning tasks.
• TPUs (Tensor Processing Units): Google's custom-designed processors optimized for machine learning.
• Third-Party Distribution Channels: Intermediaries involved in selling hardware, such as GPUs.
• Oversaturation: A market condition where supply exceeds demand, leading to price reductions.
• AI Ecosystem Flywheel: A self-reinforcing cycle of investment and innovation in the AI space.
• Venture Capital (VC) Driven Demand: Demand for AI infrastructure fueled by investment from venture capital firms.

AI Landscape: Competition and Shifting Fortunes

The current AI landscape is characterized by a dynamic interplay between major tech players, particularly Alphabet and NVIDIA, with recent events suggesting a potential shift in market sentiment. Alphabet's record-breaking rally has paused, while NVIDIA has seen a rebound. A key point of contention is the competition for AI chip partnerships, with Meta reportedly exploring a deal with Google, a move that could impact NVIDIA's customer relationships.

Lowered Barriers to Entry in Large Language Models (LLMs)

Brian Kirschman, Portfolio Manager at KCG Partners, argues that the market has been "overexcited" about the AI opportunity. The recent launch of Gemini 3 by Google, which was previously perceived as lagging, has demonstrated that the barriers to entry for developing Large Language Models (LLMs) are "a lot lower than people really originally anticipated." This suggests that companies like Google, once seen as "buggy whip suppliers," can now take a leading position.

• Evidence: Kirschman points to the emergence of numerous open-source LLMs, including those from China, that are competing effectively with models like Gemini 3 and OpenAI's offerings.
• Technical Detail: These Chinese models are achieving competitive performance with "a lot less compute power," indicating a reduced reliance on high-end GPUs or TPUs. This underestimation of Chinese AI capabilities is a significant factor.

Demand for AI Infrastructure: Near-Term vs. Long-Term Outlook

While the increased competition might suggest a slowdown in demand for AI hardware, Kirschman believes that in the "near term," demand for infrastructure companies like NVIDIA will likely continue. However, he anticipates a shift in the longer term.

• Near-Term Softness: Evidence of "softness" in GPU pricing has been observed within "third-party distribution channels." This suggests that the channel is becoming "a little bit oversaturated."
• Longer-Term Concerns: If this oversaturation continues, the weakness is expected to "trickle back over to the likes of NVIDIA." Kirschman questions the "headroom on where this spending continues on a longer term basis."

The AI Ecosystem Flywheel and Venture Capital Influence

The interconnectedness of the AI ecosystem is a critical factor. Kirschman highlights the "flywheel effect" where innovation and investment in one area drive growth in others. NVIDIA and OpenAI are described as "linchpins" of this ecosystem, driving significant AI investment.

• Impact of Cracks: If there are "cracks" in the OpenAI thesis, or if competition intensifies, it could slow down this flywheel. This slowdown would "limit the funding" available for AI development.
• VC-Driven Demand: A significant portion of the incremental demand for cloud platforms from hyperscalers is attributed to "incremental AI venture capital driven type of demand." A curtailment of VC funding would directly impact this demand, raising questions about the remaining growth potential.

Key Arguments and Perspectives

• Market Overexcitement: The prevailing sentiment that the market has been overly optimistic about the AI opportunity.
• Underestimated Competition: The significant and growing competition in the LLM space, particularly from open-source models and those developed with less compute power.
• Shifting Demand Dynamics: The expectation of continued near-term demand for AI hardware, followed by potential weakness due to market saturation and reduced VC funding in the long term.
• Interdependence of the AI Ecosystem: The critical role of key players like NVIDIA and OpenAI in driving the AI flywheel, and how disruptions can have cascading effects.

Conclusion

The AI market is undergoing a period of re-evaluation. While the immediate future may see continued demand for AI infrastructure, the increasing competition in LLMs and the potential for reduced venture capital investment raise concerns about the long-term sustainability of the current growth trajectory. The market's rationality regarding the AI opportunity is being tested, with a greater emphasis on the underlying competitive dynamics and the true drivers of demand.