Jefferies report shows that the competitive landscape for large AI models is undergoing profound changes. New entrants such as Singapore’s Sapiens AI and Korea’s Motif have entered the global top model rankings with just $20 million in funding, as distillation technology has significantly lowered the barrier to development.

Author: Li Jia

Source: Wall Street Journal

Competition in large AI models is undergoing new changes: the supply of models continues to grow, and technologies like distillation are continuously lowering the barriers to entry, enabling new players to enter the high-performance model market with far lower capital investment than leading firms. Jefferies believes this trend is pressuring API prices and challenging the previous industry logic that relied on massive capital expenditures to build competitive barriers.

According to a research report released by Jefferies on September 16, four new models were added in September to the global top 14 language models tracked by Artificial Analysis (AA), originating from Singapore, South Korea, the United Arab Emirates, and the United States. The organizations behind these models generally have limited funding, with some having raised only about $20 million in total, yet they have already entered the global top-tier model rankings.

Jefferies analyst Edison Lee and others believe that the growing number of participants in the LLM market, increasing regional support for local models, and the widespread adoption of distillation technologies are further lowering industry entry barriers. Meanwhile, narrowing performance gaps between models and intensified API price competition are imposing new pressures on the AI supply chain characterized by high capital expenditures and low return certainty.

New players can enter at a low cost, with distillation technology lowering the barrier to entry.

The report shows that the four new models entering the top 14 of AA in September are Agnes 3.0 Flash by Sapiens AI from Singapore, Motif-3-Beta by Motif Technologies from South Korea, K2 Horizon by MBZUAI from the UAE, and Apodex 1.1, headquartered in California, USA, with its development team in Singapore.

Among them, Sapiens AI has raised approximately $20 million in total funding, and Motif Technologies has completed a $16.6 million Series B round, with overall investment levels still significantly lower than those of leading AI companies.

These cases illustrate that the capital barrier to entry in the large model competition is declining. Some new entrants are not replicating the path of leading companies, which involves大规模 training of foundational models, but are instead rapidly iterating by leveraging open models, smaller teams, and more efficient training methods.

Distillation technology is a key driver. By training new models using outputs from existing large models, developers can acquire some of the advanced capabilities of these models at a lower cost. Jefferies believes that such technologies are difficult to completely block, and as their adoption grows, the cost for newcomers to catch up to leading models may continue to decline.

However, low-cost entry does not mean that leading players have lost their advantages. Training state-of-the-art models still requires substantial computational power, data, and engineering investment; new entrants are gaining competitiveness primarily in specific capabilities and niche use cases. The real change lies in the growing capital gap between entering the market and achieving market leadership.

API prices are under pressure, and models are beginning to enter price competition.

The increase in the number of models has begun to impact API pricing. According to Jefferies data, the API prices of some new models are already significantly lower than those of leading models, with Apodex 1.1’s hybrid API price at $0.3 per million tokens, making it one of the lowest-priced models among the top 15 models in AA.

Meanwhile, leading manufacturers have not broadly joined price cuts but are instead seeking higher pricing through capability upgrades in their next-generation models. The report suggests that this price increase not only reflects improved model capabilities but may also serve as a way to demonstrate commercialization potential to capital markets—leveraging higher intelligence to justify higher prices and thereby improving market expectations around profitability and return on investment.

The issue is that price increases and decreases are happening simultaneously. Leading models are attempting to maintain a premium through performance upgrades, while new entrants are competing for developers and application demand through low costs. Jefferies notes that there are currently seven large model providers in the U.S. market alone, and increased participation suggests that price competition in the API market may continue.

Shift from competing on intelligence to competing on efficiency

Another shift in model competition is the industry’s growing focus on balancing model intelligence with reasoning efficiency. Jefferies introduced the AutomationBench-AA benchmark this month to evaluate models’ performance in agent-based automation tasks, using the results to recalibrate historical data.

Under this context, some model providers have begun reducing inference costs through model architecture and memory optimizations. On September 10, DeepSeek released the V4.1 Flash model, lowering its hybrid API price to $0.2 per million tokens—a 74% reduction from the previous version.

The model further reduces hardware requirements through technologies such as the Casual Encoder-Decoder design, Engram-conditioned memory mechanism, and Compressed Sparse Attention 2, lowering HBM demand by approximately one-quarter and SSD demand by approximately one-eighth. Jefferies believes this approach emphasizes computational and memory efficiency rather than solely pursuing model intelligence.

This also means that the competitive dimensions for AI models are expanding. As demand for inference tokens grows rapidly, model providers must not only enhance capabilities but also reduce the cost per token generated. For application developers, if lower-cost models can already handle a sufficient number of tasks, price differences between models may become more important than performance differences alone.

Capital expenditures are still growing, but returns need to be revalidated.

Jefferies' concerns about the AI supply chain are not that AI demand will disappear, but rather whether the pace of capital investment can keep up with eventual returns. On one hand, new entrants are already able to secure funding in the tens of millions of dollars to enter the global top model rankings; on the other hand, leading companies continue to expand their investments in computing power, data centers, and infrastructure.

From the demand side, AI inference demand continues to grow rapidly, and improved model capabilities are constantly creating new use cases, so the fundamental basis for compute demand remains strong. The issue lies in how long it will take for new infrastructure investments to generate sufficient revenue and profit once model supply becomes more abundant and API prices are pressured by competition.

Jefferies therefore believes that the AI industry may gradually enter a phase that places greater emphasis on capital efficiency. For leading companies, continuing to increase capital expenditures may still be an important way to maintain technological leadership, but the industry as a whole needs to pay closer attention to redundant construction, resource utilization, and the actual commercial value corresponding to new computing power.

Ultimately, Jefferies expects the industry to reduce overall capital expenditures through consolidation, collaboration, and more efficient resource allocation. Demand for AI infrastructure remains strong, but "demand growth" does not automatically equate to "sufficient returns on all capital investments," which will become a key variable in the market's next phase of reassessing AI valuations.