Where does the AI ​​semiconductor ecosystem go from here? Via Investing.com

Investing.com — The trajectory of the AI ​​semiconductor ecosystem is marked by an evolving landscape driven by growing demand for the computing power needed to power advances in artificial intelligence.

According to analysts at Barclays, the sector is at a critical juncture as the global appetite for AI-based solutions, particularly for large language models, continues to outstrip the supply and performance of existing chips.

A sell-off in AI chip names such as NVIDIA (NASDAQ: ) following earnings reports fueled concerns that the market had peaked.

However, Barclays argues that the future of the industry is still full of growth, driven by the ever-increasing computational needs of AI models.

Barclays notes that the AI ​​semiconductor ecosystem is in the early stages of growth, and this period is characterized by significant supply constraints.

Projections indicate that the computational resources required to train the next generation of LLMs, some as large as 50 trillion parameters, are monumental.

The brokerage’s estimates suggest that by 2027, nearly 20 million tokens will be needed to train these models alone. This figure underscores the stark reality that the demand for AI computing is growing at a much faster rate than current chip technology can keep up with, even as the performance of AI accelerators improves.

The gap between AI computing demand and chip supply becomes even more apparent when looking at the training requirements for models like GPT-5, which is expected to require a 46-fold increase in computing power compared to GPT-4.

However, during the same period, the performance improvement of high-end chips such as NVIDIA’s next-generation Blackwell is expected to be only seven times.

This problem is compounded by limited chip manufacturing capacity, with Taiwan Semiconductor Manufacturing Company (NYSE: ) , for example, constrained to produce about 11.5 million Blackwell chips by 2025.

Adding to the complexity is the forecasted demand for inference chips. Inference, the stage where AI models generate results after they have been trained, is set to consume a large portion of the AI ​​computing ecosystem.

Barclays notes that the deduction could account for as much as 40% of the AI ​​chip market, as evidenced by NVIDIA’s claims that a large portion of its chips are used for this purpose. Total chip demand in both training and inference could exceed 30 million units by 2027.

As the industry grapples with these challenges, Barclays suggests a two-pronged approach to the AI ​​accelerator market, where both commercial and custom silicon solutions can thrive.

On the one hand, companies like NVIDIA and AMD (NASDAQ: ) are well positioned to provide chips for large-scale frontier AI model training and inference. On the other hand, hyperscalers—companies that operate massive data centers—will likely continue to develop custom silicon for more specialized AI workloads.

This bifurcated approach will allow flexibility in the market and support various use cases outside of the large LLM domain.

Inference is expected to play an increasingly critical role, not only as a demand driver but also as a potential revenue generator.

New inference optimization methods, such as reinforcement learning applied in OpenAI’s latest ‘o1’ model, signal the potential for advances in AI performance.

With better resource allocation and cost-effective inference strategies, the ROI for AI models could improve significantly, providing incentives for continued investment in both training and inference infrastructure .