Are Large Language Models Now a Commodity?

Are Large Language Models Now a Commodity?

Pausing expenditures may not be feasible for the industry leaders seeking to remain at the forefront of the development of artificially generated intelligence, Dominic Rizzo, Portfolio Manager at US asset manager T. Rowe Price, writes.

A Chinese artificial intelligence (AI) company, DeepSeek, recently announced it had developed an open-source large language model (LLM) that is relatively inexpensive to train and requires less energy and processing power than the leading applications.

The result was a significant sell-off in US technology stocks – especially semiconductor companies specializing in AI application chips. Are LLMs now a commodity, limiting the need for large-scale training?

Focus on Performance Rather Than on Cost

This remains unclear, but a positive perspective exists for continued large-scale training. DeepSeek has demonstrated that reinforcement learning is effective, and it is reasonable to assume that it could improve with increased computational power and more data. This would suggest that AI laboratories might benefit, somewhat ironically, from increasing their spending.

We believe the primary objective for AI laboratories should be to focus on performance rather than cost differentiation. Pausing expenditures may not be feasible for the industry leaders seeking to remain at the forefront of the development of AGI (artificially generated intelligence).

This could be positive for the uptick in the adoption of AI models

• We expect there will be a variety of models – some large and some small. Large models may be required for optimal consumer applications and could be more costly. However, the availability of inexpensive models is beneficial for widespread AI adoption. Smaller models should benefit from advancements in software and hardware, as well as insights gained from larger models.
• More affordable models like DeepSeek’s might be crucial to boosting real-world demand for AI applications. This suggests that in the medium term, the demand for training and inferencing may actually rise as we see AI adopted everywhere.
• As with all new dynamics, we want to remain humble and nimble and reserve the right to change our minds. Regardless, we will follow our investment framework to make investment decisions, which means we will be looking for 1.) linchpin technologies, 2.) innovation in secular growth markets, 3.) improving fundamentals, and 4.) reasonable valuations.

Dominic Rizzo is Portfolio Manager, Global Technology Equity, at US asset manager T. Rowe Price.