AI market likely to split between premium frontier models and lower-cost alternatives: Bernstein
New Delhi, June 18
The global artificial intelligence market is likely to evolve into a two-tier structure, with premium frontier AI models serving highly specialised use cases while lower-cost models dominate mainstream applications, according to a report by Bernstein.
The brokerage said recent developments in the AI industry, including concerns over the rising cost of advanced AI models, indicate that users are becoming increasingly focused on the return on investment (ROI) from AI deployments.
Bernstein noted that the high cost of advanced models has prompted developers and enterprises to rethink how they use AI. It said AI users are likely to choose different models depending on the value of the task being performed, rather than relying on a single platform for all applications.
The report argued that AI commoditisation will not necessarily be driven by convergence in model intelligence, but by whether users perceive competing models to be sufficiently capable for specific tasks.
According to Bernstein, "consumer-focused AI (ordering bubble tea, booking hotel rooms) will commoditise relatively quickly," while more complex enterprise functions such as coding, financial modelling and advanced scientific research will continue to require frontier AI systems.
The report added that once AI becomes reliable enough for a specific use case, competition will increasingly be determined by factors such as cost, reliability and availability rather than superior reasoning capability.
"Once the AI supporting a given use case becomes 'good enough', we'd expect the arbiter of competition and market share to shift from reasoning capabilities to availability and price," the report said.
Bernstein expects leading US AI labs to remain at the technological frontier and continue commanding premium pricing for advanced and specialised applications. However, it believes Chinese AI companies could gain significant market share in mainstream AI applications by offering lower-cost alternatives.
The report said Chinese AI firms may benefit from lower operating costs and their ability to follow global state-of-the-art developments rather than bearing the full expense of exploratory research.
According to Bernstein, "Chinese open source models" are likely to emerge as the value option in the global AI market, particularly for users seeking affordable AI capabilities.
Despite geopolitical and data-security restrictions limiting adoption in some Western markets, Bernstein estimates that Chinese AI companies could still access between 35 and 40 per cent of the global AI total addressable market (TAM).
The report further suggested that AI companies may eventually see improving profitability as more use cases become effectively solved. As a result, research and development resources can be redirected toward newer and more challenging problems.
"Once a use case becomes 'solved', the return on doing incremental R&D on the same topic diminishes," Bernstein said, adding that this could eventually slow the growth of AI training costs and improve operating leverage for AI developers.
— ANI
Reader Comments
Finally someone acknowledging the cost reality. I work in a mid-size Indian startup, and we're already experimenting with open-source models from China for basic tasks like customer email responses. They're surprisingly good for 80% of what we need. The frontier models from OpenAI are overkill for most Indian businesses.
Interesting analysis but the data security concern with Chinese models is real. We've seen clients in government and finance sectors outright ban any Chinese AI tools. The 35-40% TAM estimate might be optimistic if geopolitical tensions worsen.
The key phrase here is "good enough". For a country of 1.4 billion with price-sensitive consumers, good enough AI will win. We already saw this with Reliance Jio vs expensive telecom plans. Indian companies better start building affordable AI stacks now or we'll be importing everything from China. 🎯
It's a shame Bernstein doesn't mention Indian AI initiatives like Bhashini or the government's AI mission. We have the talent and dataset scale to compete in the "commoditised" tier. With our IT services background, India could be the bridge between expensive frontier models and affordable deployment.
I appreciate the nuance - not all AI needs to be a PhD-level genius. As a software architect, I use GPT-4 for complex code reviews but happily use cheaper models for boilerplate generation. This tiered approach is actually more efficient. The real challenge will be training Indian SMEs to adopt these tools effectively.
We welcome thoughtful discussions from our readers. Please keep comments respectful and on-topic.