The global technology market may be underestimating a profound structural shift in Artificial Intelligence. Following the release of increasingly competitive Chinese AI models, including Moonshot’s Kimi K3, AI-related equities experienced renewed volatility as investors questioned whether more efficient open-weight models could trigger a price war, compress margins and challenge the economics underpinning today’s Artificial Intelligence investment cycle.
However, short-term market reactions risk overlooking the bigger investment question: where does economic value ultimately accrue when intelligence becomes increasingly abundant?
The answer lies in the tension between two competing models. One focuses on monetising proprietary intelligence through differentiated Artificial Intelligence models; the other focuses on providing the infrastructure required to deploy intelligence at global scale.
Two AI Strategies, One Investment Question
The foundation of the US Artificial Intelligence boom has been built around a proprietary, capital-intensive model. Companies such as OpenAI, Anthropic and Google have pursued increasingly advanced AI systems supported by enormous investments in data centres, specialised chips, proprietary data and computing capacity. This strategy is based on the belief that the most valuable asset in Artificial Intelligence will be ownership of the most capable models.
China’s AI ecosystem has developed under different constraints. Export restrictions on advanced semiconductors have encouraged Chinese developers to focus heavily on computational efficiency, algorithmic optimisation and alternative model architectures rather than relying solely on unlimited access to cutting-edge hardware.
One example is the growing adoption of Mixture-of-Experts architectures. These models operate like a large organisation with hundreds of specialised teams, but instead of activating every department for every task, only the relevant specialists are engaged. This allows models to achieve strong performance while using significantly fewer computational resources.
The result is a narrowing performance gap between leading US and Chinese Artificial Intelligence models across many mainstream applications, while highlighting a key investment debate: does superior intelligence remain scarce enough to justify premium pricing?
From Model Scarcity to AI Abundance
The market is increasingly separating Artificial Intelligence applications into two categories.
The first is the everyday use case: customer service, routine coding, document processing and basic translation. These applications are becoming increasingly exposed to lower-cost alternatives as open-weight models improve.
The second is high-stakes reasoning: advanced scientific research, complex engineering, cybersecurity and mission-critical business applications where reliability and accuracy remain essential. In these areas, leading frontier models continue to command a premium.
This distinction is crucial. If Artificial Intelligence capability becomes sufficiently commoditised for everyday tasks, pricing power may shift away from model providers and towards the companies enabling widespread AI adoption.
The Enterprise Migration
Early enterprise adoption trends suggest businesses are actively seeking more cost-efficient Artificial Intelligence solutions. Companies are increasingly experimenting with open-weight models and alternative providers to reduce AI-related spending while maintaining acceptable performance.
However, this pressure may affect software providers more than infrastructure companies.
The Artificial Intelligence software market resembles a premium bottled water industry. If a competitor begins offering high-quality water for free, the companies selling expensive bottled alternatives face immediate pressure on pricing power. However, the utility company supplying the pipes, infrastructure and distribution network remains essential regardless of which water brand consumers choose.
The same dynamic applies to Artificial Intelligence. The companies developing individual AI models may face greater competition, but the providers of cloud computing, networking, storage and semiconductor infrastructure remain critical to the entire ecosystem.
The Limits of Spending More
The emergence of more efficient Artificial Intelligence models raises a broader question: are we reaching the limits of what additional capital expenditure can achieve?
For years, the prevailing assumption was straightforward: spend more money, acquire more computing power and build a better model. While scale remains important, recent developments demonstrate that algorithmic innovation, engineering efficiency and better data utilisation can narrow capability gaps without requiring unlimited increases in spending.
This does not mean investment in frontier models is becoming irrelevant. The US maintains significant advantages in advanced computing infrastructure, research capabilities and security-sensitive applications. If leading Artificial Intelligence companies can preserve meaningful performance advantages, premium pricing may remain justified.
However, for many standard enterprise applications, the competitive landscape is becoming increasingly balanced.
Conclusion: The Infrastructure Opportunity
The key determinant of long-term AI investment returns will be whether the remaining capability gap between leading AI models is sufficient to preserve pricing power. If the gap remains significant, frontier AI developers can continue generating attractive returns on their substantial investments. If it narrows further, the economics of proprietary Artificial Intelligence models become increasingly challenging.
Yet, a decline in model-level pricing power does not weaken the broader AI investment opportunity. Instead, it may signal a shift in where value is created.
As Artificial Intelligence becomes cheaper and more accessible, adoption is likely to accelerate across industries. This reflects the principle of Jevons’ Paradox: when a resource becomes more efficient and affordable, demand often expands rather than contracts.
This dynamic places major technology platforms such as Alphabet, Amazon and Microsoft in a strategically advantaged position. Regardless of which Artificial Intelligence model ultimately succeeds, enterprises will continue to require secure cloud platforms, advanced computing capacity, networking infrastructure, storage, cybersecurity and data management solutions.
The market remains heavily focused on who will own the intelligence layer. However, the more durable long-term opportunity may lie with the companies providing the essential infrastructure that enables AI adoption to scale across the global economy.
Written by
Mark Muscat
Portfolio Manager, ReAPS Asset Management Ltd
The information contained in this article represents the opinion of the contributor and is solely provided for information purposes. It is not to be interpreted as investment advice, or to be used or considered as an offer, or a solicitation to sell/buy or subscribe for any financial instruments nor to constitute any advice or recommendation with respect to such financial instruments. This article was issued by ReAPS Asset Management Limited, a subsidiary of APS Bank plc. ReAPS Asset Management Limited (C77747) with registered address at APS Centre, Tower Street, Birkirkara BKR 4012 is regulated by the Malta Financial Services Authority as a UCITS Management Company and to carry out Investment Services activities under the Investment Services Act 1994 and is registered as an Investment Manager under the Retirement Pensions Act.
