This transformation has far-reaching implications for competition, regulation and market dynamics.
In this Q&A, we explore what’s driving this change, whether it challenges cloud dominance or simply shifts control elsewhere, and what it means for businesses, consumers and policymakers in 2025.
1: Your article suggests a big shift is taking place in AI competition dynamics. What’s behind this move from cloud-based AI to local inference?
AI models go through three key stages: training, fine-tuning and inference. Training and fine-tuning require substantial computational power and will remain cloud-based. However, inference — the process of applying a trained model to generate outputs — is increasingly moving to devices rather than relying on cloud infrastructure. This shift is being driven by cost efficiencies, faster response times and enhanced privacy.
Recent advances in model efficiency, such as OpenAI’s 4o-mini and DeepSeek’s R1 distillations, are accelerating this trend. Distillation is a technique that compresses a large AI model into a smaller, faster version while maintaining much of its capability. These smaller models allow AI to run directly on consumer devices, making AI more accessible while reducing reliance on the cloud. However, cloud services will remain crucial for training new models, running large-scale applications and supporting AI integration into enterprise systems.
2. For years, competition regulators have worried about AI reinforcing cloud monopolies. Does this shift to on-device AI change that concern?
It adds complexity rather than resolving the issue. While on-device AI may reduce reliance on cloud computing for certain tasks, it does not necessarily lead to greater competition. Instead, control is shifting to other parts of the AI value chain, such as operating systems, semiconductor supply chains and digital platforms. Regulators who focus solely on cloud dominance risk missing where competitive bottlenecks are actually forming. The key question is whether this shift will genuinely create opportunities for new entrants or simply consolidate market power in different areas.
3. If AI moves away from the cloud, does that mean major cloud providers such as Microsoft, Amazon and Google are at risk?
Not at all. Cloud providers remain central to AI’s development, as training and large-scale deployments still require substantial computing power. While smaller, efficient models are enabling more on-device inference, OpenAI’s release of Deep Research moves in the opposite direction: it relies on extremely heavy compute even for inference, reinforcing the role of cloud infrastructure in running the most advanced AI models. Rather than being displaced, cloud providers are adapting. Many are embedding AI directly into their products, ensuring they maintain influence over how AI is accessed and used. Their focus is likely to shift towards hybrid AI models, enhanced hardware integration and AI-powered enterprise services. The cloud will continue to be a foundational element of AI, even as inference becomes more distributed.
4. What are the biggest implications of on-device AI for businesses and consumers?
For businesses, on-device AI reduces operational costs, improves performance and enhances data privacy. It allows companies to offer AI-driven products that do not require constant internet connectivity or expensive cloud-based infrastructure. For consumers, benefits include faster response times, improved security and greater control over personal data, as sensitive information can be processed locally rather than being transmitted to external servers. However, as AI becomes increasingly embedded in operating systems and hardware, it may also create new dependencies, potentially reinforcing the market power of large technology firms.
5. How does this shift affect regulatory priorities? Will competition authorities need to rethink their approach to AI?
Yes, regulators will need to adapt. Until now, much of the regulatory focus has been on cloud concentration and access to computing power. However, if AI is increasingly integrated into consumer devices, enterprise software and proprietary ecosystems, competition concerns will extend beyond cloud infrastructure. Authorities will need to consider how AI is controlled and distributed across different layers of the digital economy, from hardware and operating systems to app stores and enterprise platforms. A narrow focus on cloud competition risks overlooking the broader competitive dynamics shaping AI’s deployment.
6. Could this decentralisation of AI give smaller players an edge, or will it simply reinforce the dominance of existing tech giants?
It presents both opportunities and challenges. The lower cost of running AI locally could allow smaller firms to develop innovative applications without relying heavily on cloud infrastructure. However, major technology firms are already positioning themselves to capitalise on this shift by embedding AI into their own ecosystems. This could make it difficult for smaller firms to compete unless there is a genuinely open and interoperable AI environment. Whether decentralisation fosters competition will depend on whether firms have meaningful access to AI capabilities without being locked into dominant platforms.
7. What is the most important question in AI competition that businesses and regulators should be asking in 2025?
Who ultimately controls access to AI? The competitive landscape is no longer defined solely by who builds the most advanced models, but by where and how those models are deployed. However, this is not just a question of competition. There is an inherent tension between keeping AI open and ensuring security, privacy and economic resilience. Governments increasingly see AI as a strategic asset, leading to policies that favour domestic innovation and tighter controls over key AI technologies. Regulators and businesses must navigate this complexity, balancing competition, security and economic growth without entrenching new forms of market dominance.