Sure! Here’s the translation:
AWS, Microsoft Azure, and Google Cloud strengthen their leadership with proprietary chips and global expansion as companies accelerate their cloud migration to deploy generative AI.
Global spending on cloud infrastructure services reached $90.9 billion in Q1 2025, reflecting a year-on-year growth of 21%, according to the latest data from Canalys (now part of Omdia). This increase is directly linked to the massive adoption of Artificial Intelligence applications, which has made the cloud a strategic component for companies across all sectors.
The rise of generative AI is redefining organizations’ technological priorities, accelerating their cloud migration processes and prioritizing cost efficiency during inference, the critical phase where trained AI models are executed. As models are deployed at scale, cloud providers are aggressively investing in optimized infrastructure, including proprietary chips and specialized services.
The Cloud Giants Continue to Dominate the Market
The market podium remains unchanged. Amazon Web Services (AWS), Microsoft Azure, and Google Cloud account for 65% of global spending, showing a combined growth of 24% compared to the same period in 2024. However, the growth rate has been uneven:
Microsoft Azure grew by 33% year-on-year, reaching a 23% market share. Its platform, Azure AI Foundry, used by over 70,000 companies, processed over 100 trillion tokens in the quarter. Moreover, Microsoft improved the energy efficiency of its AI solutions by 30% and reduced the cost per token by more than 50%.
Google Cloud maintained a 10% market share and grew by 31%, despite slight capacity constraints. Its Gemini 2.5 Pro model has led benchmarks in the generative space, and the adoption of its Gemini API and Google AI Studio has increased by more than 200% since January.
- AWS, while still leading with a 32% share, saw its growth slow to 17% due to supply limitations. To compete with NVIDIA, Amazon has enhanced its offering with the Trainium 2 chip, which offers a 30% to 40% improvement in cost-performance ratio. Additionally, it expanded its Bedrock service with models like Claude 3.7 and Llama 4 and announced a $4 billion investment for a new cloud region in Chile in 2026.
AI as an Economic Engine and Challenge
The report highlights that the transition of AI from research to large-scale enterprise deployments focuses on inference costs, which are far more critical than initial training. "Unlike training, which is a one-time investment, inference is an ongoing operational cost," explains Rachel Brindley, Senior Director at Canalys.
Furthermore, many AI services operate under pay-as-you-go models (per tokens or API calls), complicating cost forecasting as usage scales. "This forces companies to limit the complexity of models or restrict their use to high-value cases," adds Yi Zhang, Canalys Analyst.
The Response: Proprietary Chips and Efficiency
To tackle these challenges, hyperscalers are investing in chips specifically designed for AI, such as Trainium (AWS) and TPU (Google). These accelerators, along with new families of cloud instances, aim to reduce total costs and enhance energy efficiency.
Meanwhile, the expansion of data centers continues at a rapid pace. Microsoft opened new facilities in 10 countries just in the first quarter, while Google added its 42nd cloud region in Sweden and allocated $7 billion to its data center in Iowa.
Conclusion
The first quarter of 2025 confirms that the cloud is no longer just a computing platform, but the critical infrastructure for the era of artificial intelligence. Global spending is growing at double digits, driven by the urgent need for businesses to modernize and stay competitive in the tech race. As AI models become more complex and their usage becomes more widespread, the battle for inference efficiency will be as decisive as that for computing power. The cloud giants are already aware of this and have set their machinery in motion.
Source: canalys