The investment cycle in artificial intelligence infrastructure is accelerating faster than expected. According to my calculations based on the latest data, the combined capital expenditures of the four largest cloud service providers (CSPs) will exceed $700 billion by 2026. This is not just a new record — it is a clear signal of the beginning of a supercycle in the semiconductor industry.

The latest reporting period has finally dispelled the doubts of skeptics. Instead of the anticipated slowdown, companies from the "Magnificent Seven" have once again revised their capital expenditure forecasts upward. Hyperscalers are seeing more and more opportunities for investment in data centers, proprietary chips, and computing infrastructure. This is not the peak of the cycle, but merely its midpoint.

Who is spending how much

The main drivers of this boom are four dominant players. Amazon (AWS) will be the largest investor, with spending of around $200 billion directed toward expanding data centers and developing its own Trainium and Inferentia chips. Microsoft (Azure) has raised its forecast to approximately $190 billion, deepening its partnership with OpenAI and expanding infrastructure for Copilot. Alphabet (Google Cloud) has increased its target to $180–190 billion, focusing spending on the TPU network and developing Gemini models. Meta Platforms has significantly raised its capital expenditure plan to $125–145 billion, allocating funds to train open-source Llama models and improve recommendation algorithms.

Tesla has also joined the race, setting a 2026 goal of over $25 billion for deploying AI servers and expanding the Dojo supercomputer. Meanwhile, Apple and NVIDIA are pursuing different strategies: Apple is betting on research and integrating AI into its own chips, while NVIDIA is focused on developing next-generation chips and investing in AI providers such as CoreWeave.

The value chain

Behind the sharp rise in semiconductor sales lies the entire value chain in AI computing. The top tier includes advanced chip manufacturing at TSMC, advanced CoWoS packaging, and HBM memory from SK Hynix. The middle tier consists of the cloud providers themselves, who connect equipment suppliers with AI developers. The bottom tier includes large language models such as OpenAI and Anthropic, as well as a growing number of AI applications.

Noticing the acceleration in demand, providers are expanding capacity — this spurs orders across the semiconductor industry. Sustained demand from cloud providers has already translated into stable orders for GPUs from NVIDIA and AMD, as well as custom ASIC chips.

Global semiconductor sales show a strong positive correlation with the capital expenditures of major tech companies, and both indicators are hitting new records. The key indicator to track for a reversal is the AI monetization coverage ratio, which relates potential AI revenue to capital expenditures. According to my estimates, the ratios for Meta (1.8), Microsoft (1.5), Google (1.1), and Amazon (0.7) confirm that these companies' investments are justified. This means the semiconductor supercycle continues.

Expert opinion: The market is clearly underestimating the scale and duration of the current investment cycle. As long as monetization ratios remain above one for most players, it is premature to talk about an AI infrastructure "bubble." Rather, we are witnessing a fundamental shift that will redefine the structure of global computing power for a decade to come.