AI Spending Spree Reduces Big Tech Buybacks

🔑 Enhanced Key Takeaways

•The combined capital expenditure (capex) of major hyperscalers, including Alphabet, Amazon, Meta, Microsoft, and Oracle, nearly tripled from $162.3 billion in 2022 to $448.3 billion in 2025, with projections reaching $755 billion in 2026.
•This strategic shift has resulted in a substantial reduction in stock buybacks, with combined buybacks by these companies plummeting to $12.6 billion in Q4 2025, marking a 74% decline from their $48 billion peak in 2021.
•A significant portion of the escalating capex is attributed to sharply rising component costs, particularly for memory chips, rather than solely an increase in physical capacity.
•AI data centers demand 3-5 times more power per square foot than traditional facilities, with individual AI server racks consuming 50-150 kilowatts, leading to considerable strain on power grids and increased water usage for cooling.
•The increasing capital intensity of AI is transforming the software industry into a capital-heavy business, as cloud and AI platforms increasingly resemble utilities due to massive fixed costs for infrastructure.

🛠️ Technical Deep Dive

AI data centers are specifically designed for computationally intensive tasks like training and running machine learning models, optimizing for parallel processing with specialized chips such as GPUs and TPUs.
Modern GPUs for generative AI consume between 700-1,200 watts per chip, a significant increase compared to the 150-200 watts used by traditional CPUs.
AI server racks typically require 60 or more kilowatts of power, substantially higher than the 5-10 kilowatts needed for standard data center racks.
The high-density computing demands of AI workloads necessitate advanced cooling solutions, with liquid cooling becoming a standard requirement in new data center builds due to the immense heat generated.
AI infrastructure requires high-performance, lossless network fabrics to facilitate efficient GPU-to-GPU communication at scale, alongside systems-level designs that can support extreme power, cooling, observability, and automation requirements over time.
Key hardware components include NVIDIA H100 GPUs, priced around $27,000-$40,000 per unit, and A100 GPUs, costing approximately $10,000-$17,000 per unit, with newer generations like the B200 also entering the market.
Custom Application-Specific Integrated Circuits (ASICs), such as Alphabet's Tensor Processing Units (TPUs) developed with Broadcom, are utilized for enhanced energy efficiency in AI inference tasks.

🔮 Future ImplicationsAI analysis grounded in cited sources

Big Tech's reliance on debt to finance AI infrastructure will intensify.

Goldman Sachs projects that hyperscalers' AI spending is on track to consume 100% of their operating cash flows, potentially necessitating an additional $400 billion in net debt to bridge the funding gap.

The shift towards AI capital expenditure will lead to a sustained moderation in S&P 500 share buyback growth.

Goldman Sachs anticipates only a 3% growth rate for buybacks across the S&P 500 in 2026, a significant deceleration from historical trends, as companies prioritize AI investments.

The extensive AI infrastructure build-out will increasingly encounter challenges related to power and water availability.

Planned data center projects in the U.S. alone could require an additional 780 gigawatts of power by 2030, surpassing the country's current peak load, while water consumption for cooling is projected to double or quadruple 2023 levels annually, with many new data centers located in high water-stress regions.

⏳ Timeline

2020

NVIDIA A100 GPU released, becoming a key component for AI training and inference.

2021

Combined stock buybacks by Amazon, Alphabet, Microsoft, Meta, and Oracle reached a peak of approximately $48 billion.

2022

Combined capital expenditure for Alphabet, Amazon, Meta, Microsoft, and Oracle totaled $162.3 billion. NVIDIA H100 GPU was launched.

2023-06

AI spending transitioned from gradual growth to a steep acceleration, marking a full-scale infrastructure buildout.

2025-12

Combined stock buybacks by major hyperscalers fell to $12.6 billion in Q4, the lowest level since early 2018.

2026

Major hyperscalers are projected to spend $755 billion on AI infrastructure.

AI Spending Spree Reduces Big Tech Buybacks

⚡ 30-Second TL;DR

🧠 Deep Insight

🔑 Enhanced Key Takeaways

🛠️ Technical Deep Dive

🔮 Future ImplicationsAI analysis grounded in cited sources

⏳ Timeline

📎 Sources (20)

👉Related Updates

vivo X300 series sales surpass 3 million units

Investors Pivot to AI Infrastructure and Power Suppliers

AI Could Slash Animation Production Costs by 90%

BlueBay Forecasts Long-Term Growth for Japanese AI Stocks