Compute Capacity

Alphabet plans a major AI infrastructure expansion, requiring substantial debt financing, including a $20 billion bond sale with a 100-year sterling tranche. This move addresses immense compute capacity demands for AI training and inference, but raises concerns about increased costs, operational complexity, and potential liabilities. The company anticipates capital expenditures potentially reaching $185 billion, more than double last year’s. While AI, particularly Gemini, shows rapid user growth, it poses a challenge to Google’s core advertising business, despite recent revenue increases. Alphabet’s investment mirrors that of other tech giants, collectively boosting capex significantly for AI development.

OpenAI aims for widespread AI adoption by 2026, focusing on integrating AI into healthcare, research, and enterprise. The company is scaling its compute capacity, projected to reach 1.9 GW by 2025, to support this growth and its monetization strategies, including potential IPO plans. A significant investment from Nvidia to bolster compute infrastructure is reportedly uncertain, highlighting the challenges of securing essential resources for AI’s future.

At Re:Invent 2025, AWS unveiled Trainium 3, a custom AI‑training chip delivering roughly four‑fold performance and energy gains, promising up to 50 % cost cuts. It also introduced AWS Factories, an on‑premise service that blends Trainium accelerators with Nvidia GPUs for a full‑stack AI solution. AWS added 3.8 GW of compute in the past year and targets over 12 GW by 2027, which analysts say could generate $150 billion in annual revenue. The dual hardware strategy aims to reduce GPU‑dependency, enhance supply‑chain resilience, and sharpen AWS’s competitive edge against Azure and Google Cloud.

Google faces the challenge of doubling its AI compute capacity every six months to meet surging demand. VP Amin Vahdat revealed a need for a 1000x capacity increase in 4-5 years, highlighting AI infrastructure competition as critical and costly. Google focuses on custom silicon like TPUs for efficiency, and leverages DeepMind research. CEO Sundar Pichai acknowledged AI bubble concerns, emphasizing cloud business strength and disciplined investment to ensure long-term sustainability and maintain a competitive edge.

Microsoft is diversifying its AI strategy by investing $5 billion in Anthropic, alongside Nvidia’s $10 billion stake. Anthropic commits $30 billion to Microsoft’s Azure compute and secures contracts for up to one gigawatt of compute capacity. Nvidia and Anthropic will collaborate on AI model and hardware optimization. This move follows Microsoft’s substantial investment in OpenAI, signaling a potential shift to broader AI partnerships. Nvidia’s CEO Jensen Huang praised Anthropic’s work.

Microsoft is pursuing self-sufficiency in data center infrastructure by increasing its use of custom-designed chips. CTO Kevin Scott emphasized the company’s commitment to securing optimal performance, currently relying on Nvidia and AMD while actively deploying its Azure Maia AI Accelerator and Cobalt CPU. Microsoft aims for complete system design, including cooling and networks, and acknowledges an industry-wide compute capacity shortage despite massive AI investments. This strategy mirrors similar efforts by Google and Amazon for performance, efficiency, and cost advantages.