Barclays U.S. Internet & Semiconductors: Framework for Modeling AI Demand & Supply — Capex 'Peak' Likely in 2028
Summaries are public. PDF access requires an active membership; all members have the same access. Sign in
Barclays introduces a comprehensive framework suggesting AI capital expenditure will peak at approximately $1 trillion in 2028, exceeding current consensus estimates by over $300 billion. This projection is underpinned by the rapid revenue scaling at leading AI labs, with OpenAI's annual recurring revenue surging from $6 billion in early 2025 to $25 billion currently, and Anthropic growing from $1 billion to $19 billion.
The report anticipates that early forms of recursive self-improvement could emerge by 2027. AI is expected to transition from assisting workflows in 2025 to agentic AI capabilities in 2026, progressing toward autonomous ecosystems by 2027. Barclays estimates the AI industry will require roughly 23 gigawatts of new computing capacity by 2029, with training compute peaking at 15.5 gigawatts for OpenAI and Anthropic alone.
Hyperscaler capital expenditure is projected to consume approximately 90% of operating cash flow in 2028 before subsequently stabilizing. The bank emphasizes that tracking AI lab compute requirements serves as the most reliable leading indicator for future infrastructure investment.
The framework carries notable uncertainty, meaning the actual spending peak could shift earlier or later depending on the pace of technological breakthroughs. Currently, OpenAI and Anthropic together account for roughly two-thirds of industry compute demand, with Google, Meta, and xAI comprising the remaining third, though the latter group is expected to gain share long-term.