TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

Morgan Stanley: AI Inference Era Begins, CapEx to Slow in 2028, Funds Shift to Software Layer

2026.09.08 - 07:22
Exclusive
Share

TechFlow reports that according to Chaoxiang Research, in its Sept. 4, 2026 research report, Morgan Stanley noted that data center capital expenditures by hyperscalers will reach $1.5 trillion in 2027, up 60% year over year, before dropping sharply to 12% in 2028. Total AI compute capacity will grow from 35 gigawatts in 2025 to 145 gigawatts in 2028, a fourfold increase. Custom chips will see their share of incremental capacity rise from 34% to 66%, driven by Google TPUs and Amazon Trainium. GenAI returns on investment range from 25% to 50%, with the model layer yielding the highest returns (approximately 46%) by operating APIs on proprietary infrastructure. Approximately 25% of S&P 500 companies have quantified GenAI gains, up from 14% a year ago.

Morgan Stanley believes the deceleration in capex growth will trigger a rotation of capital from the hardware and semiconductor layers to the enablement and software layers, with GenAI enablers such as Amazon, Meta, Google, and Microsoft poised to benefit from earnings upgrades and valuation expansion. The addressable market for digitizing global knowledge work is approximately $20 to $30 trillion, while the consumer segment stands at roughly $30 trillion. AI adoption is expected to spread faster than public cloud computing, with enterprise AI spending potentially reaching $800 billion in 2027. The firm maintains a positive outlook on on-site power generation suppliers BE, INIO, SEI, and electrical enclosure providers WULF, HUT, CIFR, and RIOT.

Add to Favorites
Share to Social Media