Citi TMT takeaways: hyperscaler backlog near $1.7T, AI capex tracking to $3.8T by 2030
sanjaykalra · x · 2026-09-08
Key disclosures from Citi's Global TMT Conference (250+ management teams):
- AI adoption is now infrastructure: hyperscaler backlog near $1.7T end of Q2, up 151% YoY; AI capex tracking toward $3.8T by 2030; new AI power capacity 19 GW this year, 45 GW by 2028.
- Inference shifts agentic: AMD's Matt Ramsay says inference has moved from single chat calls to agentic runs that plan, retrieve, call tools, self-grade and retry dozens of times — breaking seat-based budget models.
- Crossing the chasm: Cognizant's A0–A4 model has AI doing primary work with human review, 40% of engineering AI-assisted; Kyndryl's CEO: "a bullet train on tracks built for 30 mph," with 1,600 agents in its own operations.
- Edge compute: robotics decides the on-device vs data-center debate via latency and data residency.
- Agentic commerce is a data problem: agents mediate discovery, SEO influence fades; your storefront is product data plus API surface.
- Advice for CIOs: own the routing/orchestration control plane; compute is rentable, your context/evals/playbooks aren't; budget per task completed; instrument agents before expanding.
Related event: Citi: Hyperscaler Backlog Nears $1.7 Trillion as AI Capex Eyes $3.8T(3 posts)→
More from Companies & People
- A full AI stack costs $1K/month vs $80K for the org doing the same work — martyjbeard · 2026-09-09
- India clears key spectrum rules, moving Starlink closer to launching satellite internet — DimaZeniuk · 2026-09-09
- Google Cloud and Accenture form FDE unit to catch up in the AI deployment wars — RebeccaBellan · 2026-09-09
- First trailer: Andrew Garfield plays Sam Altman in OpenAI drama 'Artificial' — kyliebytes · 2026-09-09
- Google Cloud partners with Accenture in enterprise AI deployment race — TechCrunch AI · 2026-09-09
- Inworld's TTS model hits GA as CEO explains its end-to-end voice stack — agihouse_org · 2026-09-09