
Data centers may face temporary power cuts to prevent blackouts on largest US grid
THE SO WHAT
If the largest US grid operator is planning to curtail data center power to prevent blackouts starting next year, uptime assumptions for AI workloads just broke. Operators with latency- or availability-sensitive systems need multi-region, multi-grid, or on-site generation strategies on the table now, not in 2028.
READ THE SOURCE
MORE FROM THE WIRE
Deep & Emerging TechAmkor will spend up to $3bn this year, close to 40% of its revenue
A packaging house putting ~40% of revenue — up to $3B — back into capex is a clear tell that advanced packaging is the new choke point, not just wafer supply. If you rely on high-end GPUs or custom ASICs, start treating OSAT capacity and packaging lead times as first-order constraints in your roadmap.
Deep & Emerging TechTFI boosts profits, navigates jump in TForce Freight volume
Higher TForce Freight volumes with rising operating income suggest LTL and logistics operators can still expand margins if they manage mix and network density. Shippers should expect continued pricing discipline — not a race to the bottom — even as capacity loosens.
Deep & Emerging TechX Money can cut off your account over what you post
Tying stored-value access to social content moderation — where a suspended account can lose X Money access and remaining balance — collapses speech and financial rails into one risk surface. Any fintech or platform considering similar integration needs a clear separation-of-powers model, or they’ll inherit the full blast radius of content policy disputes.
Deep & Emerging Tech‘Those two jobs need different physics’: Rebellions CEO says training and inference need different chips
The Rebellions CEO arguing that training and inference “need different physics” reflects a broader shift toward specialized inference silicon as its own market. Cloud and infra teams should start modeling separate vendor strategies and cost curves for training clusters versus inference fleets instead of assuming one GPU class fits both.