The Ledger
- Anthropic's custom-silicon bet on inference is the clearest sign yet that model labs now treat compute as a design constraint, not just a procurement line.
- Nuclear restarts have moved from press release to signed power purchase agreement — the financing structure is what actually gets a reactor back online.
- OpenAI's chip-supply deals are quietly turning into equity stakes for the chipmakers on the other side of the table.
- AI tooling is reopening a conversation executives thought was closed: offshoring, on new math.
- A handful of gigawatt-scale datacenter customers are now large enough to force utilities to rewrite decade-old grid plans on their own.
Datacenter power demand forces utilities to rewrite decade-old grid plans
Load forecasts built for slow, predictable growth are being redone around a handful of gigawatt-scale customers.
Coverage of the compute economy, as it happens.
Enterprise software vendors race to bundle agents before customers build their own
Incumbent SaaS vendors are shipping agent features fast, betting that a bundled offering beats losing the workflow entirely to a customer-built alternative.
Anthropic doubles inference capacity with a custom-silicon bet
A second-generation accelerator order shifts a meaningful share of serving traffic off general-purpose GPUs for the first time.
Behind-the-meter gas turbines become the fast lane around grid queues
Datacenter developers are increasingly bypassing multi-year interconnection queues by generating on site.
Enterprise AI budgets shift from pilots to production line items
CFOs are moving AI spend out of innovation budgets and into recurring operating expense — a signal the pilot era is ending.
Nuclear restarts move from press release to power purchase agreement
The first wave of retired-plant restart announcements is now converting into signed, priced contracts with datacenter operators.
Executives quietly restart offshoring conversations as AI reshapes headcount math
AI-assisted productivity gains are changing the math on where routine engineering and support work gets done, not just how much of it is needed.
The compute economy's top stories, delivered on a schedule you choose.
AI
View allAnthropic doubles inference capacity with a custom-silicon bet
A second-generation accelerator order shifts a meaningful share of serving traffic off general-purpose GPUs for the first time.
Enterprise AI budgets shift from pilots to production line items
CFOs are moving AI spend out of innovation budgets and into recurring operating expense — a signal the pilot era is ending.
Power
View allDatacenter power demand forces utilities to rewrite decade-old grid plans
Load forecasts built for slow, predictable growth are being redone around a handful of gigawatt-scale customers.
Behind-the-meter gas turbines become the fast lane around grid queues
Datacenter developers are increasingly bypassing multi-year interconnection queues by generating on site.
Nuclear restarts move from press release to power purchase agreement
The first wave of retired-plant restart announcements is now converting into signed, priced contracts with datacenter operators.
Tech
View allEnterprise software vendors race to bundle agents before customers build their own
Incumbent SaaS vendors are shipping agent features fast, betting that a bundled offering beats losing the workflow entirely to a customer-built alternative.
Executives quietly restart offshoring conversations as AI reshapes headcount math
AI-assisted productivity gains are changing the math on where routine engineering and support work gets done, not just how much of it is needed.