I noted that the new standard for everything to be measured is going to be related to power (in kWh/MWh etc).
I recorded that the Maia 300 accelerator delivers 40 % more inference tokens per megawatt compared to leading GPU‑based systems, and up to 7 × the rack‑scale token throughput of Maia 200.
I concluded: “Put the data center on the power.”
I noted that “Modular units can sit next to a wind farm, a solar field, a dam, or a gas well and start running in months.”
These two lines capture the core shift I’m advocating: colocate compute with generation and use modular units for rapid deployment.
I captured the insight that AI infrastructure is not just a compute problem; it is an energy problem too.
I added that if your cloud GPU bill is growing, GPU availability is becoming a bottleneck, or you’re thinking about dedicated compute infrastructure, I’d love to talk.
I recorded that the new standard for everything to be measured is going to be related to power (in kWh/MWh etc).
I noted the new frontier is energy whether you believe it or not.
I concluded that moving power is costly and slow, so I am shifting to a "power‑on‑site" model by colocating modular data‑center units directly beside generation assets.
Adopt a power‑on‑site model for new data‑center capacity.
Moving power is slow and expensive. Moving data is not.
Decision: prioritize development of an operating layer that unifies energy generation, compute, and grid management.
On 2026-09-30 at 15:50 UTC I communicated that AI companies face rising cloud GPU costs, availability bottlenecks, and emerging power reliability constraints, prompting EVORA to offer an integrated compute‑plus‑energy solution.
- GPUs selected for workload profile
We’re currently speaking with a small number of teams in India to understand their requirements and work on few deployments.
Open questions include the exact power‑reliability metrics needed, cost comparisons versus cloud spend, optimal cooling technologies, and scaling constraints beyond the Indian pilot.
- Power provisioning (UPS, redundancy, grid vs generator)
- Cooling architecture (liquid, air, hybrid)
- Deployment tooling (racks, enclosures, remote management)