The Shift
From space to tokens to utility.
How data-centre value moved — and why the integrated operator wins what comes next.
For twenty years, data centres sold space: square metres, kilowatts, uptime. The tenant owned the servers — and the upside.
Cloud changed the unit to GPU-hours. Intelligence stayed the customer's problem.
AI factories make intelligence the product. The unit is the token — a metered piece of finished reasoning. Not all tokens are equal: the right model on the right silicon, at the right latency, is worth many times a badly produced one.
Refining is the next step. Crude tokens are a commodity with a falling price. Refined tokens — combined with sector data, workflows and partners — are a product with a rising one: a fraud decision, a triage suggestion, a citizen service in Bahasa Indonesia.
Four eras of value
Where the value sat, and who captured it.
1 · Colocation - Space & power
Sold m², kW and uptime. Landlords captured the value.
2 · Cloud - AI as a service
Sold GPU-hours. Clouds captured the value.
3 · The token factory
Sells metered tokens. The integrated operator captures the value.
4 · The token refinery
Sells utility. The refiner captures the value.
What “best tokens” means
Four tests, applied to every workload.
Best fit
Each workload runs on the silicon that gives the highest quality per cost — GB300 for training, accelerators for inference, FPGAs for deterministic tasks.
Best cost
Contracted power, PUE below 1.3, KV-cache and disaggregated serving.
Best quality
Models tuned for Indonesian language, data and regulation.
Best delivery
Metered, billed and governed on one platform, with data residency guaranteed.
Why integration is the moat
You cannot refine what you do not control.
Every layer we own removes a margin, a hand-off and a failure mode — and lets a token travel from GPU to application without leaving our environment.