Established
The grid already learns to orchestrate devices.
EVs, batteries, heat pumps and other controllable assets can be scheduled around price, carbon intensity, network conditions and customer needs.
Not affiliated with Octopus Energy or Kraken · working note · September 2026
FROM POWER
TO INTELLIGENCE
Four testable questions at the edge of energy, compute, flexible infrastructure and accountable AI agents — deliberately framed as experiments, not as a grand “AI strategy”.
A battery stores electrons.
A compute node consumes electrons and produces intelligence.
Why shouldn’t the energy system orchestrate both?
A compute node consumes electrons and produces intelligence.
Why shouldn’t the energy system orchestrate both?
01 / COMPUTE FLEX
KRAKEN / FLEXIBILITYHypothesis
What comes after the battery, EV and heat pump?
Could compute itself become a controllable workload class?
Not every workload is movable. But batch inference, embeddings, rendering, fine‑tuning and selected agentic tasks often are.
workload→scheduler→energy signal→compute
02 / REVENUE PER GRID CONNECTION
GENERATION / INFRASTRUCTUREEstablished
A grid connection is more than generation.
Storage, charging, private networks, flexibility and heat reuse can all change the economics of an energised site.
Hypothesis
Stop asking only “how many MW?”
Ask: what is the highest‑value stack that can sit on this grid‑connected footprint?
generation→storage→charging→connectivity→compute→heat/services
EXTREME CASE / POWER SOURCE ≠ CONTROL LAYER
Microreactor field console ↗
Wind is variable. Nuclear is steady. AI inference is spiky. Batteries are fast. The orchestration problem exists in every case. A microreactor is therefore interesting here not as a new Octopus/Kraken thesis, but as a stress-test of the same Compute Flex idea.
Separate remote / industrial infrastructure note — not a proposal for the Drysdale urban micro-pilot.
03 / ACCOUNTABLE AGENT AUTHORITY
AI FOUNDATIONS / AGENT SAFETYEstablishedGap to test
Safe is necessary. Authorised is different.
Authentication, permissions, guardrails, evals and observability can constrain an agent. Regulated systems also need to answer a different question:
Who had the authority to let this agent do this?
Working prototype
MandateEvidence
A compact evidence object connecting principal → issuer → capacity → case → powers → scope → validity → revocation → provenance.
The point is not another identity layer. It is independently verifiable authority and post‑action accountability across organisational boundaries.
04 / THE DRYSDALE MICRO‑PILOT
ONE LONDON BLOCK / REAL ASSETSPilot thought experiment
What if the first Power‑to‑Intelligence node is just… an ordinary building?
Use a single urban block as a living‑lab boundary: existing housing, a public building, roofs, weekday loads, local users and real planning constraints.
Test sequence
fabric retrofit→heat pump→rooftop PV→battery→local compute→private AI service→flex scheduling
Measure both kWh and human outcomes.
This is a prototype frame, not a claim that any specific building, owner, council or company has agreed to participate.
FIG.04 // The Drysdale Block — Spatial IndexN1 6LT–6ND · 1:200 · ISO 30°
A jobcentre sits at 30 Drysdale Street, metres from a tech gold rush. The state's AI there is real, funded and pointed at fraud. The AI pointed at getting you a job — that one doesn't exist yet. This is the case for building it, offline, for the price of a phone.
The experiment is not “can we add AI to energy?” It is whether existing energy capabilities can make intelligence physical, schedulable, financeable and accountable.
WHY I’M ASKING
NOT A GENERIC JOB PITCH1×regulated digital bank built and licensed from scratch
20+early fintech investments across multiple markets
3current frontier threads: AI infrastructure, digital identity, accountable agents
20 minis enough to tell me where this reasoning is wrong
Worth comparing notes?
I’d rather test one sharp question with someone who actually builds the system than turn it into another 40‑page memo.