A Smarter Way to Build Data Infrastructure

Stanford's Intelligence Per Watt programme finds 77% of AI workloads run on local hardware. Microsoft BitNet runs them on CPUs at 96% less energy. Net-Positive Data Centers (NPDC) is the distributed-compute architecture that hosts them — community-governed, citation-anchored, and built for the infrastructure decade ahead.

Distributed by design

Centralised hyperscale concentrates risk. A federated mesh of 100 kW nodes degrades gracefully.

3 AWS facilities affected, March 2026

Community-governed

Host-community councils hold binding veto. Partnership, not extraction.

$64B in projects redesigned by community input

Net-positive resources

Waste heat and recycled water return to the host community as measurable benefit.

1.45B gal/day projected demand by 2030

The convergence is measured, not asserted

Three independent research threads — Stanford's Intelligence Per Watt programme, Microsoft's BitNet 1-bit LLM framework, and Dr Luci Attala's water anthropology — converge on the same conclusion: distributed, low-power, community-governed compute infrastructure is both technically superior and ethically necessary. NPDC is the architecture that answers all three.

Stanford IPW

Intelligence Per Watt

77% of AI requests are practical tasks that local models (≤20B parameters) answer at 88.7% accuracy. Hybrid local-plus-cloud routing cuts energy, compute, and cost by 60–80%. IPW improved 5.3× from 2023–2025. Read the summary.

Microsoft BitNet

1-bit LLMs on CPUs

BitNet b1.58 runs at 96% less energy than full-precision peers. Ternary weights, 0.4 GB memory, outperforms LLaMA 3.2 1B on standard benchmarks — and runs on commodity CPUs without GPU farms. Read the summary.

Attala (UWP)

Water as constitutive force

Dr Luci Attala's three case studies — Wales, Kenya, Spain — establish water as a constitutive human force, not a resource to extract. Hyperscale water consumption is therefore cultural extraction, not just environmental cost. Read the summary.

March 2026

Three AWS facilities affected

Regional conflict in the Gulf disrupted three AWS facilities in UAE and Bahrain. Centralised infrastructure made the disruption a concentration event; a distributed mesh would have degraded by single-digit percentages.

2024–2026

$64B redesigned by community input

Communities in Naperville, Ohio, and dozens more set higher standards for water, land, and local benefit. $64 billion in planned hyperscale capacity has been delayed or redesigned. The legitimacy floor has moved.

Jan 2027

6 GW grid shortfall

PJM Interconnection projects a 6 GW capacity shortfall across 13 states. Operators who reduce grid dependence — distributed load, on-site solar, waste-heat recovery — gain structural advantage.

77%
of AI workloads that run on local hardware (Stanford IPW)
96%
energy reduction running BitNet on CPUs (Microsoft)
5.3×
Intelligence Per Watt improvement, 2023–2025 (Stanford)
$64B
in projects redesigned by community input (2024–2026)
1.45B
gallons/day projected hyperscale water demand by 2030
6 GW
PJM grid capacity shortfall projected for 2027
Global data-centre industry landscape showing infrastructure shifts, community-driven redesigns, and resource-efficiency opportunities
Crisis map — the 2026 watershed where hyperscale concentration meets community redesign, grid pressure, and water demand. Source: NPDC convergence dossier.

Four pillars, one architecture

Net-Positive Data Centers operates as a distributed-compute architecture across four reinforcing pillars. Each pillar is anchored in one of the convergence research threads above. Together they describe an infrastructure built for the next decade, not the last one.

Epistemic Governance

Compute allocation is treated as a knowledge-justice question, not an engineering optimisation. NPDC's seven-dimensional governance layer (Historical, Indigenous, Cross-cultural, Scientific, Artistic, Marginalised, Future-generational) routes priority compute to underserved knowledge frameworks. Stanford's Intelligence Per Watt finding — that 77% of inference runs locally — makes this allocation tractable in a way it never was on hyperscale.

  • Priority queues for Global South researchers and indigenous institutions
  • Long-term climate modelling over short-term commercial inference
  • CognioEngine integration for dimensional balance

Distributed Resilience

A federated mesh of 100–500 kW nodes replaces the megacampus. The March 2026 Gulf incident affected three AWS facilities and disrupted regional banking — a concentration event the distributed model would have absorbed at single-digit cost. Stanford's hybrid local-plus-cloud routing model maps directly to this mesh: local nodes handle the 77%; cloud escalation handles the residual.

  • No concentration risk: ten nodes offline equals ten percent capacity, not total failure
  • Distributed grid and water load — no single community absorbs the spike
  • Federated routing across 50–100 nodes per regional cluster

Symbiotic Resources

Measurable net-positive returns to host communities. Waste heat routes to district heating or greenhouse agriculture; recycled water augments municipal supply; revenue-shares from compute flow to local government. Microsoft BitNet eliminates the GPU dependency that made hyperscale water-hungry — 96% less energy and CPU-only operation mean a 100 kW NPDC node does not need a reservoir to run inference. Attala's water anthropology grounds the principle: extraction is cultural violence; return is civic infrastructure.

  • Waste heat to district heating, greenhouse agriculture, or industrial process
  • Water recycling returns to municipal supply at greater-than-input volume
  • Revenue-sharing on compute sales flows to host-community council

Transparent Governance

Real-time public dashboards expose water, power, and heat at the node level. Quarterly audits land binding recommendations. Most consequentially, the reversibility clause: a host-community council can trigger decommissioning at NPDC's expense if the net-positive metrics — co-defined with the council at deployment — are not met after twelve months. Trust is not asserted; it is engineered.

  • Real-time public dashboards for water, power, heat, and revenue-share at each node
  • Quarterly audits with binding recommendations
  • Reversibility clause: community can trigger decommissioning at NPDC expense
Network topology — megacampus to MicroDC mesh Two end-state schematic. The megacampus end-state shows a single dense node anchoring the page. The MicroDC mesh end-state shows a distributed lattice of nodes radiating from the same centre. On scroll, the megacampus collapses and the MicroDC mesh resolves. network topology · megacampus → microdc mesh distributed-compute substrate · 50–100 nodes · kinetic resilience 100+ MW single point of failure high-value target state · megacampus · centralised draw 100 kW · ref node distributed targets · low value each 10 nodes lost · 10% capacity state · microdc mesh · distributed substrate 600ms · cubic-bezier(0.16, 1, 0.3, 1) · single signature motion
Network topology — single megacampus vs. distributed MicroDC mesh of 50–100 nodes. The signature motion fires once per session as the figure enters the viewport at 35% intersection.

Megacampus vs MicroDC mesh

Metric Megacampus (current) MicroDC mesh (NPDC)
Power draw 100+ MW per site 100–500 kW per node × 50–100 nodes
Water use ~5M gallons/day 5–10k gallons/day per node, distributed
Land footprint 300+ acres Under 0.5 acres per node
Concentration risk Single point of failure No single node is critical
Grid impact Spike requires major upgrade Distributed load absorbs without upgrade
Community benefit Minimal jobs, concentrated tax Jobs, waste heat, revenue-share, water recycling
Governance Corporate; no community veto Tri-partite: investors, community, epistemic stakeholders
Reversibility None (sunk cost) Community can trigger decommissioning if metrics unmet
100 kW MicroDC reference architecture: technical diagram of power input, compute core, cooling, networking, water system, waste-heat output, community dashboard, footprint, and optional solar
100 kW MicroDC reference architecture — power input, compute core, cooling, networking, water system, waste-heat output, community dashboard, footprint, and optional solar.

Community-governed, net-positive returns

NPDC is the first data-centre architecture designed to return more value than it takes — measurably, transparently, reversibly. The calculator and governance model are the load-bearing primitives.

The Net-Positive Calculator

Community costs (inputs)

  • Water consumed (gallons/day)
  • Power consumed (kWh/day)
  • Land occupied (acres)
  • Utility-rate impact ($/household/month)

Community benefits (outputs)

  • Jobs created (FTE local employment)
  • Tax revenue (property + income)
  • Waste heat delivered (kWh thermal value)
  • Water recycled (gallons returned)
  • Subsidised compute access (schools, libraries, non-profits)
Formula: Net-Positive Score = (Benefits / Costs) − 1.0
Target: > 0.2 (20% net-positive)
Accountability: < 0.0 (extractive) → host-community council can trigger decommissioning

Tri-partite governance

Three stakeholder seats; no single seat can override the others. Decisions require consensus across all three.

Investors

Provide capital. Vote on technical and financial decisions.

Host communities

Provide land, resources, legitimacy. Binding veto over deployment and continuation.

Epistemic stakeholders

Universities, indigenous institutions, Global South researchers. Define knowledge priorities.

No single stakeholder overrides the others. Decisions require consensus across all three seats.

The 2026 pilot programme

NPDC is seeking one forward-thinking host community to deploy the first 100 kW MicroDC reference node.

Requirements

  • A host-community council willing to co-define net-positive metrics
  • Local government open to consortium co-design
  • Existing grid capacity for a 100 kW draw
  • A real use case for waste heat (district heating, greenhouse, industrial process)

What the host community gets

  • Net-positive metrics co-defined by the council at deployment
  • Binding veto over continuation after the twelve-month review
  • Real-time public dashboard (water, power, heat, revenue-share)
  • Revenue-sharing on compute sales
  • Waste heat delivered to the council-nominated use case
  • Decommissioning at NPDC expense if net-positive targets are not met
Apply to the 2026 pilot programme

The consortium

Three pathways to participate. Each seat carries a different responsibility; each carries a different vote.

For investors

The distributed model is risk-adjusted in NPDC's favour:

  • Lower concentration risk: no single high-value site
  • Lower political risk: community governance builds durable consent
  • First-mover advantage in the post-megacampus architecture

Seat: capital contribution plus a voting seat on financial and technical decisions.

Enquire about consortium membership

For host communities

A genuine partnership model, not a host-fee transaction:

  • Measurable net-positive returns, calculator-verified
  • Binding veto: the council governs, not just hosts
  • Reversibility: decommissioning at NPDC expense if metrics are not met
  • A template for epistemic and resource sovereignty

Seat: hosting a node plus a voting seat on deployment and continuation.

Apply to the 2026 pilot programme

For epistemic stakeholders

The first data-centre architecture treating compute as knowledge justice:

  • Seven-dimensional governance: Historical, Indigenous, Cross-cultural, Scientific, Artistic, Marginalised, Future-generational
  • Priority compute access for underserved knowledge frameworks
  • Cogniosynthesis principles embedded in infrastructure decisions

Seat: universities, indigenous institutions, Global South researchers. Voting seat on epistemic priorities.

Enquire about epistemic partnership

The technical whitepaper

Net-Positive Data Centers v1 is in preparation. It will cover the 100 kW MicroDC reference architecture, the epistemic-governance API, the community value-return calculator, distributed-resilience analysis, and regulatory strategy.

The whitepaper does not yet exist as a published document. Leave an email and NPDC will write to you the day v1 publishes.

Get involved

Net-Positive Data Centers Consortium

Distributed-compute infrastructure as civic architecture.

Email: consortium@netpositivedatacenters.org

Research: The convergence dossier

FAQ: Frequently asked questions