The Data Center Bargain

The debate has changed

Data centers are now public-scale infrastructure.

Lawrence Berkeley National Laboratory’s 2026 update projects a 2030 reference case of 649 TWh for U.S. data centers, approximately 11.8% of national electricity use. Its compounded uncertainty range is 521–843 TWh. FERC now requires regional grid operators to address cost shifting, transparency, generation adequacy, and flexible large loads. Pennsylvania has moved toward full energy-cost recovery, local approval, public disclosure, environmental protection, and community benefit agreements.

Benefits can be national.

AI capability, corporate value, and national productivity can rise far from the host community.

Burdens stay local.

Substations, transmission, water withdrawals, generator emissions, tonal noise, roads, emergency services, and land-use change remain physically concentrated.

The accounting can hide the bargain.

A profitable facility can still shift costs to ratepayers or leave the most affected household worse off. Aggregate growth is not proof of local benefit.

The rules, written as math

Pay all infrastructure costs first, then share what is left.

In plain terms: a project must first pay for everything it actually costs the community (grid, water, roads, safety, cleanup) before any profit-sharing begins. Only what remains after those costs is available to distribute.

Let verified gross public rent or committed project payment be $G$. Let $C$ hold every senior claim: grid, reliability, water, environmental, public-service, security, audit, and decommissioning. The distributable reserve is:

\[R=(G-C)_+.\]

Gross committed value $G$

Grid and reliability

Water, land, emissions

Public services and audit

Decommissioning and safety

Available to share $R$

For household $i$, let $L_i$ be the verified project-caused loss, $B_i$ the direct project benefit that actually reaches that household, and $m$ the universal resident floor. Define:

\[d_i=(L_i-B_i)_+,\qquad h_i=\max\{m,d_i\}.\]

The transfer $h_i$ is the smallest payment that both gives the household the universal floor and ensures the project does not make it worse off. Across $n$ households, the total reserve needed is:

\[H(m)=\sum_{i=1}^n\max\{m,d_i\}=nm+\sum_{i=1}^n(d_i-m)_+.\]

Local admission condition: the cash bargain is funded only when $G\ge C+H(m)$. A promised compute floor $c$ is separate and requires $Q\ge nc$. Compute cannot silently replace cash compensation.

Why the floor matters

An equal dividend can still leave some households worse off.

Suppose a data center causes $2 of harm to one household (well dries up, property value drops, noise) and gives every household a $1 dividend. That household gets $1 back but lost $2. They are still down $1. The dividend is fully funded and sounds fair, but it did not protect the household that was actually harmed.

−1

The smallest counterexample

One household loses 2 from the project and gets a dividend of 1 like everyone else. The dividend is fully funded, but the household is still worse off by 1. The correct payment to that household is 2, not 1.

The fix: floor plus top-up. Pay every household the base amount $m$. Then add a top-up only to households whose verified loss exceeds the base. After the floor and top-ups are funded, distribute any remaining surplus to households with greater need first.

Important: physical limits are vetoes, not prices. A community benefit payment must not buy permission to breach a critical water, reliability, safety, or pollution boundary.

A policy local governments can use

Fourteen specific policy moves a local government can make.

Conditional use only

Require special-exception or conditional-use approval above local load, area, water, generator, or expansion thresholds. Do not treat a fast-changing high-energy facility as an ordinary by-right warehouse.

No secret project

Require beneficial ownership, end users, peak MW, water, PUE, WUE, backup power, noise, jobs, incentives, phasing, and infrastructure disclosures before the first discretionary hearing.

Growth pays for growth

Make the project fund attributable generation, transmission, distribution, substations, water, roads, emergency services, studies, stranded-cost security, and decommissioning.

Hard physical envelopes

Put enforceable peak-load, ramp, water, drought, noise, generator, emissions, lighting, setback, habitat, and safety limits into the permit.

Security before operation

Require bankruptcy-resistant escrow, bonds, letters of credit, or parent guarantees sufficient for remediation, unfinished infrastructure, public-benefit nonperformance, and site restoration.

Surplus-sharing agreement

Combine a nonavoidable capacity or concession payment, full senior-cost recovery, and a variable share of audited surplus value. Never rely on accounting profit alone.

Universal benefit + top-ups

Use lawful bill credits, rebates, resident trust distributions, energy assistance, or equivalent broad benefits. Then target extra compensation at households with verified deficits above the floor.

Distinct host-zone share

Give neighborhoods closest to substations, corridors, cooling, generators, and construction an additional upside share beyond make-whole compensation.

Verified load flexibility

Require a tested curtailment amount, response time, duration, rebound limit, emissions boundary, periodic drills, and penalties for nonperformance.

Conditional incentives

Make each year of a tax benefit depend on current compliance. Suspend or claw back benefits for broken reporting, job, infrastructure, environmental, or community-benefit commitments.

Public evidence and challenge

Use independent meters, machine-readable reports, signed epoch receipts, resident complaints, audit rotation, conflict disclosure, and escrow for contested amounts.

Anti-speculation milestones

Expire or charge for unused capacity reservations. Reopen dormant permits, inflated forecasts, serial transfers, and projects that miss financing, construction, or energization milestones.

Protected rule review

Require annual review, a major periodic review, and reopening after material expansion or ownership change, while preventing amendments from quietly weakening senior claims or hard limits.

Regional compact

Coordinate minimum disclosure, bond, water, noise, audit, tariff, and benefit standards so communities do not compete by shifting costs across the same grid and watershed.

Open the standalone local-government playbook Inspect the policy source

Post-AGI economics

More AI capability does not automatically lower prices for households.

Let $a_t$ be effective productivity that actually reaches the household bundle after automation coverage, diffusion, access, and pass-through. Let $r_t$ be the scarce-resource and infrastructure factor, and $\mu_t$ the markup and rent-extraction factor.

\[\frac{P_{t+1}}{P_t}=\frac{r_t\mu_t}{a_t}.\]

Prices fall in that period exactly when $a_t>r_t\mu_t$. Raw model capability can double while diffusion is blocked, energy grows scarcer, physical bottlenecks remain, or monopoly rents absorb the savings. And even falling prices do not guarantee household abundance if wages and asset ownership fall faster.

The open question is distribution. The Alignment Theorem does not assume that AGI automatically benefits everyone. It provides a set of rules that can direct verified surplus into cash, compute, resilience, and constrained household agents, and it rejects any deal that is not fully funded.

Why the full version requires Tau Language

The rules must be executable, inspectable, and safely revisable.

A conventional chain can hold money. A legal document can state duties. Lean can prove theorems. Ordinary software can execute a rule. The full Alignment Theorem needs all of those functions in one living rulebook: the rules must be formal, decidable, executable, inspectable, and safely revisable even as they change. Tau is the only language that unifies all of these.

Layer	What it does well	What stays external
Statute / permit / contract	Democratic and legal authority	Machine execution, exact rule replay, automatic invariant checking
Conventional software	Fast operational execution	Policy meaning is buried in code; amendments may change unintended behavior
Conventional smart contract	Custody and deterministic settlement	Reasoning about a proposed replacement rulebook is normally off-chain
Lean / theorem prover	Deep proofs about a fixed formal model	Live policy execution, collaborative rule evolution, custody, and settlement adapters
Tau Language	Decidable, executable rules that can reason about proposed changes to themselves before adoption	Physical truth, legal authority, identity, meters, custody, and production hardening still require external systems

Only Tau can do this, and the mechanism is patented by Ohad Asor. No other language combines all of these functions. The full version requires a language where the rules themselves are first-class objects that can be reasoned about, proposed changes can be checked against requirements before adoption, and accepted rules remain executable. Tau is the only language with that combination.

The ten-fact local rule gate

The current research packet admits one bounded project action only when all ten verified facts are true:

✓Current policy root

✓Authenticated project identity

✓Authenticated local consent

✓Incremental energy costs reserved

✓Reliability and curtailment plan valid

✓Water, emissions, and land limits valid

✓Decommissioning bond funded

✓No-harm compensation funded

✓Universal cash and compute floor funded

✓Public audit receipt current

allow[t] =
  policy_root_ok[t]
  & project_identity_authenticated[t]
  & local_consent_authenticated[t]
  & incremental_energy_costs_reserved[t]
  & reliability_curtailment_plan_ok[t]
  & water_emissions_land_limits_ok[t]
  & decommissioning_bond_funded[t]
  & no_harm_compensation_funded[t]
  & universal_dividend_compute_floor_funded[t]
  & public_audit_receipt_current[t]

Tau does not authenticate the facts; it prevents the settlement layer from pretending that an absent, stale, disputed, or failed obligation has passed.

The trust boundary is external. The gate enforces that the declared relation was followed. It cannot verify that the facts fed into that relation were true. A meter can be miscalibrated, an oracle can be bribed, and an attestation can be forged; the chain faithfully executes on whatever validly-signed inputs it receives. This is why the architecture requires independent meters, audit rotation, conflict disclosure, dispute mechanisms, and escrow for contested amounts. The theorem assumes honest authentication; it does not provide it.

Local law and agreement

Signed meter, bond, audit and approval facts

Tau rule gate

Escrow / permit / settlement

Cash, compute and public receipts

Proof-constrained revision

What has actually been established

What has been proven, what runs today, and what still needs external authority.

Evidence lane	Result	Safe interpretation
Lean 4.33	Social-contract proofs and axiom audit pass in GitHub Actions	The formal model confirms that each household receives its minimum guarantee, the total budget is feasible, no household is worse off, cash and compute benefits are tracked separately, and the ten-fact gate implication holds
Exact finite campaign	65,100 profiles; 1,348,200 individual checks; 319,417 reserve boundaries; zero modeled welfare regressions	Strong evidence from exhaustive testing within the model, not a substitute for empirical project data
Tau rule packet	1,024 complete Boolean rows; only the all-true row admits	The checked-in gate is a complete conjunction and every single-fault row rejects
Actual Tau candidate	Earlier V1.1, dividend, and wealth packets ran correctly on a local build pinned to a specific source commit	Evidence for those exact packets on that build only
New local data-center gate	Native replay in development	The exact new 1,024-row packet has not yet executed on the reviewed Tau binary or on Tau Net
Tau Testnet	Alpha system with explicit limitations	Useful for testing architecture; not production authority or a safe home for real public funds

The theorem verifies the rules. The world still needs separate verification: lawful authority, independent measurement, causation methods, identity and appeals, secure custody, exact interpreter provenance, security review, and empirical welfare evaluation.

A practical path

Local governments can start before Tau Net is production-ready.

0–30 days

Immediate

Initiate conditional-use zoning, standard disclosures, applicant-funded independent review, and an inventory of pending applications and grandfathering exposure.

31–180 days

Near-term

Adopt hard envelopes, bonds, a community-benefit term sheet, clawback rules, public meters and dashboards, a complaint process, and regional minimum standards.

Ongoing

Shadow Tau pilot

Translate every legal obligation into a verified Boolean fact. Run the Tau gate beside human administration without moving value. Compare every decision, dispute, appeal, and amendment before production use.

The significance. The data-center debate is usually framed as acceleration versus obstruction. The Alignment Theorem reframes it as rule design. Build when the project can prove that growth pays for growth, hard limits are respected, the most affected household is protected, broad residents receive a floor, compute promises are real, and the rules cannot be weakened in the dark.

The goal is not to tax intelligence into scarcity. The goal is to make expanding intelligence lower real household scarcity and to refuse any project that cannot fund the promises used to obtain consent.

Primary sources

Is the bargain funded?

Data centers are now public-scale infrastructure.

Benefits can be national.

Burdens stay local.

The accounting can hide the bargain.

Pay all infrastructure costs first, then share what is left.

An equal dividend can still leave some households worse off.

The smallest counterexample

Fourteen specific policy moves a local government can make.

Conditional use only

No secret project

Growth pays for growth

Hard physical envelopes

Security before operation

Surplus-sharing agreement

Universal benefit + top-ups

Distinct host-zone share

Verified load flexibility

Conditional incentives

Public evidence and challenge

Anti-speculation milestones

Protected rule review

Regional compact

More AI capability does not automatically lower prices for households.

The rules must be executable, inspectable, and safely revisable.

The ten-fact local rule gate

What has been proven, what runs today, and what still needs external authority.

Local governments can start before Tau Net is production-ready.

Immediate

Near-term

Shadow Tau pilot

Public evidence behind the policy context.