Every node running eWatts is a continuous bid — ~12.5 Wh per block — competing for a fixed reward of 100 eWatts. Entry requires no capital purchase: the hardware is commodity DRAM people already own. Competition drives margins toward zero, and the network finds its own size. From that single condition, the energy anchor emerges as a market identity, not a protocol parameter.
Source: constants.rs, v0.5 (AOPS migration, protocol version 0x0005). Consensus constants — the protocol knows nothing else. Everything below the fold is market, not code.
Network size N is not an input here — it is the output. Given a market price P, electricity cost, hardware wall power, and the electricity share of marginal cost, free entry determines how many nodes the reward can sustain. The protocol never sees any of these variables; they act on it from outside.
Ten years of hardware efficiency drift at the selected rate. Wall power per node falls; free entry converts every efficiency gain into more nodes; the kWh-per-eWatt anchor stays flat. Indexed to 100 at year 0. This is the mechanism that replaces programmed work escalation.
N* = φ·100·P / c. Higher price sustains more nodes; cheaper power sustains more nodes at any price. Log scale.
VR = φ·P / p_elec — a straight line whose slope is set by electricity cost. The anchor follows price; it does not prop it up.
Consensus-side: ~5.26M eWatts/year, effectively linear regardless of network size or price. Predictable issuance without a halving calendar.
Declines as 1/t from fixed absolute issuance. References: BTC 0.83%, gold ~1.7%, USD M2 ~6.5%.
Each node consumes W watts. Over one 600-second block, its electricity cost is:
With free entry and exit, competition compresses the mining margin toward zero. Electricity is the φ-share of marginal cost (the rest is wear, connectivity, attention):
Hardware wall power W appears in N*, but cancels in E and VR. When efficiency improves, cost per node falls, the margin opens, and entry closes it — more nodes fit inside the same energy envelope. A 5% efficiency gain becomes 5% more nodes and an unchanged anchor. Note the direction precisely: N* depends inversely on W; it is the network energy and the anchor that are invariant to it.
Why this replaces programmed work escalation. A protocol-side counterweight to efficiency drift would need to observe watts — which no permissionless, Sybil-resistant protocol can do (splitting one miner into two identities must change nothing, so per-miner metrics are unobservable by design). The market observes watts perfectly, through every miner's electricity bill, and prices them into the entry decision. The counterweight exists; it just lives outside the protocol.
| Scenario | P | p_elec | N* | kWh/block | VR |
|---|---|---|---|---|---|
| Bear | $0.10 | $0.10 | 8,000 | 100 | 1.0 |
| Base | $1.00 | $0.10 | 80,000 | 1,000 | 10.0 |
| Bull | $5.00 | $0.10 | 400,000 | 5,000 | 50.0 |
| Expensive power | $1.00 | $0.30 | 26,667 | 333 | 3.33 |
Realized values sit below this bound by the factor φ (electricity share of marginal cost, ~0.85 for commodity hardware with near-zero capex — far higher than ASIC mining, where hardware amortization consumes a large share of revenue).
Bitcoin obeys the same identity — production cost tracks mining revenue (the basis of the Cambridge CCAF consumption estimates). The difference is adjustment speed: ASIC entry requires months of fabrication lead time and heavy capital, so Bitcoin's equilibrium tracks with quarters of lag. eWatts entry is switching on a machine that already exists — minutes, not months, and no capital purchase. Free entry in eWatts is closer to the theoretical ideal than in any existing PoW. The commodity-hardware choice is not a compromise on the anchor; it is what makes the anchor's arbitrage nearly frictionless.
Miners burning eWatts for extra reward weight would also absorb efficiency margins — but into supply destruction, concentrated among incumbents who already hold eWatts, adding a proof-of-stake-like entry barrier. Free entry absorbs the same margin into more independent nodes. For a protocol built on accessibility and neutrality, distribution strictly dominates destruction. The auction the protocol needs is the one it already has.
VR = φ·P / p_elec means the energy embodied in each eWatt converges to its market price expressed in electricity terms. Read the causality in the correct direction:
The anchor is descriptive, not prescriptive. Production cost follows price — it does not prop price up. If P falls, miners exit, the network shrinks, and the cost floor moves down with it. What the protocol guarantees is the relationship: at all times, acquiring an eWatt by mining costs real, measurable energy, and competition keeps that cost pinned to what the market says an eWatt is worth. "Energy-backed" means the backing is continuously re-priced by physics and free entry — the same way gold's production cost tracks gold's price across cycles, not a fixed redemption promise.
Consensus-side, issuance stays near-constant (~5.26M eWatts/year) and percentage inflation declines as 1/t — 10% in year 10, 5% in year 20 — without a halving calendar or any programmed decay.
The free-entry equilibrium holds under specific assumptions about miner behavior, hardware symmetry, and market access. Each assumption can be challenged, with different probabilities and time horizons.
A state or company with access to zero- or negative-marginal-cost electricity (stranded renewables, curtailed hydro, flared gas that would otherwise be wasted) can mine with a cost base far below market electricity prices. This distorts the anchor because the marginal miner’s pelec diverges from the market rate, and the identity VR = φ·P/pelec is only as tight as the marginal miner’s true cost. This risk is shared with every proof-of-work system that exists, including Bitcoin, and is partially self-limiting: subsidized miners who sell eWatts on the open market push P down, reducing the incentive to enter further. But a determined state actor could sustain below-cost mining for years as a strategic expense.
If 2–3 mining pools control >80% of effective bandwidth, they can coordinate to slow or block entry by new miners, throttle the hist_avg window to manipulate emission, or suppress commitment efficiency to degrade the anchor’s signal. The free-entry equilibrium requires decentralized response to margin signals; pool concentration introduces a coordination layer that can delay or sever that response. Mitigation is structural: low barrier to entry (commodity DRAM) makes pool exit cheap, but the early network is the most vulnerable to this dynamic.
The DAG cache is ~64 MB for the initial 8 GB DAG (DAG_size / 128). A custom ASIC could store this cache in on-chip SRAM (Apple M3 Ultra already has 192 MB of SRAM cache) and compute DAG elements on demand via a dedicated SHA512 pipeline, potentially eliminating the DRAM bandwidth bottleneck. Each DAG element requires chained SHA512 + FNV mixing through the cache, making on-the-fly computation slower than a DRAM read — but a sufficiently aggressive pipeline could close or invert that gap. Three defenses limit this vector: (1) the DAG grows 512 MB/year, forcing corresponding cache growth; (2) SRAM at that scale (>64 MB) is expensive, ~100× the cost per bit of DRAM; (3) even a successful ASIC would gain at most 2–3× efficiency over commodity DDR5, a far cry from the 105× advantage Bitcoin ASICs hold over CPUs. The protocol does not depend on impossibility, only on economic irrationality.
Samsung HBM-PIM and SK Hynix PIM integrate simple logic inside DRAM banks, reducing data movement overhead. For eWatts mining, this could give a ~2× efficiency improvement per watt by eliminating bus traversal. PIM does not break the anchor: the memory is still DRAM, available to any buyer, and the efficiency gain is proportional across all miners. It is the closest analogue to a “DRAM ASIC” that exists today, and its impact is to shift the equilibrium node count (N* changes since effective W per meaningful access falls) without shifting the anchor (E and VR remain invariant to W by construction).
With few nodes, the response to a margin signal (P increase, efficiency gain, power cost drop) can take days or weeks — the time for new miners to learn about eWatts, configure hardware, and commit bandwidth. During that lag, returns for existing miners are higher than the equilibrium would predict, and the anchor can drift above or below the identity. This is the single most important thing the testnet can measure: how fast does entry respond to margin? The lag is shorter than any ASIC chain (no fabrication, no capital outlay) but cannot be eliminated in a permissionless system. Modeling the lag as a first-order system with a time constant τ (hours to days) is the recommended framework for testnet data analysis.
Entry lag is the real risk — not efficiency drift. The equilibrium holds only when participation responds to margins. In a small, illiquid bootstrap network, the response has delay; during that delay, efficiency gains erode returns for incumbents and the anchor drifts below the identity. This risk is real, measurable, and characteristic of every young PoW network — eWatts minimizes it relative to ASIC chains (no fabrication lag, no capital barrier) but cannot eliminate it. The testnet hypothesis: does entry respond to margin? That is the single most informative thing the testnet can measure.
P is endogenous in the full system. This page treats price as an input; in reality price emerges from supply and demand for the token itself. The scenarios here are conditional statements ("if the market sustains P, then…"), not predictions.
Electricity heterogeneity. Miners face costs from ~$0.03 (hydro, stranded energy) to ~$0.30/kWh (European residential). The marginal miner — the most expensive one still profitable — sets the equilibrium; cheaper miners earn Ricardian rents. The p_elec in the formula is the marginal miner's, which shifts as the network composition changes.
Long-term security budget. With near-constant issuance, per-block rewards dilute across a growing miner set. Transaction fees must eventually carry miner compensation, as on Bitcoin. Whether fees will suffice cannot be predicted in advance.
The equilibrium is a bound, not a law. Frictions (φ < 1, lag, lumpy participation, price volatility) keep the realized network inside the identity bound. Bitcoin's fifteen-year record shows production cost tracking price through cycles with lag measured in months — the best empirical calibration available for how tightly such identities bind in practice.
Sources: eWatts v0.5 source code (constants.rs, reward.rs, vr.rs, proof.rs) · Cambridge CCAF (Bitcoin electricity consumption methodology) · US EIA · World Gold Council · Federal Reserve (M2)