orbital-zec

How the model works

• Orbital $/kWh = (launch $/kg ÷ specific power W/kg + non-compute hardware $/W) ÷ (lifetime hours × 0.99 illumination). Compute silicon is excluded on both sides, this is the cost of the watt, not the chip.
• Launch cost declines at a fixed %/yr from the 2026 price toward a floor. Specific power and $/W of bus hardware ramp linearly from their 2026 to 2036 values.
• Earth $/kWh compounds in two phases: a grid squeeze to 2030 (default 12%/yr: wholesale +11% in 2026, PJM +41%, data-center demand +22%/yr against single-digit grid additions), then a flatter phase (default 2%/yr) as behind-the-meter gas, solar-plus-storage PPAs and hyperscaler-funded interconnects catch up and falling LCOE caps the marginal watt. The 2026 default is an all-in datacenter power price; miners chasing stranded power ride the squeeze less than this. On top of the trend the price swings with the AI cycle (default half the inference amplitude, so ±17.5%): booms are demand spikes, busts strand capacity and cheapen power.
• ZEC price follows two halving cycles: log-linear from the 2026 price to a 2028 peak, retrace to a 2030 trough (% of peak), rise to a 2033 peak after the Nov 2032 halving, retrace to a 2035 trough, then compound growth. That adoption path is multiplied by (earth $/kWh ÷ 2026 earth $/kWh)^β. β is the elasticity of ZEC to the cost of a kWh, the cost-of-production floor argument. β=1 means a 50% rise in power price lifts ZEC 50% on top of adoption.
• NSM (NU7, Nov 2026): ZIP 235 removes 60% of transaction fees from circulation into a reserve; ZIP 233 lets anyone remove more. Reissuance of the reserve through block rewards starts Feb 2031 (96.6% coinholder vote), modelled as a fixed %/yr of the balance (default 16%, roughly a halving-equivalent decay). ZIP 234 issuance smoothing was rejected 98.9%, so halvings remain. Fee volume today is tens of ZEC/day, ~1% of subsidy, so NSM only matters if usage grows by orders of magnitude or after the 2036 halving shrinks subsidy to ~180 ZEC/day to miners. Fees are modelled in dollars (usage × fee-per-action, which re-denominates as ZEC rises) and converted to ZEC at each year's price; the reserve is held in ZEC.
• Security budget = ZEC issued that year × ZEC price. Issuance is 1,800 ZEC/day today, halving to 900 in Nov 2028, 450 in Nov 2032 and 225 in Nov 2036. NU7's 25-second blocks divide the per-block subsidy by three (ZIP 218), issuance per year is unchanged. NU7's faster blocks don't change issuance.
• Earth hashrate starts from today's network (~30 GSol/s, ~100 MW at Z15 Pro efficiency) and grows at a hardware-supply rate, because today miners earn ~7–14× their power cost and the constraint is ASIC supply, not electricity. New earth capex stops when the power bill exceeds 70% of revenue, and running miners switch off when it exceeds 85% (hosting, opex and debt sit in the gap, nobody runs to the raw electricity line). Both thresholds are shared with orbital hashrate, so earth miners are squeezed out as orbit grows. Miner budget is 80% of issuance (20% to dev fund / lockbox).
• Mining revenue per kWh = miner budget ÷ total network draw including orbital, so orbital hashrate dilutes its own earnings as it grows. The equilibrium floor (earth $/kWh ÷ power share) is where it would settle with unlimited hardware. Inference revenue per kWh defaults to $1.80, between CoreWeave's realized ~$1.15/kWh (2026 exit run-rate $18.5–19.5B on >1.85 GW active) and Blackwell at contract rates and real utilization (~$1.5–2.5). SpaceX's own sheet uses $40B/GW/yr = $4.57/kWh, which is Blackwell list price at 100% utilization; set that for the bull case. Decay is fleet-level: same-chip prices fall ~30%/yr but the fleet refreshes to chips that charge more per watt, and 2026 contract prices rose, so 10%/yr is the central case. On top of that a boom/bust cycle: revenue per kWh swings ±amplitude on a sine with the given period and peak year. Default peak 2027, period 5 yr: AI leads crypto by about a year, so AI troughs (2029.5, 2034.5) roughly coincide with ZEC troughs (2030, 2035), both being liquidity-driven. Set peak 2026 for the pure capital-rotation view where AI busts land on crypto peaks. Idle fraction swings the opposite way by the idle multiplier × amplitude, because a bust is when capacity goes unsold. The base idle fraction also declines each year (default 5% in 2026 falling 15%/yr: ~3% in 2029, ~1.2% in 2035, ~0.5% in 2040) as scheduling, batching and spot markets absorb slack; the cycle swing is applied on top of that year's base.
• Mineable capacity: only satellites launched from the Terafab start year, times the share of launches carrying Terafab silicon. Nvidia HBM-based satellites have no idle SRAM to run Equihash on. SRAM-resident inference chips (Groq/Cerebras/Dojo class) hold the 144 MB Equihash working set on-die at ASIC-class bandwidth, so the efficiency multiple defaults to 1×, parity with a Bitmain Z15 per watt, with no extra mass. Raise it only if you believe in dedicated cores on top. Orbital efficiency gain per vintage is absolute and applies only to satellites launched that year. The earth fleet improves too (default 15%/yr), and earth-equivalent watts are measured against the earth fleet of that year, so an orbital vintage loses relative ground every year after launch. The fleet's earth-equivalent hashrate is a cohort-weighted sum divided by current earth efficiency. Net relative drift at defaults is ~13%/yr for new vintages, negative for old ones.
• Orbital hashing behaviour: orbit hashes only enough to hold its target share. At the default 100% it hashes just past the earth switch-off point (miner budget × switch-off share ÷ earth $/kWh), which keeps the whole budget and never lets earth back in, at a fraction of available capacity; this pins revenue per orbital kWh to roughly earth electricity price × fleet efficiency. A target below 100% models an operator that deliberately leaves hashrate to earth miners for optics, at the cost of that share of the budget. It also never hashes below its marginal-cost floor (default $0.02/kWh: wear, scheduling, the spot-inference alternative). Earth miners respond to last year's orbital hashing, a one-year lag. The dashed line in the hashrate chart is what orbit could hash, the solid one is what it does.
• Second orbital bloc: a Chinese constellation modelled as a fraction of SpaceX's capacity on the same curve, lagged by a few years. It mines from its own start year with dedicated cores on every satellite it launches from then on, no Terafab-class silicon required, since a state actor accepts the mass penalty. It hashes all of its idle silicon, because a state actor mines for control rather than profit, so it shows none of SpaceX's restraint. In compete mode SpaceX treats it as a rival when sizing its own hashing and the headline tile reports the largest single actor; in collude mode the two act as one. Two operators is better for Zcash than one: neither alone may reach 51%.
• Blue Origin: a third orbital bloc, commercial, modelled like China in capacity terms (fraction of SpaceX, lagged) but behaving as a price-taker: it hashes all its idle silicon whenever that clears its marginal cost, with no monopolist restraint because it isn't the monopolist. It never colludes in the model.
• Orbital capacity is log-interpolated from ~0 in 2026 through the end-2028, end-2030, end-2036 and end-2040 inputs. The 2040 default of 1 TW is SpaceX's stated long-run target. Specific power and bus $/W hold at their 2036 values after 2036. The 2028 default (2.8 GW) is SpaceX's own investor projection: 40 MW in Q3'27, 135 Starship launches for orbital compute in 2028 at 10–25 MW each. 2030 and 2036 are extrapolations of that launch cadence, not SpaceX numbers. The same projection carries $40B/yr revenue per GW ($4.57/kWh nameplate, flat through 2028) and $30B/GW all-in capex. Idle fraction is the slice that could hash with no inference trade-off, decode chips are the natural host since they run bandwidth-bound and below TDP.