Edge OS — T1 Open
$2,000 / GPU / yr
Measured boot + signed runtime (2-GPU node ~$4k · dense ~$16k)
Telcos / ISPs — batch & non-sensitive lanesEdge / Cloud OS turns GPU nodes on business premises into an attested, placeable, metered inference fabric. The capital model is deliberate: telcos and ISPs own and deploy the nodes for strategic value — fiber retention, enterprise SLAs, jurisdiction-pinned sovereignty deals — while hosts lend the premises for zero capex and real earnings. nventr sells the software, not the hardware.
The fee scales with the hardware class. Attestation tier sets the per-GPU rate.
Measured boot + signed runtime (2-GPU node ~$4k · dense ~$16k)
Telcos / ISPs — batch & non-sensitive lanesSEV-SNP + GPU CC + enclave TLS + dual-VLAN (2-GPU ~$6k · dense ~$24k) — attestation the hyperscaler stack doesn’t ship
Telcos / ISPs — default business nodesT2 + jurisdiction pin + audit adder + higher SLA weight
Sovereign / regulated dealsPrivate mode · trial included with hardware
Homeowners (post-trial)Don’t buy the metal — host it. Lend a rack or closet, keep zero capital at risk, and your ISP’s node earns for you: a guaranteed availability floor plus a revenue share (8–15% of node gross, 12% modeled), stacked on free symmetric fiber + static IP from the telco partner — and an optional thermal kit turns waste heat into a utility offset. NV-HOME owners can opt in on the same terms, privacy first.
Annual revenue-share dollars per node (12% of node gross). Each cell: commodity $0.25/M · sovereignty $0.40/M.
| Node config | Light · 30% util | Baseline · 50% util | High · 80% util | + Thermal kit | Total @ baseline* |
|---|---|---|---|---|---|
| NV-BIZ-MID 2× RTX PRO 6000-class | $480 · $768 | $800 · $1,280 | $1,280 · $2,048 | +$850/yr | ~$14,000 / yr |
| NV-BIZ-MID 2× H200 NVL-class | $2,280 · $3,648 | $3,800 · $6,080 | $6,080 · $9,728 | +$850/yr | ~$17,000 / yr |
| NV-BIZ-DENSE 8× MI300X-class | $8,880 · $14,208 | $14,800 · $23,680 | $23,680 · $37,888 | +$2,600/yr | ~$28,000 / yr |
*Total host position at 50% utilization, commodity lane: floor + fiber value + share (thermal kit additional).
NV-HOME is a privacy appliance first — income is optional and best-effort, never required. With contribution mode opted in (hard off-switch, private data never enters the pool):
Nobody should buy these nodes for token yield — the numbers below make that plain, and they are published on purpose. Owning the metal is for ISPs and operators underwriting strategic value: a hosted node retains a business fiber account (worth $6–18k/yr on its own), creates enterprise SLA and jurisdiction-pinned products hyperscalers don’t offer, and the token margin merely offsets opex so the strategic capacity runs near self-funding. For transparency, the token economics of ownership (before Edge OS license; 50% utilization):
| Build | Margin/yr @ $0.25/M | Margin/yr @ $0.40/M | Payback (before OS fee) |
|---|---|---|---|
| 2× RTX PRO 6000 · $38k capex | –$13,600 / yr | Modest positive | Never → long |
| 2× H200 NVL · $72k capex | –$5,600 / yr | +$8,000 – 12,000 / yr | Never → ~6 yr |
| 8× MI300X · $212k capex | +$21,600 / yr | +$45,000 – 55,000 / yr | ~10 yr → 4–5 yr |
Only dense high-bandwidth builds clear positive operator margin at commodity token prices — mid-tier configs need the sovereignty premium or higher utilization. Revenue is bandwidth-bound, not capacity-bound.
Buying and operating the node yourself? The Edge OS license is the biggest line between gross token margin and your net. Fee take and ROI by hardware tier and use case, at the per-GPU price book (50% utilization) — hosts skip this entirely: on the preferred path the ISP/operator pays the OS license, and host floor + share are net.
| Build | Open · T1 $2k/GPU · $0.25/M | Confidential · T2 $3k/GPU · $0.40/M | Sovereign · T3 $3.5k/GPU + $5k · $0.40/M | Breakeven OS fee |
|---|---|---|---|---|
| 2× RTX PRO 6000 · $38k capex | –$17,600 net token margin negative pre-fee · Not viable | –$4,000 net fee > margin · Not viable | –$10,000 net fee > margin · Not viable | ~$2k (sovereignty) · none (commodity) |
| 2× H200 NVL · $72k capex | –$9,600 net token margin negative pre-fee · Not viable | +$4,000 net 60% of gross · 5.6%/yr · ~18 yr payback | –$2,000 net fee > margin · Not viable | ~$10k (sovereignty) · none (commodity) |
| 8× MI300X dense · $212k capex | +$5,600 net 74% of gross · 2.6%/yr · ~38 yr payback | +$26,000 net 48% of gross · 12.3%/yr · ~8 yr payback | +$17,000 net 66% of gross · 8.0%/yr · ~12.5 yr payback | ~$21.6k (commodity) · ~$50k (sovereignty) |
| NV-HOME (opt-in contribution) | Edge OS runtime $15–40/yr after trial vs $50–200/yr opt-in contribution → fee takes 8–80% of contribution, net $10–185/yr (+~$85 thermal). A privacy appliance, not an income product — the OS fee is buying private local AI, not pool access. | |||
Per-GPU pricing scales the fee with the hardware class automatically: a dense node pays ~$24k at T2 — close to the old flat band — while a 2-GPU node pays ~$6k instead of $20–30k, turning mid-tier owner-operators from impossible to merely thin. What pricing cannot fix: mid-tier token economics are weak at commodity prices regardless of fee (D2’s sovereignty premium is the real lever), and T3’s adder only pays for itself when jurisdiction-pinned deals carry premium $/M or higher placement weight. A caution on the ROI column: at the plan’s stated 4-year hardware life, even the viable builds do not recover capex at 50% utilization (8–18 yr simple payback vs a 4-yr depreciation clock). Owner-operating pencils only at high utilization (dense sovereignty at 80% util pays back in ~4 yr), a longer effective life (5–7 yr inference duty is plausible), or meaningful resale residual — which is exactly why the plan prefers the host path, where the ISP carries the hardware-life risk.
Illustrative and directional (pre-pilot), derived from the plan’s stated assumptions: 12% host share of gross (8–15% band), $1,200/yr availability floor, $0.12/kWh power (reimbursed), 4-year hardware life, fiber value modeled at the $12k midpoint of $6–18k/yr. Revenue share scaled linearly with utilization and token price from the 50% / $0.25-per-M anchors. Actual earnings depend on placement, demand, SLA performance, and contract terms.