Growth · DePIN GPU
Aethir ATH
MC ~$80M
FDV ~$170M
ARR TTM $147M
Q3 2025 rev $39.8M
Composite46.4 not ranked · thin data
Buyback yield—
P/S— sector 177.9×
Below 1y peak89% 8.8× to reclaim
Float48% FDV $189.4M
Moat40 judgement
Innovation60 judgement
Moat 40. GPU aggregation is close to a commodity with several well-funded competitors. Competes primarily on price and supply availability.
Innovation 60. GPU aggregation for AI workloads is real infrastructure, but it is a crowded design with several well-funded competitors and little architectural differentiation.
Innovation 60. GPU aggregation for AI workloads is real infrastructure, but it is a crowded design with several well-funded competitors and little architectural differentiation.
Measured from live data, except Moat and Innovation which are hand-set judgements · snapshot 2026-08-23 · full model →
What it does
Aethir is a decentralized GPU cloud running enterprise NVIDIA H100 hardware for AI training / inference and cloud gaming. It aggregates enterprise-grade capacity and provisions it to AI labs, gaming studios, and inference customers.
Fundamentals
- $147M ARR TTM — the highest actual revenue of any GPU DePIN.
- $39.8M revenue in Q3 2025 alone.
- 440K+ GPU containers across 94 countries; 150+ paying enterprise clients.
- Dwarfs io.net's ~$12.5M in the same category.
Tokenomics — the caveat
- 55%+ of supply still unlocks through 2028.
- 471M ATH unlocking Aug 12, 2026 (~3.6% of MC in one event).
- Monthly unlock cadence continues through 2028 — real dilution headwind.
- Buybacks are funded from revenue; skepticism exists about how much of the "revenue" is cloud-gaming vs AI inference.
Catalysts (6–12mo)
- $344M ATH-staked Strategic Compute Reserve deployment through 2026.
- Institutional AI client onboarding at scale.
Risks
- Emission dilution actively fights price appreciation.
- Revenue quality debate — cloud gaming vs AI inference mix.
- Competition from io.net, Akash, and centralized clouds on unit economics.
Strong Buy (asymmetric). At <1× revenue on TTM basis, priced for failure. If buybacks scale with revenue and unlocks are absorbed, this is a 3–5× asymmetric setup. Size accordingly — layer entries around unlock cliffs rather than front-loading.
How it compares to its peers
| AI / Compute | Mkt cap | Revenue/yr | Growth/yr | Bull y3 rev | P/S | Fwd P/S y3 | Buyback |
|---|---|---|---|---|---|---|---|
| Bittensorpeer | $2.1B | — | — | — | — | — | — |
| Renderpeer | $733.3M | $2.2M | -32% | $1.7M | 332.7× | 678.4× | 0.3% |
| Virtualspeer | $439.3M | $19.0M | -39% | $7.7M | 23.1× | 57.2× | 0.0% |
| Grass | $195.9M | — | — | — | — | — | — |
| Akashpeer | $157.6M | $0 | 40% | $0 | — | — | — |
| Aethir | $90.8M | — | — | — | — | — | — |
Peer revenue growth runs at a median of -32% a year, with 2 of 3 shrinking — so this multiple sits inside a contracting sector.