Voltage Park
US neocloud selling on-demand and reserved bare-metal NVIDIA HGX H100 clusters from $1.99/hr; acquired TensorDock (2025) and merged with Lightning AI (Jan 2026).
A genuinely live, actively developed and CertiK-audited decentralized-inference network with transparent on-chain pricing, but an opaque legal entity, tiny on-chain activity and token liquidity only via wrapped WGNK on Uniswap.
site reachable · checked 11 h ago · daily check
by the bestcloudmining editorial desk · curated review · prose updated 2026-10-02 · facts as of 2026-10-02 · how we verify
Gonka is a decentralized AI-inference network (GONKA token) where users pay per token to run LLM inference on distributed GPU providers - not a GPU rental service [1]. Bitfury invested 50 million dollars in December 2025 [3]. The project self-describes as "not a company" [3], which is the most transparent thing about its corporate structure. The docs were freshly stamped "epoch 412, updated 2026-10-02" on our check date - actively maintained [4].
LLM inference paid per token on-chain - users select from 5 bookable models with a pricing floor of 0.001 GONKA per 1M tokens, dynamically priced by brokers [2]. There is no GPU-hour rental, no VM deployment, and no storage product [2]. You do not own hardware; you consume inference through the network's API [2]. The per-token model means costs scale with usage, which is efficient for burst workloads but unpredictable for sustained inference [2]. The on-chain pricing API is publicly readable without an account [2].
No operating company is disclosed [3]. No KYC policy is published [2]. No SLA is published [2]. Payment runs in GONKA on-chain [2]. The terms are minimal - the product is a smart-contract-mediated inference network, not a hosting service with consumer protections [2]. There is no refund policy, no uptime guarantee, and no dispute-resolution mechanism published anywhere [2].
The on-chain pricing API is publicly readable without an account [2]. GitHub was pushed today [4]. What stays unverifiable: the claimed 10,000 H100-equivalents (unaudited), actual inference quality, and per-provider reliability [3]. The claimed scale is the most aggressive in the decentralized-compute space [3]. No independent audit of the network's infrastructure exists [3].
Bitfury's 50-million-dollar investment (December 2025) is press-documented [3]. The token trades only on Uniswap - no centralized exchange has listed GONKA [3]. Press claims of scale ("over 10,000 H100-equivalents") are unaudited [3]. No fraud or exit events surfaced [3]; community discussion is focused on staking and GPU-provider earnings [5].
Per-token LLM inference: network floor 0.001 GONKA per 1M tokens, dynamically priced by brokers [2]. No GPU-hour rate exists [2]. No KYC [2]. No minimum purchase [2]. Payment runs exclusively in GONKA on-chain [2].
Gonka is a real inference network with on-chain pricing and Bitfury backing [3] - but it is not a compute provider in any rental sense [2]. For users who want LLM inference on a decentralized network, it is a verifiable option [2]; for GPU buyers, look elsewhere [2]. The Bitfury investment is the strongest credibility signal [4]; the Uniswap-only listing and the absent legal entity are the two structural limitations that prevent comparison with conventional clouds [3]. If a centralized listing materialises and an operating company is named, the risk profile improves significantly [3]. Until then, this is a DePIN experiment with real infrastructure and institutional backing but no consumer protections [4].
For teams evaluating decentralized inference networks, Gonka is worth monitoring but not yet a production option [3]. The dynamic pricing mechanism is structurally interesting - it lets supply and demand determine the per-token rate in real time, unlike fixed-rate clouds where pricing is set by the operator [2]. The absence of a legal entity means that if the network experiences a disruption or a provider disappears, there is no contractual counterparty to pursue [3]. This is a known trade-off in the DePIN space [5].
| Documented price | see provider site |
|---|---|
| Payouts | per epoch (rewards vest and release daily) |
| Notes | Inference is priced per token on-chain and dynamically per model (EIP-1559-style stability zone 40-60% utilisation; floor 1 nicoin/token; max +-2% per block). Snapshot 2026-09-24: 0.001 GNK per 1M tokens for all five models, i.e. the on-chain floor. This is the protocol price, not the buyer price: community brokers set their own retail rates; one community post (2026-08-29) quotes ~$0.00123746 per 1M tokens. There is no per-GPU-hour rental price; hosts supply hardware and earn rewards. |
Run the long math before paying: cloud mining calculator →
Built only from what is documented on this page as of 2026-10-02 - anything not documented stays a gap on purpose.
Everything happens on gonka.ai - we count the click, nothing else changes for you.
Accepted: GNK, WGNK, USDT. Check the network and the deposit address inside your account, never from an email.
Proof is documented as partial - Verifiable inference by design but probabilistic, not hardware-attested: Transformer-based Proof-of-Work ('Sprint') measures real model compute. Compare the first payout with the pool or wallet before scaling up.
Payout frequency: per epoch (rewards vest and release daily).
Then run the math: mining calculator · contract vs. HODL · all GPU compute providers.
Short answer: Based on our checks as of 2026-10-02: Gonka (Gonka.ai) scores 5.8/10 in our weighted review. We logged no red flags. User reports and operator background are documented with dated sources in the meta-review below.
| Status | active |
|---|---|
| Operator | Gonka (decentralized protocol; incubated by Product Science, Inc.) (low) |
| Founded | 2024 |
| Proof | partial |
| KYC | not documented |
No red flags logged as of 2026-10-02.
A real, operating provider can still lose you money: with a contract you own no hardware, revenue follows coin price and difficulty, and most published contracts are priced so the seller comes out ahead. "Legit" means the operator exists and pays what its terms promise - not that the deal is profitable. 23 dated sources back this page; they are listed under sources.
Same category, scored on the same criteria - so the comparison stays honest. Compare only in the same unit, and read the minimum, KYC and proof columns before switching.
US neocloud selling on-demand and reserved bare-metal NVIDIA HGX H100 clusters from $1.99/hr; acquired TensorDock (2025) and merged with Lightning AI (Jan 2026).
Distributed consumer-GPU cloud with published per-second prices and measured host earn rates; no token, fiat only.
Established US GPU cloud: on-demand H100/B200 instances, 1-Click Clusters and Superclusters, billed per minute with no ingress/egress charges.
Open permissionless marketplace for containerised GPU/CPU compute; providers bid, tenants pay per block.
Ready to try Gonka (Gonka.ai)? Start small, verify one payout, and only scale up once the hashrate or compute actually matches what you paid for.
Based on our checks as of 2026-10-02: Gonka (Gonka.ai) scores 5.8/10 in our weighted review. We logged no red flags. User reports and operator background are documented with dated sources in the meta-review below.
No comments yet - be the first. Posting runs a short proof-of-work in your browser (anti-spam), no account needed.