Market cap $2.20 T −3.96 % over 24 hours24h volume $73.16 BBTC dominance 58.1 %ETH 10.0 %Fear & Greed 10 Extreme fearCoins 21,490

Project · AI Infrastructure

Akash Network (AKT): how the GPU cloud marketplace works and its Solana vote

Active Risk 4/5

Metrics

Token price$0.548418 +5.43 %
Market cap164.55 M USD

About the project

Akash Network is a marketplace for cloud computing that runs on its own blockchain: people who need servers or graphics cards post a request, independent providers bid to host it, and the winner is paid from an on-chain escrow account. Since 26 October 2026 the AKT stakers who govern it have been voting on whether to leave that chain. Proposal 346 asks them to move the AKT token, and later the whole marketplace, to Solana. Voting closes on 2 November 2026 at 15:02 UTC.

At 19:27 UTC on 27 October, 51.90 million AKT had voted yes and 0.96 million no, a 98.19% yes share among votes that took a side (97.97% of all votes, abstentions included). Turnout reached 59.2% of staked AKT, well above the 33.4% quorum. Nothing is final before the deadline, and our report on the Akash migration vote follows the count.

The vote lands on a network with a long record and a small turnover. Akash has run since September 2020 and has rented out GPUs since 2023, yet its tenants spent about $5,760 in the 24 hours to that same reading, according to its own stats API. Since March 2026 the token also works differently: tenants burn AKT to pay for compute.

What is Akash Network: a cloud market where providers bid for work

Overclock Labs, the company that created Akash, built it around a reverse auction instead of a price list. The buyer names the most it will pay and sellers compete under that ceiling. Akash calls the buyers tenants and the sellers providers, and its documentation describes the flow in four steps.

  1. The tenant writes a deployment file in SDL, the Stack Definition Language: a YAML file that names the container image, the CPU, memory and storage it needs, the ports to open and the price it is willing to pay.
  2. Providers, data centres and smaller operators running Akash’s provider software on Kubernetes clusters, answer with bids.
  3. The tenant picks one bid and creates a lease, which locks in the price and the resources.
  4. The provider pulls the image, starts the app and is paid automatically from the tenant’s escrow account every block, roughly every six seconds, until the lease is closed.

Escrow is now funded in ACT, a dollar-pegged compute credit explained further down; the docs say AKT is used instead while a circuit breaker is in force. There are two ways in. The managed Akash Console lets people sign in with an email or a social account, bills a payment card and funds each deployment from the account’s credits; its free trial went from $10 to $100 in September 2025. Console Air, released on 5 May 2026, is an open-source copy that a user hosts alone and connects to a Keplr wallet. “Console Air is the new home for self-custody,” the team wrote at launch.

An auction gavel above a signed contract, a processor chip and stacked coins
Providers bid under the tenant's price ceiling, and the accepted bid becomes a lease paid from escrow every block.

Under the marketplace runs akashnet-2, a proof-of-stake chain built with the Cosmos SDK. Its own modules keep the deployments, the market, provider records, escrow, certificates, provider audits and the protocol fee, called the take rate, as listed in Akash’s migration specification AEP-79. At 19:27 UTC on 27 October the chain stood at block 29,256,331 and had processed 43.6 million transactions.

GPUs, AkashML and confidential compute

Akash started with ordinary CPU servers. GPUs arrived with Mainnet 6, which The Block reported on 31 August 2023: providers could list Nvidia H100 and A100 chips as well as consumer cards, tested beforehand by 1,300 participants, and the team called the result “Supercloud”. Proposal 228, opened on chain the same day, made axlUSDC, the USDC bridged by Axelar, the first stable currency accepted for payment.

In November 2025 the project added AkashML, a managed inference service with an OpenAI-compatible interface. A developer sends prompts to hosted models (Llama 3.3 70B, DeepSeek V3 and Qwen3-30B-A3B at launch) and never touches a lease, while AkashML runs the models on Akash providers. Akash says the service costs 70% to 85% less than the big clouds, without publishing the benchmark, and its Q1 2026 report put AkashML at 1.7 billion tokens a day as a provider on OpenRouter. Neither figure can be checked from outside.

Confidential compute went live on 28 July 2026. A tenant adds one SDL setting, params.tee, to ask for a trusted execution environment: an Intel TDX or AMD SEV-SNP processor, optionally an Nvidia GPU in confidential mode, with the workload sealed in a Kata Containers micro-VM. The launch post admitted that provider capacity was early-stage, so some requests may draw no bids, and that remote attestation tooling was still being written. It advised users who need a verified environment to wait for it.

Akash Network stats: 662 active leases and about $5,760 spent a day

Akash publishes usage through a public API, the one behind stats.akash.network. These are the project’s own numbers and nobody audits them, but DefiLlama’s fee tracker for the Akash chain shows an all-time total of $6.47 million, within 0.3% of Akash’s figure, which suggests both read the same chain data.

Metric (Akash stats API, 27 October 2026, 19:27 UTC)Value
Active leases662 (652 a day earlier)
New leases in the previous 24 hours1,219
Leases created since launch631,600
Spend in the previous 24 hoursabout $5,760 (about $5,842 a day earlier)
Spend since launchabout $6.49 million
GPUs in active leases146 (153 a day earlier)
GPUs reported by providers373 in total, 279 marked active, 88 available
Active providers57
Memory and storage in active leases20.3 TB of RAM, 42.2 TB of storage

Start with scale. About $5,760 a day comes to roughly $2.1 million a year if nothing changed, a snapshot and not a forecast, and DefiLlama counts $2.03 million of fees over the twelve months to 27 October 2026. AkashML’s own launch page compares itself with reserved AWS SageMaker setups that it says can cost more than $50,000 a month. On that yardstick, everything tenants spend on Akash in a month equals three or four such customers.

The GPU figures need care. The API counts 146 GPUs in active leases and 373 reported by providers, 279 of them marked active, without saying what “active” means in the second count. Akash’s 2025 review claimed more than 1,000 GPUs deployed during the year and 60% average utilisation. Coindaro could not reproduce either number from live data: on 27 October, leased GPUs equalled 39% of the units providers reported.

The same review gave year-on-year figures. Tenants spent $3.15 million in 2025 against $1.38 million in 2024, up 128%, while new leases barely moved, at 126,025 against 123,342. The average fee per lease more than doubled, from $11.23 to $25.03, so the growth came from bigger jobs, not more of them. In April 2026 the Q1 report said total compute spend had passed $5 million, with no breakdown that matches the API.

AKT tokenomics: burn-mint equilibrium since March 2026

AKT is staked to secure the chain and carries the votes in governance. Its role as payment changed on 23 March 2026 at block 26,063,777, about 14:00 UTC, when burn-mint equilibrium (BME) went live with Mainnet 17. Proposal 318 had approved the upgrade with 61.8 million AKT for and 166 AKT against.

Under BME a tenant no longer hands AKT straight to a provider. It burns AKT at the oracle price to mint ACT, the Akash Compute Token, a credit pegged to the US dollar that cannot be transferred and only pays for compute. When a lease consumes ACT, the chain burns it and mints fresh AKT for the provider at the price of that moment. Akash’s explainer of 18 March 2026 gave an example: with AKT at $1.14, about 877 AKT mints 1,000 ACT. If AKT then rose to $1.50, settling the same $1,000 would mint only about 667 AKT, and the other 210 AKT would be gone for good. A falling AKT price works the other way and adds to supply.

The switch shows in the data. In the 24 hours to 19:27 UTC on 27 October, the stats API recorded all spending in ACT and none paid directly in AKT.

A circuit breaker watches the peg by comparing the value of the AKT held by the BME module with the ACT outstanding. At 0.95 the module enters a warning state. At 0.90, the level its on-chain parameters call the halt threshold, new ACT minting is throttled while payments to providers continue. Prices come from an Osmosis time-weighted average over 30-minute windows and from external feeds, combined through a median. On 27 October the module read healthy, with a collateral ratio of 1.79, a 0.25% fee on minting ACT and none on settlement.

BME did not switch off inflation, although the explainer said it “replaces traditional inflation”. The chain’s mint module still issues new AKT at 4% a year, inside a band of 3% to 4% that targets 67% of supply staked. The community tax is set at 70%, so most of that issuance goes to the community pool, not to stakers. The published yield fits: 30% of 4% of 299.25 million AKT, shared among 89.47 million staked, gives about 4.0% (Coindaro’s arithmetic), against the 4.01% shown by the stats API. The pool held 3.52 million AKT on 27 October.

AKT supply and staking (27 October 2026)FigureSource
Total supply on chain299.25 million AKTakashnet-2 bank module
Maximum supply388.54 million AKTCoinGecko
Staked (bonded)89.47 million AKT, about 30% of supplystaking pool
Inflation / staking yield4.0% / 4.01%mint module, Akash stats API
Community pool3.52 million AKTdistribution module
Unbonding period21 daysstaking parameters

The take rate is the protocol’s cut of each lease payment. Proposal 228 set it at 20% for axlUSDC in September 2023, and proposal 267 cut it to 1% for AKT and 2% for axlUSDC in September 2024. The BME explainer now says providers earn “without take rates”, and the visible charge on chain is the 0.25% minting fee, but the take module’s settings are not exposed by the public API, so Coindaro could not confirm the current rate. How supply, emissions and burns interact is set out in our tokenomics glossary entry.

Akash history: from a 2020 mainnet to the Solana vote

The chain went live on 25 September 2020, the date set in Akash’s blog and reported by ForkLog, with transfers, staking and governance only. Tenants arrived with Mainnet 2 on 8 March 2021; the team’s update that month counted 234 applications deployed by 57 developers.

  • 31 August 2023: Mainnet 6 brings Nvidia GPUs, and proposal 228 adds axlUSDC payments, the take rate and a deployment deposit.
  • September 2024: proposal 267 lowers the take rate to 1% on AKT and 2% on axlUSDC, with 84.5 million AKT in favour.
  • March 2025: a spam attack hits the network; validators and the core team change transaction costs within hours, according to the 2025 review.
  • 21 May 2025: Greg Osuri testifies at a hearing of a House oversight and investigations subcommittee on using federal land to power technology, speaking about decentralised compute and the energy needs of AI.
  • 13 October 2025: Osuri tells The Block that Akash will retire its Cosmos chain and calls Solana a “strong contender”. On 28 October, Mainnet 14 moves the chain to Cosmos SDK 0.53.
  • 22 November 2025: AkashML launches. On 15 December Osuri opens AEP-79, the request for proposals on a new home for the chain.
  • 23 March 2026: BME, CosmWasm smart contracts and an on-chain oracle fed by Pyth go live with the same upgrade.
  • 28 July 2026: confidential compute launches.
  • 26 October 2026: Akash explains why it picked Solana, and proposal 346 opens.

Akash Network and Solana: what proposal 346 would change

The proposal is a signalling vote and carries no code. A yes approves the migration plan in AEP-79 and allows AKT to be issued as an SPL token on Solana once two independent audits of the token and the migration portal are done, which the text expects two to three weeks after the vote passes. The marketplace stays on akashnet-2 for now, and no cutover date is set.

The case for leaving is cost. The proposal text says “Akash pays the full cost of a sovereign chain without using its powers”: about 100 validators bond AKT only to secure consensus, and the core team maintains its own forks of the Cosmos SDK, CometBFT and gogoproto. Akash says it weighed seven options against 15 hard requirements and that two passed every one, Solana and an existing EVM layer 2. A dedicated Arbitrum Orbit rollup is kept as a fallback, though it would bring back the chain-running work Akash wants to drop, and Ethereum mainnet was rejected because its gas price spikes under load. The Solana Foundation has offered engineering, audit and gas-subsidy help; the plan says it does not depend on it.

Akash’s own estimates, published with the Solana announcement, put a marketplace operation about 9 times cheaper on Solana than on an EVM layer 2 on a quiet day and about 300 times cheaper under load. Placing a bid would cost about $0.0008 against $0.007. Fees for the whole marketplace would come to about 0.075 SOL a day, plus a refundable deposit of about 0.024 SOL per lease, roughly $3.60 with SOL at $150. These are design figures, not measured costs.

What a yes changes for AKT holders and stakers

The swap would be one-to-one, under the same ticker, with six decimals. Each AKT sent through the portal would leave circulation on the Cosmos chain, so total supply would not change, and holders would not need SOL to migrate. The portal would run one way for two years after the token launch. “There is no path back,” the proposal says.

Staked AKT could migrate in one transaction, which starts the 21-day unbonding, claims pending rewards and delivers both on Solana when the unbonding ends. AKT held on Osmosis or other IBC chains must first come back to akashnet-2. Console-managed wallets would be moved automatically, and exchanges would migrate customer balances on their own timetable. ACT would stay non-transferable through Solana’s Token-2022 standard. Readers new to locking tokens can check how unbonding works in our staking glossary entry.

Staking rewards and Cosmos-side inflation would stop at cutover. Validators would keep their role and income until the old chain halts, 90 days after cutover, and a continuity payout is planned, its size unpublished. On Solana, staked AKT would no longer secure any chain; Akash’s post says “new mechanisms for staking will be announced prior to cutover”. For tenants and providers, running leases would continue for those 90 days while new deployments go to Solana. SDL files stay the same, and providers would upgrade their software once and register again.

After the vote come several gates: two audits before the token launch, then a provider testnet of at least six weeks with 25 or more independent operators, a 14-day dress rehearsal and a separate cutover proposal. Upgrades to wind down and halt the chain would each be submitted 21 days ahead, and IBC partners get 60 days’ notice. Governance or validators could still stop the marketplace move until about six hours before cutover, but every token migration is final once the portal opens.

A ballot box and a bridge between two stacks of blocks beside a calendar
A yes on proposal 346 opens a one-way path to Solana, with audits, a provider testnet and a separate cutover vote still to come.

Several points are still open in the proposal itself: how AKT left on the old chain at halt reaches the portal, what happens to AKT not migrated within two years, the emissions curve and the size of the validator payout. Akash’s post adds that the treatment of unvested AKT is not settled. In October 2025 Osuri had said the new home would remain IBC-compatible; under this plan, Cosmos links after halt would go through third-party bridges. The precedent Akash cites is Helium, which moved from its own chain to Solana in 2023. Noble, a stablecoin chain, left Cosmos for its own EVM chain on 18 March 2026.

Governance and validators: 79 seats filled out of 100

Decisions on Akash are taken by stake-weighted votes over seven days. A proposal needs a quorum of 33.4% of staked AKT and more than 50% yes among votes that are not abstentions. If “no with veto” passes 33.4%, it fails and its deposit, 1,000 AKT for proposal 346, is burned. On the 27 October count, veto votes stood at 0.0015%.

The chain allows up to 100 validators, each charging at least 5% commission, and 79 were in the active set on 27 October; the “about 100” in the proposal is closer to the cap than the count. Akash’s Solana post also mentions a validator count of about 770 to 800, but that figure describes Solana’s network, not Akash’s.

Development is paid largely from the community pool. Holders approved $340,000 in April 2025 for the Accelerate conference, held in June, and in October 2026 proposals 344 and 345 paid Overclock Labs for client engineering and support work in the third quarter. Protocol changes go through Akash Enhancement Proposals (AEPs) on GitHub. AEP-79 names Overclock Labs as the main evaluator of the migration bids, so the company that builds Akash also judged where it should move.

Akash Network founders: Greg Osuri and Adam Bozanich

An Akash blog post from September 2022 lists Greg Osuri as chief executive and co-founder of Overclock Labs and Adam Bozanich as its chief technology officer and co-founder. Osuri is the public face of the project. He wrote AEP-79, presented a talk on decentralised cloud at Nvidia GTC in March 2025 and testified to Congress two months later.

No funding round could be confirmed from a primary source. Data aggregators give conflicting seed figures for Overclock Labs between 2017 and 2020, so this page leaves them out. The size of the team is not published.

Akash vs Render, Flux and the big clouds

When Akash added GPUs in 2023, The Block named Render and Flux as its decentralised rivals. Others have joined since, among them io.net and Aethir, which also pay hardware owners in tokens; Coindaro has not verified their revenue figures and does not compare them here. Bittensor takes a different route: it pays its token for AI output scored by validators, while Akash sells raw capacity at a price agreed in advance.

The harder competition is the hyperscalers and GPU clouds such as CoreWeave, which sell reliability, support and long contracts. Akash competes on price and on open access: anyone with a wallet can deploy through Console Air or the command line. Its savings claims come from the company itself.

Is Akash Network legit? Audits, risks and our score of 4 out of 5

Akash is a working network, with more than six years since its mainnet, open-source code on GitHub and a usage record anyone can query. That does not make it low risk. Coindaro rates it 4 out of 5, a high risk, for these reasons.

  • No public security audit of the chain, the provider software or the BME module was found as of 27 October 2026. Community “auditors” check providers’ hardware, not the code.
  • The Solana move is a one-way migration through a portal that has yet to be built and audited. Akash itself lists fake portals, a supply mismatch and thin provider capacity after re-registration among the threats.
  • After halt, Akash would lose native IBC and its own validator set and would depend on Solana’s block times, fees and outages. Live slashing is not yet available on Solana.
  • Demand is small: 662 active leases and about $5,760 of spend over the 24 hours to 19:27 UTC on 27 October, measured by the project’s own API.
  • BME has run only since March 2026, and the take rate and the emissions after a move are not public.

For the Solana programs Akash promises four independent audits, an economic review and a public audit contest, with reports published within 30 days. Until those reports exist, the Solana code is unaudited. Automated security scores such as CertiK’s are not audits.

What to watch next: 2 November, then the token launch

The first date is 2 November 2026 at 15:02 UTC, when the vote on proposal 346 closes. If it passes, AKT on Solana would follow two to three weeks later, after its audits, which places it in mid to late November at the earliest. Akash’s blog says two weeks; the proposal text, which is what stakers vote on, says two to three.

Then comes the slower part: the provider testnet, the rehearsal and a cutover proposal with no date. Before that proposal, Akash still has to publish the staking design, the validator payout and the emissions curve, and name the single official domain of the migration portal.

The 2026 roadmap has a backlog of its own. On 27 October it still gave estimated dates, not completion dates, for hardware verification with trusted execution (AEP-29, due 30 May), private overlay networking (AEP-48, 30 July) and virtual machines (AEP-49, 30 August). AEP-79, the migration itself, is due by 31 December 2026.

Usage gives the baseline: 662 active leases and 146 leased GPUs on 27 October. If proposal 346 passes, the next date Akash has to announce is the token launch, through its official channels and on one portal domain.

Strengths and points of vigilance

Strengths

  • Prices are set by providers bidding under a ceiling the tenant chooses, and payment runs per block from escrow.
  • Usage can be checked by anyone through a public API: 662 active leases and about $6.49 million spent since launch on 27 October 2026.
  • The network sells GPUs, managed AI inference through AkashML and, since July 2026, confidential compute.
  • The Solana plan was put to an on-chain vote with published terms, gates and a list of open questions.

Points of vigilance

  • No public security audit of the chain, provider software or BME module was found as of 27 October 2026.
  • Demand is small: about $5,760 spent in a day and 146 GPUs in active leases on 27 October 2026.
  • The move to Solana is one-way, ends native IBC and validator income, and leaves emissions and payouts unpublished.
  • The take rate under BME and real GPU utilisation cannot be checked independently.

Team

  • Greg Osuri Co-founder and CEO of Overclock Labs; author of AEP-79, the migration plan
  • Adam Bozanich Co-founder and CTO of Overclock Labs

Roadmap

  1. 2020-09-25Mainnet 1: la catena akashnet parte con trasferimenti, staking e governanceDone
  2. 2021-03-08La Mainnet 2 apre la rete alle distribuzioni dei tenantDone
  3. 2023-08-31La Mainnet 6 aggiunge le GPU Nvidia; la proposta 228 aggiunge i pagamenti in axlUSDC e il take rateDone
  4. 2025-10-28La Mainnet 14 porta la catena al Cosmos SDK 0.53, due settimane dopo l'annuncio del piano di lasciare CosmosDone
  5. 2025-11-22Parte AkashML, l'inferenza di IA gestitaDone
  6. 2026-03-23Entrano in funzione il burn-mint equilibrium (ACT), CosmWasm e un oracolo on chain alimentato da PythDone
  7. 2026-07-28Calcolo confidenziale con Intel TDX, AMD SEV-SNP e GPU NvidiaDone
  8. 2026-10-08Completato lo strumento unificato da riga di comando (AEP-89)Done
  9. 2026-11-02Si chiude alle 15:02 UTC il voto sulla proposta 346 (passaggio a Solana)
  10. 2026-11AKT emesso come token SPL su Solana due o tre settimane dopo un sì, a due audit completati
  11. 2026-12-31Completamento stimato dell'AEP-79, la migrazione; nessuna data di passaggio fissata

Crypto-assets are volatile and you can lose all the money you invest. Nothing published on Coindaro is investment advice.

Related