Akash Network

What Akash is

Akash is a decentralized compute marketplace. Workload deployers (tenants) post deployments specifying compute requirements as SDL (Stack Definition Language, a Kubernetes-flavored YAML). Provider operators bid; the lowest bidder gets the workload. Settlement happens on-chain in AKT.

Compared to AWS:

• No account; payment via wallet.
• Sanctioned, under-banked, or pseudonymous users can deploy.
• Provider diversity beats hyperscaler concentration.
• Container-only — Kubernetes Pod-equivalent workloads.

Compared to Phala / Marlin:

• No TEE attestation by default; trust the provider operator (or layer TEE on top).
• Mature Kubernetes-style scheduling.
• Larger active provider set.

Architecture

Tenant         Akash chain (Cosmos SDK)        Provider
   │                  │                            │
   ├─ post SDL ──────►│                            │
   │                  ├─ broadcast deployment ────►│
   │                  │                            ├─ submit bid
   │                  │◄────── bids ───────────────┤
   ├─ select bid ────►│                            │
   │                  ├─ create lease ────────────►│
   │                  │                            ├─ run workload
   ├─ direct conn  ─────────────────────────────►  │ (container)

Once the lease is created, the tenant talks to the provider directly (port forwarding, ingress, logs); the chain handles only payment and lifecycle events.

SDL — the workload spec

version: "2.0"
services:
  web:
    image: nginx
    expose:
      - port: 80
        as: 80
        to:
          - global: true
profiles:
  compute:
    web:
      resources:
        cpu:
          units: 0.5
        memory:
          size: 512Mi
        storage:
          size: 1Gi
  placement:
    westcoast:
      pricing:
        web:
          denom: uakt
          amount: 1000
deployment:
  web:
    westcoast:
      profile: web
      count: 1

A Helm-chart-shaped YAML that Akash providers can run. Most existing Docker / Kubernetes apps port over with minimal changes.

GPU support

Akash added GPU support for AI workloads in 2023-2024. Provider operators can advertise:

• GPU model (RTX 4090, A100, H100, H200, B200, etc.).
• VRAM size.
• Custom labels for filtering.

Tenants specify the required GPU type in SDL. The reverse auction picks providers offering it. By 2025-2026 the Akash GPU marketplace had thousands of GPUs across consumer and datacenter classes; the long tail of consumer GPUs makes Akash a low-cost option for inference workloads that don't need the latest H200 or B200 hardware.

AKT and economics

• AKT token for payment, staking, and governance.
• A small take rate is charged on every lease, going to staking rewards and the DAO treasury.
• Provider rewards based on staked AKT and uptime.
• Provider services. Providers can also offer specialized features (TEE, IPFS, persistent storage) at premium pricing.

Akash's economic story is closer to "marketplace with a token" than to pure DePIN tokenomics. AKT exists primarily to coordinate, not to be speculatively staked for unrelated workloads.

Recent developments (2024-2026)

BME (Burn-Mint Equilibrium) — Live March 23, 2026

The most significant tokenomics change in any major DePIN compute network. AEP-76 (BME) activated on Mainnet 17 following Proposal 318:

User payment → buys AKT on open market → AKT burned → produces ACT (non-transferable stablecoin)
ACT settles transaction with providers
Providers mint ACT back to AKT at market price

• Every compute transaction burns AKT, creating deflationary pressure as usage grows.
• Providers receive stable ACT, satisfying enterprise settlement needs.
• Post-launch: Akash crossed $5M monthly compute spend in Q1 2026, confirming the tokenomics thesis.
• Testnet: 250+ participants, 10+ test categories, $10,000+ incentives distributed before mainnet launch.

AkashML — 1.7 Billion Tokens/Day on OpenRouter

AkashML launched November 2025 with an OpenAI-compatible API. As of Q1 2026:

• Processing 1.7 billion tokens/day via OpenRouter integration.
• OpenAI-compatible API: swap base_url to point at AkashML endpoints.
• Dominant player in the AI inference DePIN ecosystem per Q1 2026 report.

Homenode Early Access — Live May 2026

Purpose-built for consumer/prosumer GPU contribution. Phase 1 supported GPUs:

• Enables average users to participate as providers — not just data center operators.
• Resilience argument: distributed home nodes eliminate single points of failure.
• Physical infrastructure argument (Greg Osuri, NEARCON SF): conflicts and grid failures in one region are absorbed by nodes elsewhere.
• URL: homenode.akash.network

Akash Agents Platform — One-Click AI Agent Deployment

Built by community member Sandeep Narahari:

• Powered by AkashML inference.
• One-click agent deployments on decentralized compute.
• No infrastructure management overhead.
• Inherits Akash's provider diversity and anti-censorship properties for agentic AI workloads.

AEP-65: Confidential Computing with TEE — Live May 19, 2026

Akash launched Confidential Computing (announced via X, May 2026):

• Provider operators cannot read tenant workloads inside TEE.
• Attestation reports prove hardware integrity on-chain.
• Composable with Phala and 0G and Confidential-AI-Inference stack.
• AEP-29 (Hardware Verification using Trusted Execution) already live for provider attestation.
• AEP-65: enabling providers with TEE hardware to offer confidential workload deployments.
• This eliminates the primary trust gap that separated Akash (trust the provider) from Phala/Marlin (TEE-attested).

DOOOR — TEE Integration on Akash (Community SDK)

Community project integrating Trusted Execution Environments into Akash Network, enabling secure and verifiable AI execution. See GitHub discussion (akash-network/discussions/872) for the open-source SDK approach.

500,000 AKT Burned via BME — On-Chain Evidence (May 26, 2026)

From on-chain data: 500,000 AKT burned via BME mechanism. This is direct evidence of real compute usage translating to token deflation — the core thesis of Akash's March 2026 tokenomics upgrade. AKT trading range ~$0.85–$0.97 in May 2026, up ~24.85% weekly as top AI/DePIN gainer.

GPU Marketplace Growth

• Thousands of GPUs across consumer and datacenter classes.
• H200 and B200 nodes added to GPU waitlist.
• AkashML OpenRouter integration making AI inference the dominant workload.

Akash Network 2025 Year in Review Highlights

• Razer partnership: $0.01/image cost via Akash's peer-to-peer GPU market.
• Venice.ai: end-to-end encrypted AI inference mode on Akash — prompts encrypted on device, stay encrypted in transit.
• Flashback Labs: built privacy-first AI platform letting users create AI twins trained entirely on their own data.
• Sigea Cloud Labs: deployed privacy-first cloud storage on Akash.
• Codex: deployed decentralized Codex Storage Nodes via Akash.
• Morpheus: integrated Akash into Smart Agent Builder compute marketplace.
• Akash Console UX is friendlier than the original CLI-only flow.
• Network-funded grants for specific use cases (AI inference, model serving, dev tools).

Q4 2026 Roadmap — Four Firm Dates Published (September 2026)

The Akash 2026 roadmap page now lists four Q4 deliverables with concrete completion dates:

What changed: the team's published schedule moved from quarterly buckets to fixed dates. AEP-46 Spot Instances has been Draft for 23 months; AEP-67 Console Bid Screening has been Last Call for 16 months. The Sep 1 weekly confirmed AEP-46 + AEP-67 were delivered Aug 30/31 — those were the preceding gates to the Q4 deliverables.

AEP-79 Shared Security is the security/economics pillar the Cosmos-to-Solana migration debate has been circling. The late-2026 target means a chain decision (or a no-migration defense) is structurally required by year-end. AKT traded $0.465–$0.561 across Sept 4–7 (CMC AI Sep 7), with a "$2 target" framing from a bullish analyst offset by persistent sell pressure testing critical support.

See Decentralized Compute Tuesday — September 08 2026 Weekly Update for the full Q4 roadmap table and AKT price context.

Comparison

Use cases

Cheap inference

Run open-weights LLMs (Llama 3, Mistral, etc.) on Akash provider GPUs. Cost is roughly 30-60% of AWS for comparable hardware. No KYC. Pay in AKT (or via fiat-on-ramps that abstract this).

Web2 service hosting on a censorship-resistant fabric

A Nostr relay, a website, a Bitcoin node — anything you'd put on a small VPS — runs on Akash with similar uptime. A provider can refuse, but the marketplace lets you redeploy elsewhere.

Distributed CI and batch jobs

Spawn ephemeral workers via SDL for periodic batch jobs (data ETL, ML training nights, scientific compute) without committing to fixed cloud spend.

TEE-as-a-service composition

Run Phala's dstack or a Marlin Oyster CVM image as your container. Akash provides the box, the TEE provides the integrity layer. The pattern is documented in MOC - Composing Primitives.

Trade-offs

Strengths

• Open source, permissionless.
• Real GPU marketplace at scale.
• Kubernetes-flavored UX, low porting cost from existing apps.
• Mature, with years of production operation.
• Lower cost than hyperscalers for many workloads.

Limitations

• No native TEE attestation. Trust the provider, or combine with attested workload images for integrity.
• Provider quality varies. Some are professional shops; others are home-rigs with unstable links.
• Limited managed services. No managed databases, queues, etc.; bring your own.
• Egress can be expensive depending on provider pricing.
• Token-coupled UX. Fiat onramps exist but are friction.

Attack surface

Operational patterns

Permissionless Nostr relay

Akash deployment with strfry container
  ↓
Public WebSocket on Akash provider IP
  ↓
NIP-11 relay metadata

If a provider drops you, redeploy elsewhere. Relay state can be replicated across Akash providers via standard Nostr replication.

Akash plus TEE = confidential compute

SDL specifies a TEE-aware image (Phala dstack, Marlin Oyster snapshot)
  ↓
Image boots inside provider's TDX/SEV-SNP machine
  ↓
Workload attests to client; client verifies

The Akash chain coordinates payment; the TEE provides the integrity layer; the workload is end-to-end confidential.

Primary sources

• Akash Network — akash.network.
• Documentation — docs.akash.network.
• Source — github.com/akash-network.
• SDL reference — docs.akash.network/readme/stack-definition-language.
• Cosmos SDK — docs.cosmos.network.

For a deployment walkthrough with an exit plan, see deploying on decentralized compute.

Continue with Compare decentralized compute platforms.