ATH vs AKT: The GPU Cloud and the Open Market

The smaller token belongs to the larger compute business.

At the cutoff, AKT carried a $170.5M circulating market capitalization. ATH carried $92.7M. The activity beneath those values ran the other way.

Aethir's latest twelve complete months showed $90.97M of gross network billing. Akash's latest twelve-month marketplace window showed $1.92M of spend. Aethir processed about 47 times more compute payments, yet the market valued AKT at 1.84 times ATH.

That is not a pricing error we can solve by dividing market cap by “revenue.” The two systems sell different products, keep different portions of a customer payment, and put decentralization in different layers. The useful question is not which project has the bigger number. It is what each number proves.

Aethir coordinates the cloud. Akash publishes the market.

Aethir assembles outside hardware into a GPU service for AI and gaming workloads. Its Cloud Host rules prescribe locations, x86 bare metal, Ubuntu, uniform GPU configurations, public networking, onboarding tests, 99% uptime, response times, and continuous monitoring.

The HostAgent fingerprints hardware, sends telemetry, maintains a WebSocket connection, accepts software updates, and supports remote management commands. Aethir's off-chain Indexer matches work to supply using price and service quality.

The buyer gets one coordinated service rather than a negotiation with each data center. Aethir takes on more sales, scheduling, integration, and support work. It also keeps more control over admission, placement, monitoring, price, and payout access.

Akash exposes the market itself. A provider registers on chain, publishes attributes and prices, and runs Kubernetes. A tenant posts a deployment specification. Providers bid. The tenant chooses a lease. Providers can decline orders and close leases.

The raw path is closer to infrastructure than a managed cloud. A wallet and command line do not require KYC. Console Air lets a user self-host the interface, keep custody of keys, and choose RPC and API endpoints. Overclock Labs can restrict users of its managed Console under its terms, but Console is not the only protocol route.

Enterprise buyers often want someone to absorb the coordination Akash leaves open. Razer's April 2026 AVA Mini case study used AkashML, a managed layer pooling RTX 4090 and 5090 capacity behind one endpoint. TensorOpera separately confirmed it used Aethir GPUs to train Fox-1 and integrated Aethir's pool. Both networks run real compute. They package it differently.

The same customer payment creates different claims.

Aethir reports a gross line and retains a documented portion. Akash reports marketplace spend and deliberately keeps no lease-settlement take.

Aethir's service-fee documentation assigns 80% to Cloud Hosts and 20% to the protocol. The protocol portion is not audited company revenue or profit. Public material does not identify the consolidated legal entity that recognizes it, and ordinary ATH holders do not automatically receive it.

Akash's application-layer documentation says lease settlement has no take rate. Under AEP-76, a user burns AKT to mint dollar-denominated ACT. A provider later redeems ACT and receives newly minted AKT at the settlement-time price. The mechanism links usage to AKT supply without converting marketplace volume into protocol revenue.

Aethir's lead survives every reasonable definition.

Aethir recorded $127.90M of gross network billing in calendar 2025. Akash recorded $3.16M of marketplace spend, a 40.5x gap. Across the latest complete windows, the figures were $90.97M and $1.92M, a 47.3x gap.

Apply Aethir's published 20% split and its implied trailing protocol share is $18.19M before incentives and operating costs. That retained layer alone was 9.5 times Akash's entire marketplace spend.

Recent activity narrows the ratio without closing it. Aethir recorded $3.43M in July. Akash recorded $250,886 across the 30 complete days ending August 20. Aethir remained 13.7 times larger on those measures.

The lead comes with a trend problem. Aethir monthly billing fell 75.4% from a $13.92M September 2025 peak to July 2026. AlphaRank's prior audit found 1,250 service-fee deposits from 61 addresses, with the ten largest producing 95.5% of approximate price-weighted net deposits. Wallets can aggregate multiple customers, but public evidence does not separate beneficial payers, contracts, grants, resellers, refunds, or related parties.

Akash begins from a much smaller base. Q2 2026 spend reached $488,090, up from $253,533 in Q1. July added $230,564 and the first 20 days of August added $175,352. The direction improved. The level still does not describe a large current compute business.

Akash's 2025 review said deployment creations rose 466% to 3.14M, while new leases rose 2.2% and active year-end deployments fell 69%. Spend still rose 128%. Trial credits and provider incentives were active. That proves more attempts and more spend, not millions of durable paying workloads.

Decentralization changes when the layer changes.

Akash gives buyers and providers a more credible exit. Aethir gives enterprise buyers more centralized accountability.

Akash is not censorship-proof. Providers can refuse bids or close leases. Validators and governance can change rules. Managed interfaces can restrict users. The difference is that no single interface is mandatory and the operating stack is inspectable and forkable.

Aethir's control can improve consistency. It also means Aethir can decline hosts, enforce identity and hardware rules, assign workloads, and control payout access. Checker licenses do not change those boundaries.

Akash shows its concentration. Aethir does not expose the comparable denominator.

The Akash dashboard showed 63 active providers, 423 total GPUs, and 258 active GPUs at 15:18 UTC. A provider-registry refresh 27 minutes later contained 62 online providers, including 35 with GPUs, and reconciled closely at 422 total and 260 active.

Seventeen online GPU-provider records carried an organization value of overclock or Overclock. The attribute schema describes the field as the organization that owns the provider, while current audit documentation calls it the operator name.

Together, those records advertised 352 GPUs and 254 active GPUs: 83.4% of capacity and 97.7% of active GPUs. The fields could represent one operator, affiliates, managed infrastructure for outside owners, or several sites sharing a common label. The narrow conclusion is operational concentration under common metadata, not common legal ownership.

Aethir reports 433,449 containers across 94 countries. Its documentation defines a Container as a virtual endpoint, not a unique physical GPU. Country rows summed to 546,909, 26.2% above the headline, without a public reconciliation. Those units cannot be placed beside Akash's physical inventory entries.

Aethir does not publish a current operator-level concentration table or independently audited physical-GPU map. Its supply may be commercially broader than Akash's registry suggests, but public evidence cannot test it. Akash's disclosure reveals an uncomfortable concentration that Aethir's disclosure does not let us calculate.

BME links usage to supply. It does not make every dollar deflationary.

Akash launched BME in March 2026. Users burn AKT at the mint-time price to create dollar-denominated ACT. Providers redeem ACT later and the system remints AKT at the settlement-time price.

At cutoff, the BME dashboard showed 1.840M AKT burned to mint $1.098M of ACT. Providers had redeemed $909,668 of ACT and the system had reminted 1.582M AKT. Net burn was 257,149 AKT.

On July 29, cumulative net burn had been 288,806 AKT. It fell by 31,657 over the next 23 days even as gross burn rose. August 20 alone showed negative daily net burn of 3,633 AKT. If AKT falls between mint and settlement, more AKT must return to satisfy the same dollar claim.

The mechanism also depends on an external price path. Akash documentation says off-chain Hermes relayers carry Pyth prices through Wormhole contracts into the chain oracle. Governance controls contract upgrades and configuration. BME is transparent and open-source, but its monetary outcome depends on relayers, oracle infrastructure, governance, and market price.

Scale matters. Net burn equaled 0.0866% of the 297.0M AKT supply. At 4% inflation, estimated annual issuance was 11.88M AKT. Cumulative net burn since activation equaled 2.16% of one year's issuance. The vault's 1.572 collateral ratio was healthy. The mechanism worked; demand had not made it a dominant deflationary force.

Both tokens sit inside the system. Neither is a claim on the business.

ATH is used in customer settlement, Cloud Host staking, Checker rewards, and supply incentives. Aethir's 20% protocol share creates a treasury resource. Public documents do not route it automatically to ordinary ATH. Half of the 42B maximum supply is allocated to provider and Checker rewards. At cutoff, 20.13B ATH, or 47.9% of the maximum, circulated.

AKT secures its own chain, votes on governance, underwrites ACT minting, and settles providers. BME connects usage to supply, but the sign and size of the net result depend on price and redemption timing. Marketplace growth can benefit providers without creating protocol income. Inflation was 4%, and observed net burn remained small beside issuance.

Neither token is equity. Neither grants a public claim on consolidated profit. Aethir has a clearer retained protocol fee and a weaker automatic holder bridge. Akash has a clearer security role and a usage-linked monetary mechanism with little current magnitude.

The market is paying for protocol qualities that revenue does not measure.

At cutoff, CoinGecko showed AKT at a $170.5M circulating market cap and ATH at $92.7M. Market capitalization prices future utility, security, scarcity, and optionality alongside current sales.

These are hypotheses, not a recovered market model. AKT's premium asks investors to pay for an open protocol whose current marketplace is small. ATH's discount comes with much larger activity, but falling billing, concentrated unidentified deposit addresses, a gated control plane, and substantial future supply.

We did not build a social-attention index. Higher market capitalization and rank are the observable attention proxies used here.

Each project is winning a different contest.

Aethir is the stronger compute business. Akash is the stronger open protocol.

The comparison turns on what the investor wants the token to represent.

Aethir processes about 47 times more disclosed compute payments. Even its implied 20% share was 9.5 times Akash's whole marketplace. The defensible claim is that Aethir runs the larger commercial network and retains a documented fee layer, not that it earns 47 times more profit.

Akash provides permissionless deployment, provider entry, open-source software, and a more distributed settlement layer. Its visible GPU supply remains concentrated, its marketplace is small, and BME has not offset much inflation. AKT's higher market cap prices the open protocol and its future use more heavily than today's compute sales.