# io.net

> Independent io.net review: IO tokenomics, the on-chain IDE burn, revenue below the marketing claims, and why its GPU count overstates the network.

- Scores: Freedom 3.9/10 (grade F), Returns 5.4/10 (grade D)
- Category: compute · Token: IO
- Published: 2026-03-02 · Updated: 2026-10-13
- Canonical: https://ownyourmind.ai/projects/io-net/
- Data (scores + integrity verdicts, JSON): https://ownyourmind.ai/api/projects/io-net.json
- Cite as: Own Your Mind (ownyourmind.ai), licence CC BY 4.0. Link the canonical URL.

Charts and interactive widgets are omitted here; their data is in the JSON above.

---

## What it does

io.net is a GPU marketplace built on Solana. You supply GPUs, you rent GPUs, and io.net sits in the middle taking a cut. The "decentralised" part is that GPUs come from independent suppliers in what io.net says are more than 130 countries. Everything else, the orchestration, matching, billing, verification, and the node software itself, is run by io.net.

The architecture has seven layers, from a ReactJS frontend down through FastAPI and GraphQL APIs, PostgreSQL and Redis databases, RabbitMQ task queues, and a Kubernetes orchestration layer running IO-SDK (a proprietary fork of Ray). GPU suppliers run the closed-source IO Worker binary, pass Proof-of-Work benchmarks and Proof of Time-Lock verification, stake IO tokens as collateral, and receive hourly block rewards plus job payments. Consumers deploy GPU clusters via IO Cloud, or access 30-plus open-source AI models through the IO Intelligence API (OpenAI-compatible endpoint).

The project launched on Binance Launchpool in June 2024 after raising $30 million in a Series A led by Hack VC at a $1 billion valuation. Backers include Multicoin Capital, Anatoly Yakovenko (Solana founder), OKX Ventures, Delphi Ventures, Animoca Brands and others. The team is estimated at roughly 100 people across New York and San Francisco.

The founding story is complicated. Ahmad Shadid founded io.net in 2022, resigned as CEO two days before the token launch in June 2024 amid allegations about his background in Arabic-language crypto projects and questions about inflated GPU metrics. Co-founder Tory Green (Stanford, West Point, ex-Merrill Lynch, ex-Disney strategic planning) took over as CEO, then moved to chair of the io.net Foundation in April 2025. Gaurav Sharma (ex-Binance VP of Engineering) was appointed CEO. Three CEOs in under two years.

## The product in practice

Grounded in our io.net review, the render-vs-akash comparison, and community sentiment from independent forums and reporting. We haven't deployed a cluster on io.net, so the hands-on notes are attributed.

**Who it's for.** AI and ML developers who want to rent GPU clusters cheaply for Ray-based or Kubernetes workloads, and GPU owners looking to earn by supplying idle hardware. IO Cloud handles cluster deployment, io.Intelligence exposes 15-plus open models through an OpenAI-compatible API, and IO Worker is the supplier client. It is crypto-native in the small ways that matter: you connect a wallet rather than create an account, and payouts settle in IO on Solana.

**What it's like to use.** On the demand side, spinning up a cluster is quick and marketed well below hyperscaler pricing, though the price only appears after you configure the job, which drew comparisons to competitors like RunPod that show rates upfront. The harder problem is on the supply side, where GPU owners on independent forums report sitting idle most of the time and picking up few paid jobs, earning mainly from block rewards rather than paid compute demand (our [io.net worker guide](https://ownyourmind.ai/use/how-to-earn-with-io-net/) walks the setup and the honest payout maths). Operational friction compounds it: by io.net's own documentation, worker job earnings can't yet be withdrawn, USDC withdrawals run through a manual support ticket, and automated payouts aren't enabled. Serious GPU renters also haven't adopted it, and independent cloud-rating systems that cover the centralised providers leave decentralised networks like io.net out of the evaluated set entirely.

**The credibility overhang.** io.net carries the heaviest trust baggage of the compute set. In 2024 its GPU counts were shown to be wildly inflated (public figures ranged from a few thousand to over 500,000, while a critic put the actually-rentable number in the low hundreds), a metadata vulnerability let attackers spoof roughly 1.8 million fake GPUs, and the founder stepped down as CEO two days before the token launch. The team since added authentication and proof-of-work verification, but the utilisation gap persisted: independent data put verified-active GPUs at around 2% of registered supply. Three CEOs in under two years hasn't helped the stability story.

**Demand, honestly.** The current pitch rests on inference rather than cluster rental: io.net said in our July 2026 read that it processes billions of AI tokens a day through OpenRouter, which still lists it as a provider, even if the throughput figure is its own. Cluster-rental revenue is thinner (an annualised figure in the low tens of millions per DePINpulse, down from a higher earlier io.net claim, and not reconciled on-chain the way Aethir's is), and the headline demand signal is a single undisclosed enterprise contract.

io.net says the June 2026 IDE upgrade routes revenue into IO buybacks, and every epoch we have indexed so far, 372 of 957 as of late September 2026, reconciles its burn to the chain, while the purchases behind it stay io.net's own figure; whether the demand is durable is the open question.

**The vision.** io.net wants to be the cheap, permissionless alternative to hyperscaler GPU cloud for AI workloads, aggregating idle supply worldwide. The technology ships and io.net reports a growing inference line, but the network still has to prove that durable, diversified compute demand exists behind the emissions and the one big contract. Watch active utilisation and named customers rather than registered GPU counts.

## Value proposition

Cheaper GPUs for AI workloads than AWS, GCP or Azure. io.net claims 70-90% cost savings. Whether you actually get that depends on the workload, the GPU model and availability.

What is interesting is the revenue trajectory, and it has not gone the way the early figures implied. io.net reported Q4 2024 revenue of $3.1 million and Q1 2025 of $5.7 million, an 82.6% quarter-on-quarter jump that annualised north of $20 million, but that pace did not hold: by mid-2026 DePINpulse tracks the annualised figure at roughly $12.5 million. These numbers trace back to io.net's own reporting through aggregators, with no independent audit of revenue or compute hours published.

The confidential compute offering is a meaningful differentiator. Intel TDX and NVIDIA H100/H200/B200 GPUs provide hardware-based data protection during processing. Users can verify attestation reports. Training data, model weights and architectures stay encrypted. For AI workloads handling sensitive data, this matters.

IO Intelligence provides an OpenAI-compatible API with access to 30-plus open-source models (Llama, DeepSeek, Qwen and others). The Co-Staking Marketplace, launched in February 2025, lets token holders stake alongside GPU operators and share block rewards without running hardware.

The counter-narrative is equally clear. Strip away the DePIN branding and you have a company that runs a GPU marketplace with token incentives. [Akash](https://ownyourmind.ai/projects/akash/) does the same thing but with open-source code, permissionless deployment, on-chain governance and consistent leadership since 2015. Akash also undercuts io.net by roughly $0.07 per GPU-hour on H100s. On the other side, centralised providers like Lambda Labs and Vast.ai offer GPU marketplaces without the token overhead.

io.net's positioning is stuck in between: not decentralised enough to win on the sovereignty thesis, and not reliable enough to win enterprise workloads from AWS. Paying customers exist, but revenue has slipped below its 2025 pace and the long-term competitive moat is unclear.

## Tokenomics

IO is an SPL token on Solana. Max supply: 800 million. Circulating supply is shown live in the fact strip above. The remaining tokens unlock through emissions and vesting schedules running to 2028.

Distribution:

- **Emissions (GPU supplier and staker rewards):** 37.5% (300M IO, originally distributed hourly over 20 years, disinflationary starting at 8% annual rate, decreasing roughly 1% per month, now transitioning to monthly emission epochs)
- **R&D and Ecosystem:** 16.0% (30% unlocked at TGE, linear vest over 35 months from January 2025)
- **Seed investors:** 12.5% (13-month cliff, then linear vest over 23 months; unlocks began July 2025)
- **Core contributors (team):** 11.3% (13-month cliff, then linear vest over 35 months; unlocks began July 2025)
- **Community (airdrops, incentives):** 10.0% (45% at TGE, linear vest over 41 months)
- **Binance Launchpool:** 2.5% (fully unlocked at launch)

Insider allocation (seed plus team plus Series A) totals roughly 34%. The Binance Launchpool provided broad initial distribution, and 50% of total supply is earmarked for community over time, but the "community" emissions primarily benefit GPU suppliers, not passive token holders.

### IDE tokenomics overhaul (live June 2026, in part)

io.net announced the Incentive Dynamic Engine (IDE) in December 2025, a restructure of how emissions and revenue flow through the token. The burn component went live on 1 June 2026 and settles on-chain every epoch. The harder reforms have not. Here is what the design promised against what io.net's own epoch explorer actually shows, as of 16 June 2026.

- **Demand-driven emissions (not live).** The design replaces the fixed schedule with rewards that rise and fall with network demand. The explorer tells a different story: epoch rewards sit at a flat 10,983.8 IO every epoch across a fortnight where active GPUs swung between roughly 800 and 1,450. Emissions are still a fixed drip, not demand-responsive.
- **Buyback-and-burn (live, but not revenue-funded).** The design routes 50% of surplus Fee Vault revenue into open-market IO buys and burns. Surplus only exists when client fees exceed supplier payouts. io.net's own explorer shows the reverse: clients bought 35,756 IO over the period while suppliers were paid 1.37 million IO, so fees run at roughly 2.6% of payouts. There is no surplus to fund a buyback.
- **The burn happens, but it is emission-side.** Roughly 488,000 IO was burned between 1 and 16 June 2026, Solana-settled per epoch and independently checkable. The per-epoch burn tracks the fixed emission at about a third of it, not usage, and at 14 times total client purchases it could not be the revenue buyback the marketing describes. On those June figures, io.net's "powered by network revenue" framing was contradicted by io.net's own data.
- **Supply reduction (partial).** The first fortnight projected to 12 million-plus IO a year, a deflationary force. Set against roughly 32 million IO of annual emissions, that would offset about a third. Circulating supply still grows, and insider unlocks continue through 2028.

By 24 September 2026 its explorer reported about 5.8 million IO purchased by clients against about 2.15 million burned, a 37% burn share, and the 294 epochs we checked on 19 September each reconcile to the burn on the IO mint.  On io.net's accounting, purchases now exceed the burn, so the June reading of an emission-funded burn no longer holds on its current figures. The purchases themselves settle partly in fiat and USDC off-chain, so they remain io.net's claim with an on-chain floor. The cumulative burn works out to roughly 7 million IO a year since 1 June, well short of the 12 million projected.

The verifiable on-chain burn ledger is a clear improvement. The maths now happens in public, which is what the token needed. At launch, the campaign wrapped around it ("IDE is live", "usage turned into scarcity") ran ahead of io.net's own June figures. Score this on what the chain shows, not the launch graphics.

### Price action

IO trades on Binance, Coinbase, Gate, Bybit, KuCoin, WhiteBIT and others. The burn now adds modest buy-side support, funded by client purchases on io.net's own accounting, but it offsets only part of ongoing dilution.

IO is 98% below the all-time high of $6.43, set on 12 June 2024, the day after listing. The all-time low was hit in February 2026.

The token was valued at $1 billion at the Series A. Investors who bought at that valuation are deeply underwater.

## How to participate

**Supply GPUs.** Install the closed-source IO Worker software on NVIDIA GPUs (minimum 4GB VRAM, 100 Mbps down / 75 Mbps up). Stake a minimum of 200 IO per chip (adjusted by GPU model multiplier). Pass a 12-hour initial stress test and maintain minimum 5-hour daily uptime. Earn hourly block rewards plus compute job payments. Higher-end GPUs (A100, H100, RTX 4090) earn significantly more.

**Co-stake IO tokens.** The Co-Staking Marketplace (February 2025) lets token holders stake alongside GPU operators and share block rewards without running hardware. Average APR is roughly 8.7%. Fourteen-day unstaking cooldown during which tokens earn no rewards.

**Rent compute.** Deploy GPU clusters via IO Cloud for AI/ML workloads. Access the IO Intelligence API (OpenAI-compatible) for inference on 30-plus open-source models. Confidential compute VMs available for sensitive workloads. Pay in IO or fiat.

**Build.** IO Intelligence APIs, IO-SDK (Ray fork) for distributed computing, Python library. Documentation is adequate but lacks depth compared to major cloud providers.

## io.net earnings: what GPU operators actually make

Before buying hardware for io.net, understand the honest picture. io.net earnings depend on three variables that the marketing tends to collapse into one: the hourly block reward rate (set by emissions), the frequency of actual paid compute jobs (set by demand), and the network-wide utilisation rate (which is far lower than headline GPU counts suggest). The first two are controllable by io.net and the market. The third is the one that kills most would-be suppliers' return calculations.

### The three revenue streams

GPU suppliers on io.net earn from:

- **Hourly block rewards** paid in IO for keeping the worker node online and passing uptime checks. Block rewards continue whether the GPU is doing actual work or sitting idle. They come from the 300 million IO supplier pool under a disinflationary schedule decreasing roughly 1% per month. The Incentive Dynamic Engine restructure (announced December 2025, implementation planned Q2 2026) will shift this to demand-driven rewards, but that system is not yet live as of April 2026.
- **Compute job payments** paid per rental when a customer actually books the GPU for AI training, inference, or rendering. Higher-end GPUs (A100, H100, H200, RTX 4090) receive proportionally more job payments because they match more rental queries. Consumer cards still earn block rewards but rarely pick up paid jobs.
- **Co-staking participation.** The Co-Staking Marketplace (February 2025) lets IO token holders stake alongside operators and share block rewards. Operators accept a slightly lower effective rate in exchange for a larger bonded stake. Co-stakers earn approximately 8.7% APR on average. Operators use this to meet stake requirements without putting up all the IO themselves.

### Participation requirements for GPU suppliers

| Requirement | Value |
| - | - |
| Minimum GPU VRAM | 4 GB |
| Network | 100 Mbps down / 75 Mbps up |
| Minimum IO stake | 200 IO per chip, adjusted by GPU tier multiplier |
| Initial qualification | 12-hour stress test |
| Daily uptime minimum | 5 hours |
| Co-staking cooldown | 14 days, zero rewards during unstaking |
| Worker software | Closed source |

### The utilisation reality that determines actual earnings

This is the part the io.net landing page buries. In Q1 2025, io.net reported 327,000 registered GPUs but the daily average of verified, active GPUs was 6,720, and only 5,350 were cluster-ready on a given day. That is 2% utilisation of the registered base, and the verified count declined 11.1% quarter-on-quarter even as company-reported revenue grew. The marketing now runs two figures at once, "30,000+ GPUs" on the homepage and "320,000+ GPUs" in the /cloud page's structured data, while io.net's own explorer showed 2,447 devices with 1,199 active when we read it on 13 August 2026.

Practically, this matters enormously for earnings. Block rewards are paid to every operator passing uptime checks. Compute job payments only flow when a renter actually books your specific GPU. If the network has surplus supply (which the utilisation figures indicate), block rewards can dominate take-home earnings while paid jobs remain sparse. Break-even on hardware purchased specifically for io.net supply is therefore driven more by the IO token price and the emission schedule than by rental demand. For consumer GPUs, that break-even is often longer than the hardware's useful life. For data-centre tier hardware (A100, H100), the math is more favourable but still depends on winning enough paid jobs to justify the data-centre economics.

### Current rates and the live dashboard

Because the emission rate and the per-tier GPU multipliers change month to month, the most reliable earnings figure for any specific GPU is the one shown in io.net's live supplier dashboard at `worker.io.net/worker/earnings-and-rewards` (authentication required). Model earnings over at least a 30-day horizon before committing capital, and factor in restructure risk from the IDE: its burn went live in June 2026, but demand-driven payouts are not active yet, so block rewards remain a flat per-epoch drip the design still intends to make demand-responsive.

For a direct comparison of io.net's revenue model against Render, Akash and Aethir, see [Render vs Akash vs io.net](https://ownyourmind.ai/tokenomics/render-vs-akash-vs-ionet/). For the head-to-head against the other Solana GPU DePIN, see [Nosana vs io.net](https://ownyourmind.ai/tokenomics/nosana-vs-ionet/).

## Honest assessment

### What works

Revenue is no longer climbing the way the early prints implied. Messari reported $3.1 million in Q4 2024 and $5.7 million in Q1 2025, but DePINpulse now puts the annualised figure nearer $12.5 million, so the growth did not hold. Both aggregators republish rather than audit, so the numbers still trace back to io.net's own reporting. A newer plus is transparency: the epoch explorer publishes Solana-settled emission and burn data anyone can check.

The confidential compute offering with hardware-based attestation (Intel TDX plus NVIDIA) is ahead of most DePIN competitors. IO Intelligence provides a practical API for AI model access. Leadership under Gaurav Sharma (ex-Binance VP Engineering) brings operational credibility after a turbulent 2024. Eighty-five-plus partnerships announced by January 2025, including integrations with Filecoin, Render Network, ChainGPT, Injective and Nillion.

### What doesn't work

GPU numbers are the headline problem, and io.net does not have one headline. It has two. The homepage offers "instant access to 30,000+ GPUs"; the /cloud page carries that same figure and, in its own structured data, "320,000+ GPUs, 80,000+ CPUs". Those are published simultaneously and differ by more than tenfold.

We read its explorer inventory on 13 August 2026 (2,447 devices, 1,199 active) and again on 23 September (2,976 devices, 1,253 active). Against the lower headline that is a gap of about 10x. Against the higher one it is over 100x. Neither is a rounding difference.

Multi-GPU hosts do not close that gap. The inventory is listed per GPU class and dominated by single-card units (August read): 709 RTX 4090s, 584 spot H100 80G PCIe, 512 H100 80GB HBM3. Only 128 of the 2,447 devices in the August read were NVLink, SXM or DGX class.

The pattern is old. In Q1 2025 the daily average of verified, active GPUs was 6,720 against 327,000 registered, 2% of the base, with only 5,350 cluster-ready on a given day. The verified count fell 11.1% quarter-on-quarter even as revenue grew.

How the gap got this wide is a story from April 2024: approximately 1.8 million fake GPUs attempted to connect to the network to farm airdrop rewards. Attackers exploited vulnerabilities that allowed them to mimic genuine GPU signals and spoof metadata. An RTX 4090 was reportedly split into infinite virtual GPUs. Martin Shkreli publicly highlighted the discrepancies, and according to community reporting at the time an io.net executive acknowledged the failure publicly.

Both the "decentralised" and "open source" branding are misleading. io.net describes itself as "The Open Source AI Infrastructure Platform" on its website. The core platform code (backend, IO-SDK, GPU orchestration, matching engine, billing, IO Worker node software) isn't open source. Only 9 public GitHub repos exist, containing setup scripts, binaries, documentation and demo apps. 131 GitHub stars total. [Akash](https://ownyourmind.ai/projects/akash/) publishes everything.

There's no governance. No DAO. No on-chain voting. No governance forum. No proposals mechanism. All decisions are made by the io.net team and foundation. Token holders have zero governance rights. The complete absence of governance is notable for a project of this size.

io.net achieved SOC 2 compliance, which covers organisational security controls and is relevant for enterprise customers. However, no smart contract audit or platform security audit from a recognised blockchain auditing firm (CertiK, Trail of Bits, Halborn, etc.) has been published, despite $30 million in Series A funding and a platform handling GPU compute workloads and a staking system.

### The risk

The founder risk has already materialised. Ahmad Shadid departed under a cloud and his subsequent project O.XYZ also faced misrepresentation allegations. Three CEOs in under two years creates leadership instability. Tory Green's move to Foundation chair suggests organisational restructuring is ongoing.

The token is down 98% from ATH, and insider unlocks began in July 2025 and continue through 2028. The IDE burn now provides some counterbalance, but its delivered pace since June is roughly 7 million IO a year, a fraction of emissions, so it cannot reverse the trend without sustained demand growth.  The harder IDE reforms announced in December 2025, demand-driven emissions and a revenue-funded buyback, are still not live as of June 2026. "If implemented" was doing heavy lifting, and for the reforms that matter it still is.

Competitively, the position is squeezed. [Akash](https://ownyourmind.ai/projects/akash/) is more decentralised, more transparent, fully open source, and cheaper on H100s. Centralised providers are more reliable and come with SLAs. io.net's differentiation (the confidential compute, the IO Intelligence API, the DePIN supply aggregation) is real but isn't yet a moat.

### My position

I don't hold IO. The closed-source core, absence of governance, founder scandal, inflated GPU metrics, and 98% price decline from ATH put this firmly in the "show me" category. The revenue growth is worth monitoring, but the decentralisation credentials are too weak for the sovereignty thesis. If you want decentralised GPU compute, [Akash](https://ownyourmind.ai/projects/akash/) is the more honest option. If you want cheap GPUs without caring about decentralisation, Lambda Labs and Vast.ai don't require a token.

## Freedom Score: 39/100

io.net scores 39/100 (F grade). Full methodology at [Freedom Score Methodology](https://ownyourmind.ai/resources/freedom-score-methodology/).

**Infrastructure decentralisation (10/20):** GPU supply comes from independent operators in what io.net says are 130-plus countries, which is architecturally distributed. Permissionless GPU onboarding is a structural positive. But the core platform (orchestration, matching, billing, API, verification) is entirely centralised and operated by io.net. The IO Worker software is closed-source binaries. The 2024 Sybil attack demonstrated that verification systems were insufficient. Its own explorer showed 2,447 devices with 1,199 active on 13 August 2026, against 30,000-plus marketed. The platform can censor or deny service to any provider or consumer.

**Governance decentralisation (2/20):** No on-chain governance. No DAO. No governance forum. No proposals. All decisions made by the io.net team and foundation board. The io.net Foundation (chaired by co-founder Tory Green) controls strategic direction. Token holders have no governance rights whatsoever. This is fully centralised governance with a "foundation" label.

**Token distribution fairness (5/15):** Insider allocation is roughly 34% (seed 12.5%, team 11.3%, Series A 10.15%). Vesting schedules exist (two to three years with 13-month cliff). The Binance Launchpool (2.5%) was accessible but required Binance KYC. Community airdrop (10%) had reasonable distribution. Not a fair launch. VC-backed at $1 billion valuation. Token declined 98% from ATH while insiders began unlocking in July 2025.

**Censorship resistance (5/15):** io.net can censor any user or provider. The platform controls account access, job allocation and the matching engine. KYC requirements for some participation. Terms of Service can restrict usage. The IO Worker binary is closed-source, so providers can't verify what code they run. The underlying Solana token is censorship-resistant for transfers. Confidential compute VMs protect data during processing. But the platform layer is a centralised censorship vector.

**Data sovereignty (8/15):** Confidential compute VMs with Intel TDX and NVIDIA H100/H200/B200 provide hardware-based data protection during processing. Users can verify attestation reports. Supports encrypted training data, model weights and architectures. Federated learning with dataset isolation is supported. SOC 2 compliance adds enterprise-grade organisational security controls. This is a notable strength. However, standard (non-confidential) workloads lack these protections, and metadata (usage patterns, billing) is held centrally by io.net.

**Open source and transparency (9/15):** Only 9 public GitHub repos containing setup scripts, binaries, docs and demo apps. Core platform code is not open source. IO-SDK (Ray fork), GPU orchestration engine, matching engine, billing system and IO Worker node software are all closed-source. No reproducible builds. The "Open Source AI Infrastructure Platform" branding on the website is misleading given the closed-source core. However, SOC 2 compliance demonstrates audited organisational security processes, Messari publishes quarterly reports with verifiable network metrics, and token economics are well documented with verifiable on-chain data (contract: BZLbGTNCSFfoth2GYDtwr7e4imWzpR5jqcUuGEwr646K on Solana).

### Path to improvement

Three changes would materially increase io.net's score:

1. **Open-source the core platform.** The IO-SDK (Ray fork), GPU orchestration engine, and IO Worker node software are all closed-source. Publishing this code under a permissive licence would address the most fundamental credibility gap. If the code is good, open-sourcing it costs nothing. If it isn't good enough to open-source, that tells you something. [Akash](https://ownyourmind.ai/projects/akash/) publishes everything and it hasn't hurt their competitive position.
2. **Implement governance.** Token holders have zero governance rights. No DAO, no proposals, no voting. Even a basic governance framework with proposal submission, quorum thresholds and on-chain voting would demonstrate that IO is more than a payment rail for a centralised company. The io.net Foundation controls all strategic decisions without community input.
3. **Extend the explorer to the metrics that still lack it.** The epoch explorer now publishes Solana-settled emissions, burns and device counts, which is progress. But the credibility gap is on the demand side: registered-versus-active GPUs, compute hours delivered, and audited revenue. Publishing those with the same on-chain rigour, rather than leaving them to self-reported quarterly figures, would rebuild trust.

## Returns Score: 54/100

IO scores 54/100 (D grade). Full methodology at [Returns Score Methodology](https://ownyourmind.ai/resources/returns-score-methodology/).

**Token utility (12/20):** IO functions as payment for GPU compute and as staking collateral for providers, which gives it more utility than most DePIN tokens. The Co-Staking Marketplace adds a passive participation layer. But the platform itself is entirely centralised. io.net controls orchestration, matching, and billing. The token is a payment rail for a company's services, not a protocol-native necessity. If io.net decided to accept USDC-only tomorrow, the token's utility would vanish overnight.

**Value accrual (10/20):** The IDE's burn went live on 1 June 2026 and settles on-chain, a step up from a whiteboard design that earns a point. But the value-capture story rests on io.net's word. In June its own explorer showed client purchases at roughly 2.6% of supplier payouts, too little to fund the burn; by September it reported about 5.8 million IO purchased against about 2.15 million burned, and the burns reconcile on chain while the purchases don't.  Until the purchase side is independently checkable, it is not the proven usage-to-token flywheel that would justify a larger bump.

**Supply dynamics (11/20):** The burn is now a deployed mechanism rather than a target: roughly 488,000 IO destroyed on-chain between 1 and 16 June 2026, with 12 million-plus a year projected; the delivered pace since is roughly 7 million a year. That earns a point over the prior score. But emissions still run at a flat per-epoch reward (10,983.8 IO in June, 10,652.7 IO on 24 September) with no sign of the demand-driven reduction the IDE promised. On io.net's own epoch totals, about 2.15 million IO burned against about 9.85 million paid in epoch rewards since 1 June, the burn offsets roughly 22% of emissions and the network stays net inflationary. Insider unlocks continue through 2028, and the token is still down 98% from ATH.

**Revenue sustainability (13/25):** The headline has come down: DePINpulse tracks io.net's annualised revenue at roughly $12.5 million in mid-2026, well below the $20 million-plus run-rate implied by Q1 2025's self-reported figures, so the growth didn't hold. io.net's own June figures were thinner still, with client IO purchases at about 2.6% of supplier payouts over the first fortnight; io.net's cumulative purchase figure has since risen to about 5.8 million IO, but it's io.net's own number and settles partly off-chain.  Treat the $12.5 million as aggregator-tracked rather than audited, and weigh it against competition from Akash and centralised providers. The two-point trim reflects the lower figure and the gap between the revenue narrative and the on-chain flows.

**Liquidity and access (8/15):** Binance listing provides baseline accessibility, and the token trades on major exchanges including Coinbase, Gate, Bybit, and KuCoin. Volume has recovered from its post-launch trough and is no longer thin relative to market cap: about a seventh of the market cap changes hands daily, spread across 59 venues, though the venue leading that flow is an obscure one rather than any of the majors. The 98% drawdown from ATH means most holders are deeply underwater. Liquidity is adequate for retail but not for institutional-scale entry or exit.

### Path to improvement

Three changes would materially increase io.net's returns score:

1. **Ship the rest of the IDE.** The burn is live and on-chain, which is the easy part. Demand-driven emissions and a checkable revenue buyback are the reforms that matter: emissions are still a flat per-epoch drip, and the purchases io.net says fund the burn aren't on chain. Turning those on would prove the team can execute on tokenomics, beyond the launch graphics.
2. **Demonstrate net deflationary months.** Publishing months where burns exceed new emissions would be the single most powerful signal for token holders. On io.net's own epoch totals the burn has offset roughly 22% of epoch rewards since 1 June 2026, so the network is still net inflationary.  The maths needs to net out in the public burn ledger itself, since a 12-million-a-year projection is a forecast rather than a delivered reduction.
3. **Accelerate the insider unlock schedule or commit to no further raises.** Two to three more years of continuous insider selling at current prices will suppress any organic price recovery. A compressed unlock with a commitment to no further dilutive funding rounds would signal confidence in the revenue trajectory.

## Score change log

| Date | Score | Change | Reason |
| - | - | - | - |
| 2026-10-13 | Data | N/A | Frontmatter turnover restated from about a seventh to about a fifth of market cap (7-day median 19.7%, 13 October). Scores unchanged. |
| 2026-10-04 | Editorial | N/A | Product-in-practice sweep: scoped the burn reconciliation to the 372 of 957 epochs indexed, dated the OpenRouter claim, attributed the revenue figure to DePINpulse, and made inference growth io.net's report. Scores unchanged. |
| 2026-09-25 | Data | N/A | The burn now offsets about 22% of epoch rewards since 1 June on io.net's own explorer totals, not a third; the June purchase-to-payout ratio is attributed to io.net rather than called on-chain. Scores unchanged. |
| 2026-09-24 | Data | N/A | Explorer inventory refreshed to the 23 Sep integrity reading (2,976 devices, 1,253 active); the gap to the 30,000+ headline is now about 10x. Scores unchanged. |
| 2026-09-24 | Data | N/A | IDE prose aligned to the current ide_burn_total match and purchase finding: io.net now reports ~5.8M IO purchased vs ~2.15M burned, so the June "emission-funded" read is dated; burn pace (~7M/yr) and epoch reward (10,652.7) updated; country count attributed. Scores unchanged. |
| 2026-09-23 | Data | N/A | Integrity records re-cut: 4 claims now quote io.net's own pages verbatim and 6 are our own measurements. Scores unchanged. |
| 2026-09-19 | Data | N/A | `ide_client_io_purchased` moved unverified → established: each io.net epoch's reported burn reconciles to six decimals against the SPL burn in its on-chain settlement transaction, 39 of 39 indexed. Purchases stay a reported figure with a verified floor. Scores unchanged. |
| 2026-09-18 | Data | N/A | `ide_burn_total` verified on chain: the mint's supply reduction from 800M implies 2,149,437 burned against io.net's reported 2,146,039, a 0.16% match. Scores unchanged. |
| 2026-09-06 | Data | N/A | WS3b liquidity rail: the still-thin-relative-to-market-cap line is inverted. Turnover is a 13.86% 7-day median across 73 tickers on 59 venues, with Aivora Exchange top at 13.5% and Binance second. Corrected on both surfaces. Liquidity & Access 8/15 held. |
| 2026-08-15 | Data | N/A | Circulating supply de-hardcoded and bound. The stated 43.31% (June 2026) was 4.4pp stale against 47.69% of max supply. |
| 2026-08-14 | Data | N/A | Labelled all 10 of the largest IO holders, which the 2026-08-13 read had left measured but unclassified. Exchange custody is 14.64% of supply (Binance, Coinbase), leaving 44.17% project-held across two "ocean foundation" Squads vaults and the vesting apparatus. Distribution fairness stays 5/15. |
| 2026-07-06 | Editorial | N/A | Added a "The product in practice" section (per `specs/product-comparison-framework.md`) covering the 2024 GPU-count credibility overhang and the pivot from cluster rental to OpenRouter inference; rental revenue is self-reported. No score change. |
| 2026-06-17 | Returns | 54 → 54 | Net-neutral re-composition after the IDE burn went live on-chain (1 June 2026). Value Accrual 9→10 and Supply Dynamics 10→11 (burn now a deployed, Solana-settled mechanism, ~488K IO burned by 16 June, 12M+/yr projected); Revenue Sustainability 15→13 (annualised revenue reconciled to ~$12.5M per DePINpulse from $20M+, and on-chain client purchases at ~2.6% of supplier payouts contradict the revenue-funded-burn framing). Prompted by the @ionet "IDE is live" campaign. |
| 2026-06-17 | Freedom | 38 → 39 | Open-source transparency 8→9: io.net's epoch explorer now publishes on-chain, Solana-settled per-epoch emission and burn data, partly closing the prior "no independent verification tools" gap. Core platform still closed-source. |
| 2026-08-13 | Data | N/A | io.net publishes two GPU counts at once: "30,000+" on the homepage and "320,000+" in the /cloud page structured data. Its own explorer read 2,447 devices with 1,199 active the same day, so the gap is ~12x or ~130x depending which headline you take. Infrastructure stays 10/20 pending the monthly. |
| 2026-08-13 | Data | N/A | Correction to the row above, same day: an earlier version credited io.net with cutting its headline tenfold to 30,000+. The 320,000+ figure is still live on /cloud, so this is inconsistency, not a correction on their part, and the credit was withdrawn. |
| 2026-08-13 | Data | N/A | Migrated io.net onto the verdict spine: the GPU-count and supply checks now anchor to Infrastructure and Supply Dynamics, plus new on-chain verdicts for the fixed SPL supply, the absent freeze authority, holder concentration and the open-source claim (inflated). |
| 2026-06-17 | Data | N/A | IDE status corrected: burn component live 1 June 2026 (on-chain), but demand-driven emissions and the Fee Vault revenue buyback are not live per io.net's own explorer (emissions flat at 10,983.8 IO/epoch; burn is emission-funded). Revenue figure reconciled from "$20M+" to ~$12.5M annualised (DePINpulse, matches research JSON). updatedDate bumped. |
| 2026-06-02 | Data | N/A | Circulating supply corrected from 37.7% (~301M, March 2026) to 43.31% (~346M of 800M) per CoinGecko, reflecting continued vesting/emission unlocks. IDE tokenomics still not live (Q2 2026 target); no score change. June monthly review. |
| 2026-04-06 | Data | N/A | Updated date only. GPU count already reflected 327,000+. No score changes (IDE not yet live). |
| 2026-03-12 | Returns | 49 → 54 | IDE tokenomics overhaul, Q1 2025 revenue $5.7M (82.6% QoQ growth), co-staking marketplace launch. |
| 2025-03-06 | Both | N/A | Initial publish. Freedom 40/100, Returns 49/100. |