Crypto Gloss

io.net

io.net launched in 2023 amid the convergence of two major trends: the GPU shortage caused by AI demand, and the DePIN movement in crypto (Decentralized Physical Infrastructure Networks). The thesis was simple but powerful: millions of GPUs sitting idle in crypto mining rigs, data centers with spare capacity, and even consumer gaming PCs could collectively provide AI companies with GPU compute at significantly lower prices than AWS, Google Cloud, or Azure. By using blockchain-based coordination and the IO token to incentivize supply-side GPU operators, io.net aimed to undercut centralized cloud by 70-90% in cost while providing flexible, on-demand GPU cluster access. The system is built on Solana for fast settlement and low transaction fees.

How io.net Works

Supply side (GPU operators):

  1. Operators install the io.net software on machines with eligible GPUs
  2. The io.net worker agent reports GPU specs, availability, and location to the network
  3. When a job is matched, the operator’s GPU is allocated and earns IO tokens
  4. Earnings depend on GPU model (H100 > A100 > A6000 > RTX 3090 etc.), uptime, and job duration

Demand side (AI/ML teams):

  1. Teams specify GPU type, quantity, and duration via io.net’s cluster deployment interface
  2. io.net’s matching engine assembles a cluster from available GPUs across operators
  3. Payment is in IO tokens or USD equivalent (abstracted for enterprise clients)
  4. Clusters are deployed using Ray (distributed computing framework) for ML workloads

Coordination layer (Solana):

  • Job matching, operator reputation, payment settlement happen on-chain
  • Fast finality and low fees make Solana practical for high-frequency settlement
  • On-chain reputation scores for operators based on uptime and job completion

GPU Hierarchy and Pricing

io.net supports multiple GPU tiers, with pricing approximately:

GPU Use Case Relative Earnings
NVIDIA H100 LLM training, frontier AI Highest
NVIDIA A100 Training, inference Very high
NVIDIA A6000 Training, rendering High
NVIDIA RTX 4090 Consumer-grade ML Medium-high
NVIDIA RTX 3090 Consumer-grade ML Medium
AMD equivalents Gaming/consumer Lower

vs. Centralized Cloud:

  • AWS H100 on-demand: ~$98/hour per GPU
  • io.net H100: ~$2.50–$6/hour per GPU (at launch pricing)
  • Discount: 70-90% — the core value proposition

IO Token

Network: Solana

Total supply: 800,000,000 IO

Launch: 2024 (TGE with airdrop to early GPU suppliers and users)

Utility:

  • Payment for GPU compute on the network
  • Staking by GPU operators as collateral/reputation bond
  • Governance over network parameters (pricing, accepted GPU types, slashing conditions)

Distribution (approximate):

  • ~40% earmarked for GPU supply-side rewards over time
  • ~15% team and advisors (vested)
  • ~15% investors (vested)
  • ~12% foundation/ecosystem
  • ~8% community/airdrop at launch

DePIN Context

io.net is part of the broader DePIN (Decentralized Physical Infrastructure Networks) movement:

DePIN thesis:

Physical world infrastructure (compute, storage, wireless, energy, sensors) can be decentralized using crypto token incentives:

  • Tokens incentivize individuals to contribute physical resources
  • Network aggregates resources into a marketplace
  • Clients benefit from competition and lower prices vs. oligopoly (AWS, AT&T, etc.)

Other DePIN networks:

  • Helium (wireless) — decentralized LoRaWAN and 5G
  • Filecoin/Arweave (storage) — decentralized file storage
  • Render Network (GPU rendering) — GPU compute for 3D rendering & AI
  • Akash Network (cloud compute) — decentralized Kubernetes/cloud
  • io.net (GPU clusters for ML) — concentrated on ML/AI workloads specifically

Challenges and Risks

Supply quality control:

Ensuring GPUs are actually available, performing as claimed, and reliably maintained is difficult at decentralized scale. io.net uses benchmarking and proof-of-work verification, but quality inconsistency remains a challenge.

Enterprise trust:

AI companies building production infrastructure are sensitive to reliability. A single GPU failure in a training cluster can corrupt hours of work. Many AI teams prefer guaranteed SLAs from AWS/GCP despite higher cost.

Token inflation:

Ongoing IO rewards for GPU operators create constant sell pressure as operators convert earnings to fiat to pay electricity/hardware costs.

Competition:

Akash Network, Render Network, Gensyn, Prime Intellect, and others are all tackling overlapping parts of the decentralized compute market.

How to Participate in io.net

As a GPU operator:

  1. Go to io.net and install the worker agent on a machine with a supported NVIDIA/AMD GPU
  2. Connect a Solana wallet to receive IO rewards
  3. Maintain high uptime to build reputation score

As a compute buyer:

  1. Go to io.net and create an account
  2. Specify GPU requirements, cluster size, and duration
  3. Pay using IO tokens or credit card (fiat abstracted)

To hold IO tokens:

  • Purchase IO on Solana DEXes (Jupiter, Raydium) or CEXes after TGE
  • Start with to get SOL/USDC
  • Store tokens:

Social Media Sentiment

io.net generated significant hype in 2023-2024 at the intersection of two of crypto’s hottest narratives: AI and DePIN. The project attracted well-known backers (Multicoin Capital, Hack VC, IoTeX) and accumulated substantial social media following. Real-world usage metrics were scrutinized heavily — GPU count claimed on the network, actual utilization rates, and verifiable AI team usage were all questioned. A critical incident in early 2024 involved questions about the accuracy of reported GPU counts (some concerned the numbers were inflated). The team addressed these concerns with improved verification mechanisms. The broader thesis — that decentralized GPU supply can meaningfully challenge AWS pricing for AI workloads — is real and validated by market conditions, but execution at scale is the open question. IO is widely traded as an AI + DePIN exposure token regardless of actual compute utilization metrics.

Last updated: 2026-04

Related Terms

Sources

Stoica, I., Morris, R., Liben-Nowell, D., Karger, D. R., Kaashoek, M. F., Dabek, F., & Balakrishnan, H. (2001). Chord: A Scalable Peer-to-Peer Lookup Service for Internet Applications. ACM SIGCOMM Computer Communication Review.

Ben-Or, M., Goldwasser, S., & Wigderson, A. (1988). Completeness Theorems for Non-Cryptographic Fault-Tolerant Distributed Computation. Proceedings of the 20th Annual ACM Symposium on Theory of Computing.

Shafi, A., Baker, M., & Carpenter, B. (2006). Phalanx: A Grid Computing Architecture for Scalable Parallel Computing. Proceedings of IEEE International Conference on Cluster Computing.

Bernstein, P. A., Hadzilacos, V., & Goodman, N. (1987). Concurrency Control and Recovery in Database Systems. Addison-Wesley.

Antonopoulos, A., & Wood, G. (2018). Mastering Ethereum. O’Reilly Media.