Skip to main content

AI/Compute

AI/Compute

AI/Compute groups projects that provide computational resources for AI workloads — the GPU and infrastructure layer of crypto AI. These are decentralized compute networks, GPU marketplaces, and orchestration platforms that let anyone contribute or consume processing power for training and running AI models.

In the DecAIHub registry, AI/Compute is one of the largest categories and the one most closely tied to physical infrastructure. The core promise is straightforward: instead of renting GPUs from a single cloud provider, users access a distributed network of compute suppliers coordinated by a blockchain protocol.

Workload types matter when reading cards: some networks emphasize training (long-running, high-memory jobs), others inference (latency-sensitive serving), and many sit in between with general-purpose job queues. The AI/Compute label applies when the positioning and artifacts describe compute for AI — not merely generic hashing or unrelated workloads.

In short: AI/Compute covers projects that supply, orchestrate, or marketplace computational resources for AI tasks.

What it means

Understanding the difference between compute-as-resource and compute-as-AI is critical when reading project cards in this category.

Compute in the AI context refers to the processing power needed to train models, run inference, or handle data pipelines. Traditional AI relies on centralized cloud providers (AWS, GCP, Azure). AI/Compute projects attempt to decentralize this by building networks where independent providers offer GPUs, CPUs, or specialized accelerators, and a protocol layer handles discovery, scheduling, pricing, and settlement.

The blockchain's role in these projects varies widely: from simple payment rails to on-chain proof-of-work verification and reputation systems for compute providers.

What separates genuine decentralized compute from marketing is measurable evidence: real GPU providers serving real jobs, verifiable utilization metrics, and on-chain settlement of completed work. Typical AI components found in AI/Compute projects include GPU orchestration engines, job schedulers, proof-of-compute protocols, compute marketplaces, and resource-matching algorithms. Some projects focus on raw GPU rental; others add higher-level abstractions like model-aware scheduling or federated training coordination.

Common subcombinations

DecAIHub uses composite categories. Below are the most frequent combinations within AI/Compute. Infra is the dominant pairing; overlaps with DePIN, AI/Training, and AI/Inference describe how the same network positions supply and demand.

Subcombination What it means
AI/Compute | Infra The largest group: GPU networks, compute orchestration layers, and decentralized cloud platforms.
AI/Compute | DePIN Physical GPU and compute infrastructure with token incentives — hardware providers earn tokens for contributing resources.
AI/Compute | AI/Training Compute specifically provisioned for model training workloads (large-scale, long-running).
AI/Compute | AI/Inference Compute specifically provisioned for running trained models in production (latency-sensitive, high-throughput).

A single project can carry multiple domain tags (e.g., AI/Compute | Infra | DePIN).

What to look for in a project card

When reading an AI/Compute card on DecAIHub, pay attention to:

  • AI component — does the card list a concrete AI component (GPU orchestration, job scheduling, proof of compute, compute marketplace), or only generic "decentralized compute" messaging?
  • Evidence of real supply — are there Tier-1 artifacts showing actual GPU providers, measurable utilization, or on-chain proofs of completed jobs?
  • Onchain necessity (onchain_score) — does the blockchain do more than issue tokens? Look for on-chain job matching, settlement, slashing, or verifiable compute proofs. See AI reality, Onchain necessity and Token necessity.
  • AI reality score (ai_score) — providing GPUs is infrastructure, not AI by itself. Does the project have its own AI logic, or does it only host other people's workloads?

What this category alone does not prove

A project labeled AI/Compute has been classified as compute-focused based on its positioning and available artifacts. The label alone does not confirm that:

  • there are real GPU providers actively serving workloads on the network;
  • the compute is actually being used for AI rather than sitting idle;
  • the AI built on top of the compute is real, proprietary, or useful;
  • the decentralized approach offers meaningful advantages over centralized alternatives;
  • the token is necessary for the compute marketplace to function.

Always check the full DecAI Fit score and the evidence block in the project card.

Do not confuse with

  • AI/Inference — focuses on the act of running a model and producing a result; compute is the resource that powers inference, but the two categories address different layers. See AI/Inference.
  • AI/Training — focuses on building and improving models; training requires compute but is a distinct activity. See AI/Training.
  • AI/Agents — builds autonomous software that decides and acts; agents may consume compute, but the category is about behavior, not resources. See AI/Agents.
  • DePIN without AI — some DePIN projects provide storage, bandwidth, or sensors rather than GPU compute for AI. The AI/Compute label applies only when the compute targets AI workloads.

Frequently asked questions

If a project sells GPU time, why does ai_score sometimes look low?

Supplying hardware is infrastructure. The AI reality score rewards verifiable AI-specific logic (schedulers tuned for models, proof systems tied to inference, and so on). Raw rental without described AI-oriented software may score differently from a stack that documents ML-aware orchestration.

How is AI/Compute different from AI/Inference on a card?

AI/Inference emphasizes serving models and outputs. AI/Compute emphasizes who provides the machines and how jobs are matched and settled. A single project can touch both; read the composite tags and the AI component field to see which layer the team stresses.

Does “decentralized” always mean many independent providers?

Not necessarily from a label alone. Check evidence: dashboards, docs, or on-chain signals that show diverse providers and completed work — not only protocol whitepapers.

External references

See also

No comments to display

Back to top