November 8, 2023 marks a pivotal moment in the convergence of artificial intelligence and cryptocurrency. Two major developments — Ritual’s emergence from stealth with a $25 million Series A funding round and io.net’s deployment of 107,000 repurposed cryptocurrency mining GPUs for AI workloads — signal that decentralized AI infrastructure is rapidly moving from concept to reality. With Bitcoin trading at $35,655 and the broader market showing renewed strength, the AI-crypto intersection is attracting both capital and computing power at an unprecedented scale.
The Synergy
The simultaneous emergence of Ritual and io.net represents two complementary approaches to the same fundamental problem: the centralization of AI computing power. Today, a handful of large technology companies control the majority of AI infrastructure, from training compute to model deployment. This concentration creates bottlenecks in access, inflates costs, and introduces single points of failure that affect the entire AI ecosystem.
Ritual addresses this challenge by building a decentralized network for AI model execution, while io.net tackles the supply side by aggregating GPU computing resources from underutilized sources. Together, they form the backbone of what could become a fully decentralized AI stack — from raw compute to model inference — all coordinated through blockchain incentives.
AI Use Cases in Web3
Ritual’s core product, Infernet, enables smart contracts to integrate AI models directly into their execution logic. This capability unlocks a range of use cases that were previously impractical. Lending protocols can automatically adjust risk parameters based on real-time market conditions analyzed by AI models running on Ritual’s network. Decentralized exchanges can implement AI-powered dynamic pricing that responds to order flow patterns and liquidity shifts with greater precision than traditional algorithms.
The implications extend beyond DeFi. Content moderation on decentralized social platforms can leverage AI models running on Ritual to identify and filter harmful content without relying on centralized APIs. Gaming protocols can implement dynamic difficulty adjustment and procedural content generation powered by on-chain AI inference. Each of these applications requires reliable, low-latency access to AI models — exactly what Ritual’s decentralized infrastructure aims to provide.
Data Privacy Implications
The decentralized AI paradigm raises important questions about data privacy. When AI models are executed across a distributed network of nodes, ensuring that sensitive input data remains private becomes a critical concern. Ritual’s architecture must grapple with the challenge of enabling useful AI inference while preventing node operators from accessing the raw data being processed.
This is where zero-knowledge proofs and secure multi-party computation become essential. By combining AI inference with cryptographic privacy guarantees, decentralized AI networks can offer the best of both worlds: the computational power of distributed systems and the privacy assurances that users and enterprises demand. The Biden administration’s recent executive order on AI safety, which emphasizes privacy-preserving techniques, adds regulatory weight to this requirement.
The Innovation Frontier
io.net’s deployment of 107,000 repurposed mining GPUs represents a creative solution to the GPU shortage that has constrained AI development worldwide. Cryptocurrency miners who invested heavily in GPU hardware during previous bull markets now have an alternative revenue stream that extends the useful life of their equipment. The partnership with the Render network enables a seamless marketplace where GPU owners can offer their computing resources to AI developers.
This model of repurposing crypto infrastructure for AI workloads could prove transformative. As Ethereum’s transition to proof-of-stake continues to reduce demand for GPU mining, thousands of data centers filled with capable hardware need new use cases. AI inference and training represent the perfect match — computationally intensive tasks that can be distributed across geographically diverse nodes.
Concluding Thoughts
The developments of November 8, 2023 suggest that the AI-crypto intersection is entering a new phase of maturity. Ritual’s $25 million funding round, backed by investors including Balaji Srinivasan, Accel, and Archetype, demonstrates that serious capital is flowing into decentralized AI infrastructure. Io.net’s massive GPU deployment proves that the physical infrastructure to support these networks already exists and can be rapidly mobilized.
As these networks grow and interconnect, the vision of decentralized AI — where computing power, model training, and inference are distributed across a global network of independent operators — moves closer to reality. The implications for both the AI and cryptocurrency industries are profound, with the potential to democratize access to AI capabilities while creating new economic opportunities for crypto infrastructure operators.
Disclaimer: This article is for informational purposes only and does not constitute financial or investment advice. Always conduct your own research before making any financial decisions.
107k repurposed mining gpus for ai workloads. finally a use case for all that hardware after the merge killed eth mining
most of those mining GPUs are older RTX 30 series though. fine for inference but you are not training anything serious on them
most repurposed mining gpus are rtx 30 series good for inference
Chen Wei exactly. RTX 30 series for inference work is fine but lets not pretend this competes with H100 clusters for training. different use cases entirely
Chen Wei RTX 3090s are actually solid for 7B model inference. not training anything big but for serving Llama derivatives they work fine
Ritual raising 25M for decentralized AI inference is interesting but the real bottleneck is model size. Most useful models barely fit on centralized clusters.
ritual 25m raise targets the decentralized inference gap
Priya R. the model size bottleneck is exactly why Ritual targets inference not training. serving a quantized 13B on distributed consumer GPUs is already doable
the convergence play is where the money is going to flow next cycle. ai plus crypto is the meta
decentralized ai inference is the low hanging fruit. training will stay centralized for a while but serving models on distributed gpus is already viable
io.net repurposing 107000 mining GPUs for AI workloads is the most practical pivot in crypto mining history. actually useful compute instead of hashing puzzles
Adesina O. 107k GPUs from mining to ML inference is genuinely smart. Ritual handling the smart contract layer while io.net provides raw compute is a clean split
107k GPUs repurposed from mining to AI inference is nice but Ritual targeting 13B models means they are competing with free Hugging Face endpoints
107k repurposed mining GPUs sounds impressive until you realize most are worn out from 24/7 mining. failure rates on those cards must be brutal
Yumi K mining GPUs ran at 90C+ for 2 years straight. the VRAM degradation on those repurposed 3090s is no joke, ive had 3 cards die mid inference job
3 dead 3090s from mining repurposed for inference is brutal. the thermal cycling on gDDR6X memory after 2 years of 24/7 mining is basically a ticking time bomb
vram_skeptic_ 3 dead 3090s is rough. mining thermal cycling on GDDR6X is basically planned obsolescence in reverse
Hana P. thermal cycling on mined 3090s is bad but the bigger issue is VRAM capacity. you need 80GB+ for 70B inference and consumer cards max at 24GB. io.net is stitching together a puzzle that doesnt fit
inference_cost_void_ 80GB VRAM for 70B is right but most inference workloads are 7B and 13B quantized models. consumer 3090s handle those fine despite the wear
trying to shard a 70B parameter model across random GPUs sounds cool until you hit the latency wall. io.net needs way better networking
70B across distributed GPUs with 50-100ms ethernet latency is rough. but quantized 7B and 13B models work fine on single 3090s repurposed from mining. Ritual is targeting the right tier
n00b_ml latency is a real problem but io.net added RDMA networking for clustered deployments. early but the benchmarks look promising for multi-GPU inference
cluster_mesh_ RDMA on distributed consumer hardware is a stretch. you need IB or RoCE for that to work properly. ethernet latency across random datacenters will kill you
rdma_skeptic you need RoCEv2 at minimum for distributed inference across consumer nodes. plain ethernet adds 50-100ms and thats game over for any interactive workload
Ritual raised 25M and shipped a whitepaper. io.net deployed 107000 GPUs and still cant beat AWS on a single inference benchmark. the decentralized AI thesis needs actual revenue not just node counts
Ritual shipped Infernet but the real test is whether developers actually use it over just calling OpenAI API. decentralization needs to beat centralized on price and latency