On October 1, 2024, as Bitcoin held steady near $60,837 and the broader cryptocurrency market capitalization hovered around $2.4 trillion, a quieter but equally significant development unfolded at the intersection of artificial intelligence and blockchain technology. Ankr, a leading Web3 infrastructure provider, announced its partnership with Ungate to operate nodes for a new EigenLayer Actively Validated Service (AVS) designed specifically for decentralized AI workloads. This collaboration represents a tangible step toward what many are calling the “decentralized AI supermesh.”
The Synergy
The convergence of AI and blockchain addresses a fundamental tension in the technology landscape. Modern AI models demand enormous computational resources, currently concentrated in the hands of a few large corporations. Blockchain-based networks offer an alternative: distributed, permissionless compute infrastructure where participants contribute resources and are rewarded transparently through cryptographic incentives.
The Ankr-Ungate partnership exemplifies this synergy. Ungate’s platform, built on EigenLayer’s shared security model, provides an “InfiniRoute” system — an AI-powered model orchestrator that routes computational tasks to the best available models and resources in real time. Rather than relying on a single provider like OpenAI, the system dynamically selects and combines multiple models to deliver faster, cheaper, and often more accurate responses than any individual model could achieve alone.
AI Use Cases in Web3
The InfiniRoute system opens several compelling use cases at the AI-blockchain intersection. Swarm intelligence applications allow developers to build AI operating system modules and models that harness the collective intelligence of distributed agents, datasets, and infrastructure. This approach mirrors how decentralized finance composability created entirely new financial products — except here, the building blocks are AI models rather than financial primitives.
Enhanced efficiency emerges as another key benefit. By routing requests to specialized models optimized for specific tasks, the network avoids the computational overhead of running a single massive model for every query. This translates to lower costs for developers and end users while maintaining or improving output quality.
For the broader DePIN (Decentralized Physical Infrastructure Networks) sector, this represents continued maturation. Projects like Render Network for GPU rendering, Akash Network for cloud computing, and Bittensor for decentralized machine learning have been building the foundational infrastructure. The Ungate AVS adds a coordination layer that could make these resources more accessible and interoperable.
Data Privacy Implications
Perhaps the most compelling aspect of decentralized AI networks lies in their approach to data privacy. Ungate’s federated learning model enables AI training and inference without centralized data custody — a critical differentiator as regulatory scrutiny of AI companies’ data practices intensifies globally. In a zero-data-custody environment, sensitive information never leaves the user’s control, yet the system can still produce useful AI outputs.
This stands in stark contrast to the centralized AI model, where user data flows through corporate servers, creating honeypots of personal information vulnerable to breaches and misuse. For an industry still reeling from $120 million in September 2024 hacks, the security implications of decentralized data handling resonate deeply.
The Innovation Frontier
The EigenLayer AVS framework itself represents an important technical innovation. By allowing protocols to tap into Ethereum’s validator set for security without launching their own blockchain, it dramatically lowers the barrier to entry for new decentralized applications. This shared security model, combined with Ankr’s proven infrastructure operations, creates a reliable foundation for AI workloads that demand both high performance and trustless verification.
Looking ahead, the convergence of AI and crypto is likely to accelerate. As AI models grow larger and more expensive to train, the economic incentives of decentralized compute become increasingly attractive. The market has already signaled interest, with AI-focused tokens gaining significant attention throughout 2024.
Concluding Thoughts
The partnership between Ankr and Ungate is not just another blockchain collaboration — it is a glimpse into how artificial intelligence and decentralized networks might evolve together. By combining shared security, distributed compute, and intelligent routing, these systems address real limitations of centralized AI infrastructure. As the technology matures and more developers build on these foundations, the AI-blockstacking convergence could reshape both industries in ways we are only beginning to understand. The pieces are being put in place, one node at a time.
Disclaimer: This article is for informational purposes only and does not constitute financial advice. Always conduct your own research before making investment decisions.
Ankr running Ungate AVS nodes on EigenLayer is a nice validation of the restaking model for something beyond ETH staking
restaking for AVS nodes is one of the few non-speculative use cases for EigenLayer so far. actual compute being secured by restaked ETH is a real utility signal
non-speculative EigenLayer use case and its barely discussed. restaked ETH securing actual compute workloads is the real pitch
Ankr running EigenLayer AVS nodes for Ungates AI workloads while BTC sat at 60K. the convergence was theoretical in 2024, now its actual infrastructure shipping
decentralized AI supermesh sounds cool but the article skips over latency. distributed compute for inference has real speed challenges compared to centralized
thats a fair point on latency. but for training workloads where you can parallelize across nodes, geography matters less than for real-time inference
gpu_lord parallelization across heterogeneous nodes is a nightmare for training. different GPU models mean different batch sizes mean stragglers everywhere
batch_size_64 heterogeneous GPU clusters for training is a real nightmare. wonder how Ankr handles checkpoint sync across different hardware
the latency issue is real but solvable for batch inference. nobody is suggesting decentralized compute for real-time trading bots. it is for training and heavy workloads where 200ms extra does not matter
ines making the latency point well. training workloads can handle distributed nodes but inference needs low latency
compute_gw training workloads can handle distributed nodes yes but synchronization overhead kills your throughput advantage. nobody mentions this in the AVS pitch decks
restake_bear synchronization overhead is exactly why distributed training across heterogeneous nodes has been stuck in research papers for 5 years. gradient averaging across different GPU memory profiles is a nightmare
Ines M. makes the right point about batch inference. 200ms latency doesnt matter for training workloads but kills any real-time use case instantly
the InfiniRoute system is basically load balancing for decentralized compute. neat idea but EigenLayer slashing conditions dont account for GPU failure rates which are way higher than validator downtime
gpu_routing_ EigenLayer slashing not accounting for GPU failure rates is a massive blind spot. validator downtime is predictable, hardware failures are chaotic
BTC at 60K and total market cap at 2.4T when this dropped. everyone focused on price while Ankr quietly built the only decentralized compute marketplace with actual enterprise clients
Ankr providing actual infrastructure for AI workloads instead of just another liquid staking wrapper is refreshing. the Ungate partnership could set a template for other restaking AVS launches
ankr providing real infra for ai workloads instead of another staking wrapper is actually refreshing
if Ungate works it validates the whole AVS model beyond liquid staking derivatives. huge for EigenLayer if it actually ships
mara is right. if ungate actually ships it proves avs works for real compute not just staking derivatives
EigenLayer slashing for compute failures is going to be brutal. a single bad GPU node corrupts a training run and the validator gets slashed. hardware failure rates are not predictable like validator downtime
Ankr running actual AVS infrastructure instead of issuing another staking derivative is genuinely refreshing. more EigenLayer AVS launches need to look like this
Ankr Ungate AVS running decentralized inference is interesting but the latency vs AWS GPU instances must be significantly worse. anyone seen benchmarks
Hannelore K. latency on Ankr AVS is roughly 3-5x worse than AWS for inference workloads. fine for batch training, unusable for anything real-time. the InfiniRoute routing layer adds overhead nobody benchmarks
Ankr running EigenLayer AVS nodes for actual GPU workloads is the only real restaking use case beyond staking derivatives. everything else is just ETH doubles counting
avs_dispatch_ the InfiniRoute load balancing is clever but EigenLayer slashing conditions dont account for GPU failure rates which are way higher than validator downtime. different hardware different risk model