Artificial intelligence has asserted itself as one of the dominant forces in technology throughout 2023, fueled by the explosive growth of large language models and generative AI applications. But behind the headlines about ChatGPT and Midjourney, a quieter revolution is unfolding at the intersection of AI and blockchain technology — one where decentralized networks are proving they can compete with traditional cloud infrastructure for the most demanding computing workloads. At the center of this convergence stand two projects that are rapidly redefining what decentralized infrastructure can achieve.
The Synergy
The connection between AI and decentralized computing is rooted in a fundamental economic reality: AI training and inference require enormous computational resources, and those resources are becoming increasingly scarce and expensive. NVIDIA, the chipmaker whose GPUs are essential for AI workloads, has seen its stock surge nearly 225% over the past year, pushing its market capitalization past one trillion dollars. This scarcity creates an opening for alternative providers — and decentralized networks are uniquely positioned to fill it.
As Bitcoin trades near $39,500 and the broader crypto market shows renewed institutional interest, the DePIN narrative — decentralized physical infrastructure networks — has emerged as one of the most compelling use cases for blockchain technology. Rather than building another trading platform or speculative token, DePIN projects are addressing real-world infrastructure gaps by creating marketplaces where anyone with computing resources can offer them to anyone who needs them.
AI Use Cases in Web3
Akash Network has positioned itself as a decentralized cloud computing marketplace where users can access high-density GPUs like the NVIDIA A100 and H100 — the same chips powering AI research at major technology companies. According to Overclock Labs founder Greg Osuri, Akash has seen remarkable growth driven directly by the AI boom. The AKT token has surged approximately 550% over the past year, rising from $0.25 to $1.65, though it remains well below its all-time high of $8.07 set in April 2021.
Perhaps the most significant milestone has been the successful training of foundational AI models entirely on Akash decentralized infrastructure. This achievement challenged the prevailing assumption that training large models requires the low-latency, tightly-coupled GPU clusters found in traditional data centers. The implications extend beyond mere technical validation — it suggests that the future of AI compute could be distributed rather than concentrated in the hands of a few cloud giants.
Daily spending on Akash compute resources has surged from approximately $70 per day to around $2,000 — a nearly 30x increase that reflects growing demand for GPU access outside traditional cloud environments. This growth trajectory aligns with the broader trend of organizations seeking alternatives to the walled gardens operated by Amazon Web Services, Microsoft Azure, and Google Cloud.
Livepeer, the decentralized video infrastructure network, is pursuing a complementary vision. CEO and founder Doug Petkanics has identified several AI-powered capabilities that align naturally with Livepeer core video processing architecture: automatic caption generation, content moderation screening, and AI-driven product identification within video streams. These features leverage the same distributed computing model that Livepeer uses for video transcoding, creating new revenue streams for network operators while reducing costs for developers.
Data Privacy Implications
The shift toward decentralized AI computing carries significant privacy implications. When organizations process data on centralized cloud platforms, they surrender control over where and how that data is stored, processed, and potentially accessed. Decentralized networks offer a different model: data can be processed across distributed nodes without being concentrated in a single jurisdiction or under the control of a single corporation.
The video streaming industry, valued at approximately $200 billion, illustrates this tension acutely. Traditional cloud providers charge around $3 per hour of ingested video stream, a cost structure that reflects the expense of maintaining idle server capacity worldwide. Livepeer distributed model could fundamentally alter this economics by pooling underutilized computing resources from operators around the globe.
For AI workloads specifically, the privacy advantages are compelling. Organizations training models on sensitive data — healthcare records, financial information, personal communications — face regulatory requirements in jurisdictions around the world. A decentralized computing marketplace allows them to process data closer to its source, potentially simplifying compliance with data residency requirements.
The Innovation Frontier
Both Akash and Livepeer are featured among the 16 nominees for Crypto Project of the Year, a recognition that reflects the growing maturity of the DePIN sector. But the innovation extends beyond individual projects. The broader ecosystem of decentralized compute is creating new primitives: on-chain treasuries for funding development, grant programs for ecosystem builders, and governance mechanisms for coordinating infrastructure upgrades.
Livepeer establishment of an on-chain treasury this year represents a particularly interesting development. Community members can submit proposals for grants to develop AI-related features, creating a self-sustaining funding mechanism that does not depend on venture capital or corporate sponsorship. This model could serve as a template for other decentralized infrastructure projects seeking to bootstrap development in specific technical domains.
Concluding Thoughts
The convergence of AI and decentralized computing is no longer theoretical. Projects like Akash and Livepeer are demonstrating that distributed networks can handle the most demanding computational workloads while offering meaningful advantages in cost, privacy, and censorship resistance. As GPU scarcity continues to intensify and AI adoption accelerates across every industry, the demand for decentralized alternatives will only grow. The projects that can deliver reliable, cost-effective compute at scale will be well positioned to capture a significant share of the expanding AI infrastructure market.
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.
NVIDIA up 225pct and Akash still trading at a fraction of AWS spend. the decentralized compute thesis is structurally correct but adoption is painfully slow
NVIDIA 225% stock surge and somehow Akash is supposed to compete with their pricing power. GPU supply chain dominance doesnt get disrupted by a DAO
NVIDIA trillion dollar valuation and decentralized compute is still a rounding error. the thesis is early but the demand curve is undeniable
nvidia up 225% in a year and were wondering why gpu compute costs are exploding. akash charging 30% less than aws for the same workload makes total sense now
30% less than AWS sounds great until you factor in reliability. akash uptime is getting better but its not parity yet for production workloads
Sebastian F. calling 30pct savings great until your job fails at 3am is the most accurate take here. Akash needs an SLA layer before production workloads migrate
30% cheaper than AWS sounds great until your job fails at 3am and theres no support team to call. decentralized compute needs an SLA layer
The 225% NVIDIA stock surge does explain why gpu compute costs are exploding but I doubt Akash’s 30% savings versus AWS can deliver consistent performance at scale.
gpu_economist underestimates how Livepeer encoding + Akash inference slashes costs versus traditional cloud providers.
been running inference jobs on livepeer for the past month. its not perfect yet but the cost savings compared to running my own gpu rig are real
running inference on livepeer for a month and the savings are real but you trade reliability for price. production workloads need parity not discounts
been testing livepeer for image gen workloads. cost is great but cold start times are rough. any tips on optimizing that?
livepeer inference is improving fast but the latency is still an issue for real time apps. batch processing is where it shines right now
batch inference on livepeer costs about 60% of what AWS charges for comparable GPU time. real time is the goal but batch is paying the bills right now
batchlord batch at 60pct of AWS is solid math. but the cold start problem is what kills adoption. nobody builds on infrastructure that randomly takes 45 seconds to provision
batch on livepeer at 60% of AWS cost is solid for offline workloads but cold starts kill it for anything user facing. needs better provisioning
The potential here is huge with forecasts of 1 million AI agents by 2025. Akash delivering 30% cost savings versus AWS pricing could really accelerate adoption in decentralized setups.
How exactly will Livepeer and Akash handle the projected 1 million AI agents by 2025 when Akash already charges 30% less than AWS amid the 225% NVIDIA surge?
Akash renting underutilized GPUs for AI training while Livepeer handles video encoding is finally competing with centralized cloud on cost.
NVIDIA at a trillion and Akash is still a rounding error on cloud spend. the thesis makes sense but adoption cycles for infra are painfully slow
Livepeer cold start issue is why I gave up. batch rendering works but any interactive workload is a nightmare. switched to RunPod and never looked back
Henrik O. adoption is slow because migrating off AWS means retraining your whole devops team. the 30% savings doesnt cover the switching cost for most companies
Henrik O. saying Akash is a rounding error but they just crossed 50M in quarterly revenue. AWS does that before lunch but growth rate is what matters here
Luka P 50M quarterly is impressive until you realize AWS does that in about 4 hours. growth rate argument only works if the base is meaningful
cold start issue is real. tried livepeer for image gen and waited 40 seconds for provisioning. batch is fine but anything user-facing is a no-go right now
batch_proxy_ 40 seconds for cold start is a provisioning issue not a protocol issue. Akash fixed this with persistent bids in v0.6 but nobody updated the benchmarks
NVIDIA at a trillion mcap and Akash trading at like 1/1000th of that. the arbitrage on compute pricing is real if they can fix the UX nightmare that is current deployment flows