Who Is Greg Osuri and How Akash Built the Decentralized GPU Cloud

Greg Osuri sits at the center of the strangest pivot any Layer-1 founder has pulled off this cycle. Eight years ago he was selling a peer-to-peer cloud marketplace to a crypto audience that mostly cared about DeFi yields. Today his network runs NVIDIA Blackwell B200 and B300 inference jobs at 70-85% below AWS pricing, AkashML went live last November, and AKT is up roughly 72% year-to-date at $0.78 with a reworked tokenomics model tying supply directly to compute usage.

Osuri is the founder and CEO of Overclock Labs, the company that built Akash Network, and the public face of the decentralized cloud thesis. He is a 25-year open-source veteran, an early AngelHack co-founder, and an ally of Erik Voorhees in the push for open and permissionless AI. The story of how he ended up running the largest decentralized GPU marketplace in crypto is also the story of why the AI compute trade keeps coming back to AKT.

From Indian Software Engineer to San Francisco Cloud Architect

Osuri grew up in India and started his professional career in 2004 at Miracle Software Systems as a technical architect, where he spent two years on enterprise networking and distributed systems before moving to the United States. He joined IBM in 2006 as a consultant in the Nashville area, working in critical infrastructure and service-oriented architecture for clients including Verizon, Sprint, JP Morgan Chase, and Blue Cross Blue Shield. That exposure to how Fortune 500 companies actually run their compute is the part of his background that shapes Akash today.

In 2008 he led the design of Kaiser Permanente's first cloud architecture, one of the earliest production cloud deployments at a US healthcare provider. AWS had launched EC2 only two years earlier, "the cloud" was still mostly a marketing term, and engineers who actually understood how to move enterprise workloads off bare metal were rare. Osuri was one of them.

From there he founded a sequence of smaller startups (SBILabs in 2008, Gridbag in 2009) before launching the venture that put him on the public map. AngelHack opened in November 2011 and grew into the world's largest hackathon organization, with more than 200,000 developers across 164 cities at its peak. Osuri served as founder and CTO for nearly two years before stepping back, and the network he built running AngelHack became the recruiting pipeline for everything that came after.

Why He Founded Overclock Labs and the Original Akash Thesis

Osuri founded Overclock Labs in June 2015. The original pitch had nothing to do with crypto. The thesis was simple. Cloud computing in the mid-2010s was overwhelmingly dominated by three hyperscalers (AWS, Microsoft Azure, Google Cloud) whose pricing power came from controlling both the supply of data center capacity and the distribution of access to it. At the same time, roughly 85% of all data center capacity in the world sat idle on any given day, locked inside corporate servers, university computer labs, mining farms, and underutilized colocation racks.

If you could build a peer-to-peer marketplace that connected unused compute on one side with developers who needed it on the other, the price floor would collapse and the hyperscaler margins with it. That is the original Akash thesis, and Osuri spent three years building the technical foundation before the public token launch.

The decision to put it on a blockchain was not an afterthought. A peer-to-peer compute marketplace needs trustless settlement, automated escrow, and a way to coordinate thousands of independent providers without a central operator. By 2017 the Cosmos SDK had matured enough to be the right substrate for that, and Osuri picked Cosmos over Ethereum specifically because Akash needed sovereign control over its own validator set and economic parameters. Akash's mainnet launched in September 2020 with co-founders Adam Bozanich and Boz Menzalji, and the AKT token began trading shortly after.

What AKT the Token Actually Does

AKT is the native token of the Akash Network and serves three core functions. It pays for compute, secures the chain through delegated proof-of-stake, and governs the protocol through on-chain voting. Until earlier this year that was the entire token model, and it had a known problem. AKT inflation issued new tokens to validators regardless of actual network usage, which meant compute demand and token demand were only loosely connected.

Project Twilight changed that. The hard fork activated on March 23, 2026, and introduced the Burn-Mint Equilibriumtokenomics model. Under BME, when a developer pays for compute, the AKT used in that transaction is permanently burned. Simultaneously, a non-transferable USD-pegged credit called the Akash Compute Token (ACT) is minted for the provider. The accounting cleans up two problems at once. Tenants get stable, predictable USD pricing for their workloads, and AKT now has a usage-driven deflationary mechanism that ties the token's supply curve directly to network revenue.

The market has noticed. AKT has run from roughly $0.45 at the start of 2026 to $0.78 as of May 9, a year-to-date gain of about 72% in a market where most large-cap alts are flat or down. The thesis is straightforward. If decentralized GPU compute is the trade, AKT is the cleanest liquid expression of it.

The AI Pivot That Reshaped Akash

The single most important thing Osuri has done as a CEO is the AI pivot. Through 2022 and 2023, Akash was largely a CPU-focused cloud marketplace competing with cheap VPS providers. The model worked but it was unexciting and the AKT token traded accordingly, hovering between $1 and $4 depending on broader crypto sentiment.

Then ChatGPT happened, the GPU shortage cracked open, and Osuri redirected the entire roadmap toward decentralized GPU compute. Akash added NVIDIA H100 support in 2023, expanded to L40S and A100 inventory through 2024, and by late 2025 had begun integrating the new Blackwell B200 and B300 architecture that hyperscalers were paying premium prices to secure. The pitch to AI startups was direct. If Microsoft and Google have locked up most of the early Blackwell supply, Akash gives you a second source through its decentralized provider network at substantially lower cost.

The November 2025 launch of AkashML closed the loop. Most decentralized GPU networks fail at the user-experience layer because deploying inference on bare-metal Kubernetes is genuinely hard, and AI engineers want to ship models, not configure clusters. AkashML wraps the underlying complexity behind a serverless interface. Pre-configured open-source models (Llama 3.3-70B, DeepSeek V3, Qwen 2.5-30B, NVIDIA Nemotron) ship as one-click API endpoints, and the costs come in 70-85% below AWS SageMaker or OpenAI's hosted inference. The product has been live for six months, and it is the single biggest reason institutional capital has started paying attention to AKT again.

Why Osuri Is the Cypherpunk in the AI Compute Conversation

Osuri's public profile in 2026 is louder than it has ever been, and he uses the platform deliberately. He testified before the US House Financial Services Committee on May 21, 2025, on cloud infrastructure and decentralization. He helped pass California's first blockchain law (AB 2658) by providing expert witness testimony at the State Senate. He is a regular speaker at Token2049, Consensus, and Permissionless. And the throughline of every public appearance is the same argument.

The argument is that AI is a power-law industry where the chip and data center stack is consolidating into the hands of three or four hyperscalers, and that consolidation is incompatible with an open internet. Osuri's framing has become the standard cypherpunk position on AI compute. If you do not own the rails, you do not own the application, and an AI economy that runs entirely on AWS is an AI economy that runs at the discretion of a single vendor.

The position has earned him a working alliance with Erik Voorhees, the ShapeShift founder turned Venice.ai builder, who runs Venice as the privacy-first answer to ChatGPT. Voorhees and Osuri spoke together at Akash Accelerate in 2024 in a session billed as "The Power of Permissionless," and the partnership has become the public template for how decentralized AI is supposed to work. Venice handles the front-end inference experience while Akash provides the back-end compute, both open-source, both running on AKT-priced infrastructure, and neither depending on a hyperscaler to operate.

What's on the Roadmap Through Mid-2026

The near-term roadmap has three concrete items worth watching. The first is the Lease-to-Lease Private Networkinglaunch scheduled for May 30, 2026. The feature introduces secure, direct communication between leased compute instances on Akash, mimicking the way a traditional Virtual Private Cloud (VPC) works inside AWS or Azure. For enterprise customers, the absence of VPC-style networking has been one of the largest blockers to migrating workloads. Once it ships, an entire category of multi-instance deployments becomes possible on Akash for the first time.

The second is the continued ramp of Blackwell GPU inventory. Akash providers are receiving B200 units through the supply chain on roughly a one-quarter lag from the hyperscaler allocations, and the B300 expansion is expected to reach the network by the second half of 2026. Each new generation of NVIDIA hardware that becomes available on Akash widens the price gap with centralized providers because most enterprises cannot get on the hyperscaler waitlists for the latest chips at any price.

The third is the Akash Homenode beta, which opened sign-ups in Q1 2026. Homenode lets individual owners of consumer-grade hardware (RTX 4090, RTX 5090, RTX Pro 6000 Blackwell) plug their GPUs directly into the Akash marketplace and earn AKT for serving inference workloads. The program is small in absolute supply terms but carries real long-tail significance, because even a single-digit percentage of the consumer-grade cards already in circulation plugging into Akash would expand the network's inference capacity by an order of magnitude with no incremental data center buildout.

Frequently Asked Questions

Is Greg Osuri still the CEO of Akash Network?

Yes. Osuri is the co-founder and current CEO of Overclock Labs, the company that builds and maintains the Akash Network protocol. He is also the public face of the project at most major conferences and on regulatory matters in the US, where he has provided expert testimony to Congress and to California state legislators on decentralized infrastructure.

What is the difference between Akash Network and centralized GPU clouds like AWS?

Akash is a decentralized peer-to-peer marketplace. The compute capacity comes from independent providers around the world bidding on developer requests through a reverse-auction model, and prices typically settle at 70-85% below AWS, Azure, or Google Cloud for equivalent workloads. AWS gives you a single vendor with predictable SLAs and integrated services. Akash gives you cheaper compute, no central operator, and access to GPU generations that hyperscalers are still rationing.

How does AKT capture value from Akash usage?

Through the Burn-Mint Equilibrium model that activated on March 23, 2026. Every dollar of compute paid through Akash burns a corresponding amount of AKT permanently, and the protocol mints non-transferable USD-pegged compute credits to the provider. The mechanism converts compute revenue directly into AKT supply contraction. Higher usage means more burn, and more burn means a tighter supply curve over time.

Is decentralized AI compute actually competitive with hyperscalers?

For inference workloads it is increasingly competitive, especially on cost. AkashML demonstrates the case at the production level by serving Llama 3.3-70B and DeepSeek V3 at 70-85% lower prices than comparable AWS or OpenAI hosted endpoints. For large-scale training runs the gap is narrower because hyperscalers still hold the multi-thousand-GPU clusters that frontier model training requires. The honest answer is that decentralized compute wins on inference today and is closing the gap on training workloads as Blackwell inventory diversifies.

Bottom Line

Osuri runs into the second half of 2026 with the strongest hand he has ever held. AkashML is shipping production inference at hyperscaler-undercutting prices, Blackwell inventory is expanding through the provider network, the BME tokenomics tie AKT supply directly to compute usage, and the Lease-to-Lease Private Networking launch on May 30 unlocks the enterprise VPC use case that has been the missing piece for two years.

Watch three things into June. First, the actual usage curve after the May 30 networking upgrade goes live, because that signals if enterprise pipelines actually convert. Second, the AKT burn rate published in the next on-chain analytics update, because the BME thesis only works if compute revenue is genuinely growing. Third, any signaling from the Akash team on a Blackwell B300 inventory commitment for the second half. If those three signals come in green, the AKT trade in 2026 looks closer to the start of a re-rate than the end of one.

This article is for informational purposes only and does not constitute financial or investment advice. Cryptocurrency trading involves substantial risk. Always conduct your own research before making trading decisions.