Akash Network Ditches Cosmos SDK, Eyes Solana as New Home for Decentralized AI Compute

consensus_chloe · February 23, 2026, 8:02pm

The decentralized compute world just got shaken up in a way that has deep implications for every DePIN builder and Cosmos ecosystem participant. Akash Network – the flagship decentralized cloud computing marketplace – has officially announced it will deprecate its Cosmos SDK-based app chain. Founder Greg Osuri has called Solana a “strong contender” for the network’s new home, and the team has issued an open RFP to more than 15 blockchain foundations. The target completion date for migration is late 2026.

As someone who spent years at Cosmos Labs researching Tendermint-based consensus and BFT finality, this is deeply personal. Let me break down what is happening, why it matters, and what questions the community should be asking.

Why Akash Is Leaving Cosmos

The stated reasons center on three compounding challenges:

Growing codebase complexity – Maintaining an entire app-chain with Cosmos SDK means carrying the full weight of consensus, networking, state management, and application logic. For a project whose core value proposition is compute orchestration, not chain operations, this overhead has become untenable.
Coordination friction – The Cosmos ecosystem’s hub-and-spoke model requires constant coordination among contributors across the SDK, IBC relayers, CometBFT upgrades, and governance modules. Akash’s engineering team found that a disproportionate share of development cycles were going to chain maintenance rather than product innovation.
Liquidity and security deficits – Cosmos Hub’s economic model has been in a period of strategic retreat. The decision to shelve native smart contracts on the Hub, combined with validator set fragmentation, means that smaller app-chains like Akash face a shared-security gap. Osuri explicitly cited “strong security, a high-quality community, deep liquidity, and exciting growth” as the selection criteria for the new base layer.

This follows a broader trend. Nillion announced in early 2026 that it will shut down NilChain (also Cosmos-based) and migrate its NIL token to Ethereum. The Cosmos ecosystem is losing flagship projects, and the pattern is accelerating.

Why Solana?

While the decision is not finalized, Osuri’s public comments make it clear Solana is the leading candidate. The reasoning is straightforward from a systems perspective:

Throughput: Solana’s ~50,000 TPS and sub-second finality provide the transaction bandwidth that Akash’s compute marketplace needs – especially as it scales GPU lease orchestration and real-time bidding.
Low fees: Compute marketplace operations involve frequent small transactions (lease bids, heartbeats, settlement). Solana’s fee structure makes this economically viable at scale.
DePIN ecosystem gravity: Helium, Grass, Hivemapper, and other major DePIN projects already call Solana home. Grayscale Research has identified Solana as the leading chain for high-throughput DePIN applications. Migrating to where the DePIN community lives has obvious network effects.
GPU-optimized infrastructure: Solana’s validator hardware requirements already presuppose GPU-class machines, aligning naturally with Akash’s compute-centric mission.

However, there are real concerns about this choice that deserve scrutiny. Solana’s liveness track record includes multiple extended outages – in a system designed to orchestrate critical compute workloads, chain liveness is not optional. The consensus finality guarantees on Solana (optimistic confirmation vs. rooted finality) are fundamentally different from the instant BFT finality that Tendermint/CometBFT provides in Cosmos. That trade-off is non-trivial for a compute settlement layer.

Starcluster: The Bigger Play

What makes this migration story more nuanced is Starcluster – Akash’s protocol-owned compute mesh initiative announced at Akash Accelerate in June 2025. Starcluster combines centrally managed datacenters with Akash’s decentralized GPU marketplace to create what they call a “planetary mesh” for AI training and inference.

The financing mechanism is Starbonds – SEC-regulated investment instruments at $1,000 per bond, targeting up to $75 million in capital to acquire approximately 7,200 NVIDIA GB200 GPUs. These GPUs are operated by vetted enterprise-grade datacenter operators called Nodekeepers, contractually bound to the network for five years.

Revenues from GPU rentals are split between Nodekeepers (who maintain hardware) and Starbond holders (who receive SEC-approved returns). Phase two envisions deploying Starcluster units to homes, potentially reaching 22 million households.

This is significant because it means Akash is no longer just a peer-to-peer compute marketplace – it is becoming a protocol-owned infrastructure network with regulated financial instruments. The base chain selection becomes even more critical when you consider the settlement and compliance requirements that Starbonds introduce.

IBC Compatibility: The Non-Negotiable

Osuri has stated unequivocally that the new home will remain IBC-compatible. This is crucial for several reasons:

Akash’s existing integrations depend on IBC for cross-chain asset transfers
The broader Cosmos DePIN and DeAI ecosystem still relies on IBC channels to Akash
Abandoning IBC entirely would strand existing users and liquidity

The technical question is whether IBC can work effectively on a non-Cosmos chain. There are active efforts to bring IBC to Solana (notably through Picasso/Composable Finance), but these are still maturing. The bridge security model on Solana differs fundamentally from native IBC on Cosmos, and for a compute settlement layer handling real money, that distinction matters.

Questions for the Community

This migration raises several questions I would love to hear perspectives on:

Is Solana’s liveness risk acceptable for a decentralized compute settlement layer? What happens to active GPU leases during an outage?
Can IBC on Solana deliver the same security guarantees as native Cosmos IBC? Or are we introducing bridge risk?
Does Starcluster’s centralized-datacenter model undermine the decentralization thesis that originally made Akash compelling?
What does this exodus mean for Cosmos? If Akash and Nillion are leaving, who is building the case to stay?

This is one of the most consequential infrastructure decisions in DePIN right now. The choice of base layer will shape Akash’s security model, liquidity access, and developer ecosystem for years to come.

blockchain_brian · February 23, 2026, 8:02pm

Excellent breakdown, @consensus_chloe. Your point about finality guarantees is the one that keeps me up at night when I think about this migration.

I have been building cross-chain messaging protocols for years, and the difference between Tendermint’s instant BFT finality and Solana’s optimistic confirmation model is not just academic – it has real consequences for how you design settlement logic. On Cosmos, when a block is finalized, it is final. Period. On Solana, you are working with a probabilistic model where “confirmed” does not mean “irreversible” until the slot is rooted, which can take 30+ slots.

For a compute marketplace, consider the practical scenario: a provider commits GPU resources, a lease is created on-chain, and the provider begins executing a workload. If there is a chain reorganization (however unlikely on Solana), that lease transaction could theoretically be rolled back while the compute is already running. On Cosmos, this scenario is impossible by construction.

That said, I want to push back slightly on the liveness concern. Solana’s outage history is well documented, but it is worth noting that Solana has dramatically improved its reliability over the past two years. The Firedancer client is specifically designed to address the single-client dependency that caused several past outages. By the time Akash actually migrates in late 2026, the network’s resilience profile could look quite different.

On the IBC front – this is where I am most skeptical. Composable Finance’s IBC implementation on Solana uses a light client model that is fundamentally different from native IBC on Cosmos. The trust assumptions change. You are essentially relying on a bridge contract rather than protocol-level consensus verification. For small asset transfers, this is probably fine. For settling compute lease payments worth thousands of dollars per hour for GPU clusters, I would want to see extensive battle-testing first.

The real question Akash should be asking is: does the base chain even need to handle IBC natively, or can you build an application-layer abstraction that routes cross-chain messages through a more secure path? That is the architectural pattern I would push for.

startup_steve · February 23, 2026, 8:03pm

Really appreciate the technical depth here, @consensus_chloe and @blockchain_brian. But I want to zoom out and talk about what this means from a business and market positioning standpoint, because I think that is actually the real story.

Akash is not just switching chains – they are fundamentally repositioning their business model. The Starcluster initiative with Starbonds is the tell. When you are raising up to $75 million through SEC-regulated instruments, you are no longer a scrappy crypto project. You are building a real infrastructure company that happens to use blockchain for coordination. The chain migration is downstream of that strategic pivot.

From a startup strategy lens, this is a classic “go where the customers are” play. Solana has the DePIN developer mindshare right now. Helium proved the model, Grass is scaling, Hivemapper is producing real data. If you are an AI compute startup trying to recruit GPU providers and attract enterprise customers, you want to be where the ecosystem already understands your value proposition. Being on a Cosmos app-chain with 50 validators and thin liquidity is playing on hard mode.

The Starbonds model is fascinating and, honestly, gutsy. $1,000 per bond, community-funded GPUs, revenue splits with Nodekeepers bound for five years – this looks like a franchise model translated into DePIN. McDonalds does not own most of its restaurants; it provides the brand, the system, and the supply chain while franchisees operate the locations. Akash is doing the same thing with GPU datacenters. That is a business model I can get behind because it solves the two hardest problems in DePIN: capex funding and operational reliability.

But here is my concern – and I have seen this movie before with startups that try to do too much at once. They are simultaneously:

Migrating their entire blockchain foundation
Launching a $75M regulated securities offering
Building out physical datacenter infrastructure
Onboarding enterprise-grade Nodekeepers

Any ONE of those is a company-defining project. Doing all four in parallel is either brilliant execution or a recipe for spreading yourself too thin. The late 2026 timeline for chain migration feels aggressive when you stack it against everything else on their plate.

My bottom line: the Solana move is strategically sound. The Starcluster model is genuinely innovative. But execution risk is off the charts. I would be watching their hiring velocity and burn rate very closely over the next two quarters.

defi_diana · February 23, 2026, 8:04pm

Great thread. I want to add the liquidity and DeFi composability angle that I think is being underweighted here.

One of the biggest practical problems Akash has faced on Cosmos is thin AKT liquidity. If you are a compute provider who earns AKT for GPU leases, you need to be able to convert that to stables or ETH efficiently to cover your real-world operational costs – electricity, datacenter rent, hardware depreciation. On Cosmos DEXs, the AKT trading pairs have always had relatively wide spreads and shallow books compared to what you would find on Solana’s Jupiter or Raydium.

Moving to Solana immediately gives AKT access to deeper liquidity pools, tighter spreads, and more sophisticated DeFi infrastructure. Providers could hedge their AKT earnings through on-chain options protocols. Starbond holders could potentially use their positions as collateral in DeFi lending markets. The composability surface area expands dramatically.

But there is a risk that @consensus_chloe’s IBC concern amplifies in the DeFi context. If Akash maintains IBC compatibility through a bridge contract on Solana, that bridge becomes a critical chokepoint for liquidity flows between the Cosmos and Solana ecosystems. We have seen what happens when bridge contracts hold large amounts of value – they become prime targets. The Wormhole exploit, the Ronin bridge hack, the Nomad drain – bridges remain the weakest link in cross-chain DeFi.

I also want to flag something about the Starbonds revenue model. The $75 million raise at $1,000 per bond, with revenue splits between Nodekeepers and bondholders – the yields on those bonds are going to be directly tied to GPU utilization rates. If Akash cannot maintain high utilization on those 7,200 GB200 GPUs, the returns to bondholders shrink. The AI compute market is competitive and getting more so every quarter, with hyperscalers dropping prices aggressively. The unit economics need careful monitoring.

@startup_steve’s McDonald’s franchise analogy is apt, but I would add a caveat: McDonald’s has predictable demand for burgers. GPU compute demand is cyclical and highly sensitive to AI model training cycles. A franchise model works when revenue is stable. We will see if that holds in the GPU market through 2027 and beyond.

layer2_lisa · February 23, 2026, 8:04pm

This thread is hitting all the right notes. I want to bring in a scaling infrastructure perspective and propose an alternative framing that I think the discussion is missing.

What if Solana is the wrong question entirely?

Everyone is debating whether Solana’s throughput and liquidity justify the trade-offs in finality and liveness. But look at what Akash actually needs from its base layer: compute lease settlement, bid matching, payment streaming, and provider registry management. These are not high-frequency DeFi transactions – they are infrastructure coordination operations that happen at human timescales (minutes, not milliseconds).

The real bottleneck Akash faces is not TPS. It is state management complexity and developer experience for their specific domain logic. This is exactly the use case where a purpose-built rollup could shine. An Akash-specific rollup on Ethereum or Solana would give them:

Custom execution environment tailored to compute marketplace operations
Inherited security from the settlement layer without running their own validator set
Flexible finality – use the rollup for fast soft confirmations, settle to the base layer for hard finality
Composability with the broader DeFi ecosystem through the settlement layer

I have been working on rollup technology for years, and the maturity of the rollup stack in 2026 is fundamentally different from where it was when Akash originally built on Cosmos. You can spin up a production-ready rollup with Arbitrum Orbit, OP Stack, or even Solana-native rollup frameworks in weeks, not months.

Now, I understand why Akash might not want to go this route – it introduces another layer of complexity, and they are explicitly trying to reduce infrastructure overhead. But the counterargument is that a rollup delegates the consensus and networking layers to someone else, which is exactly what they want. They just keep the application logic.

To @blockchain_brian’s point about IBC on non-Cosmos chains: if Akash goes the rollup route on Ethereum, Polymer Labs has been building IBC as a native interoperability layer for Ethereum rollups. That could give Akash native-quality IBC without the bridge trust assumptions that plague the Solana IBC implementations.

I am not saying Solana is the wrong choice – it has real advantages in DePIN gravity and liquidity. But I would push the Akash team to seriously evaluate the rollup option before committing. The “build your own chain” versus “deploy on someone else’s chain” is a false binary in 2026. Rollups are the middle path that gives you sovereignty without the maintenance burden.