As AI compute becomes increasingly concentrated among a few cloud giants, Gonka is building a decentralized alternative. Since its mainnet launch in August 2025, the Layer-1 network focused on high-efficiency AI compute has attracted 2,200 developers and 12,000 GPU-equivalent capacity, and just secured a $50 million investment from Bitfury.
1. What Gonka is and the Problem It Addresses
Gonka is an L1 decentralized network designed for efficient AI computation. Its core thesis: the bottleneck in AI is no longer models but control over compute. Advanced GPUs are highly concentrated among a handful of manufacturers and hyperscale cloud providers, making AI compute expensive, opaque, and geopolitically constrained. Startups and entire regions face price volatility, capacity shortages, and vendor lock-in. Gonka rethinks this at the protocol level: compute becomes open, verifiable infrastructure rather than a gated service. It uses a Transformer-based Proof-of-Work mechanism that directs nearly all GPU power toward meaningful AI workloads (today primarily inference), not abstract hash puzzles. Hosts earn rewards based on verified computational contribution, not capital allocation.
2. How Its PoW Differs from Bittensor and Others
The main difference lies in what each network defines as 'work.' Projects like Bittensor focus on coordination at the model or network level, with incentives shaped by staking, delegation, or peer evaluation, not always proportional to raw compute. Gonka is a compute-first network where 'work' is verifiable AI computation. Voting power and rewards are directly tied to verified computational contribution. On efficiency: many decentralized systems consume significant compute on consensus or duplicated work (e.g., Bittensor allocates ~60% of rewards to staking, which does not produce AI compute). Gonka's Sprint-based design minimizes overhead, allowing nearly all GPU resources to be used productively.
3. Why Focus on Inference First, Not Training
This was a matter of sequencing, not limitation. Inference is where the majority of real-world AI usage happens today, and the infrastructure bottlenecks are most visible. From a design perspective, inference workloads are continuous, measurable, and well-suited to validate the core principles in a decentralized environment. Training is a different class of problem, but Gonka plans to introduce it in the future. The network allocates 20% of all inference revenue to support future model training.
4. How It Verifies That Miners Actually Performed the Work
Verification is built into how the network measures compute. Inference tasks are executed during short, time-bound periods called Sprints. Hosts are asked to run inference on large Transformer models randomly initialized for each cycle. These tasks cannot be precomputed or simulated. The network checks if outputs match expected results. For efficiency, Gonka does not recheck every computation; it continuously verifies a portion of results and increases checks for suspected participants. If work fails validation, fees are not paid. Hosts that consistently submit correct results gain greater participation in the network.
5. Competing with OpenAI, Google, Microsoft
Gonka doesn't see itself competing with these giants in the traditional sense. Centralized providers control massive infrastructure, but that control comes with trade-offs: gated access, opaque pricing, and capacity shaped by internal priorities. Gonka is designed as open infrastructure, with compute supplied by a decentralized network of hosts and availability determined by real supply and demand. This enables use cases structurally underserved by centralized platforms—workloads requiring openness, predictable access, and infrastructure-level transparency. By letting hardware providers compete directly on performance and efficiency, Gonka also drives down costs for a broader range of developers and regions.
6. Adoption Drivers: 2,200 Developers and 12,000 GPUs
Adoption isn't driven by short-term hype but structural alignment. On the supply side, hosts look for alternatives to centralized models that underutilize their hardware. On the demand side, developers face pricing volatility, capacity constraints, and closed APIs. As more hosts join (independently or through pools), the network becomes more useful for developers. As more workloads come online, sustained demand attracts more infrastructure. This feedback loop drives organic growth. Coordination happens openly in the Gonka community, including the Gonka Discord.
7. $50M from Bitfury without Sacrificing Decentralization
Gonka is decentralized by design at the protocol level, not just in narrative. Governance is tied to verifiable computational contribution, not capital ownership. Bitfury's support brings expertise in building large-scale compute infrastructure, but does not grant special privileges within the protocol. Funding itself does not convert into influence in Gonka.