On July 28, 2026, Sogni AI announced Sogni Unlimited—a $20/month subscription for unlimited access to over 100 AI models, powered by a decentralized GPU network. The press release reads like a victory lap: Supernet mainnet one year old, 1.58 billion creations delivered, timing perfect as centralized platforms retreat from unlimited plans. The hook is seductive. But in my thirteen years of tracing the silent bleed from 2017's broken logic, I've learned that when a project claims to solve the hype cycle's failure with a new layer of complexity, the real failure is often just masked. Sogni Unlimited is not a scam—it is a smart business model sold as a decentralized revolution. The code doesn't lie, but the marketing does.
Context: The AI Subscription Graveyard
Over the past six months, Midjourney and OpenAI have quietly throttled or eliminated their unlimited tiers. The reason is simple: inference costs are exploding. Running a single high-end query on a cluster of H100 GPUs can cost cents to dollars per call. For platforms serving millions of users, unlimited plans are actuarial disasters. Enter Sogni, positioning itself as the decentralized savior: 'We don't need datacenter GPUs because we use idle consumer-grade hardware from our community. We don't need venture capital because our subscription revenue pays operators directly. We don't need tokens because real dollars work better.' The narrative is clean, almost too clean. The founder, Mauvis Ledford, is a former CoinMarketCap executive. The CTO, Mark Ledford, has an open-source AI background. The company is headquartered in Singapore. All signals point to legitimacy. But legitimacy is not the same as transparency.
Core Insight: The Operator's Dilemma and the Fair Use Trap
Let's dissect the economic model. A user pays $20 per month. Sogni deducts payment processing fees, taxes, and refunds. The remaining 'net' was 51% to the operator who provided the GPU that processed the user's request. The other 49% stays with Sogni. On the surface, this is a fair split: the operator gets the majority. But note: 'net' is self-defined. There is no on-chain verification of the revenue pool, no smart contract enforcing the distribution. Sogni controls the books. Operators must trust that the company is honest about total subscription revenue and deductions. In my audit experience of DePIN projects, this single point of trust is where most failures originate. Complexity is just laziness wearing a tech suit.
Furthermore, the 'fair use' policy is a floating throttle. The article describes a scheduler that prioritizes users based on recent usage, preventing abuse. But the criteria are opaque. At what point does a 'heavy user' get deprioritized? Is it after 50 generations per day? 100? 500? The lack of clear limits is a feature for marketing but a risk for users who build workflows around the service. If Sogni decides to tighten limits to preserve margins, the 'unlimited' promise breaks. The company retains full discretion. This is not decentralization; it is a vending machine with a hidden algorithm.
The GPU Real Estate
The network runs on consumer-grade GPUs—think RTX 4090s, not A100s. This is a deliberate choice. Consumer cards are cheaper, more abundant, and easier for individuals to operate. But they are also less reliable, less powerful, and lack the memory bandwidth needed for cutting-edge models. Sogni's library of 100+ models includes Krea 2 Turbo, Stable Diffusion variants, and LTX-2.3 video—all open-weight medium models. No Llama 3 70B. No GPT-4 scale. The network is optimized for creativity, not heavy lifting. That is fine for the target audience, but it means the 'decentralized compute' narrative is exaggerated. Sogni is a distribution layer for open models, not a general-purpose AI compute market. The code never lies, only the auditors do—and there is no audit of the model scheduling logic.
Contrarian Angle: What the Bulls Got Right
Now, the fairness of the critique. Sogni Unlimited does something genuinely hard: it aligns incentives without token inflation. Operators get paid in dollars, not speculative coins. This eliminates the price volatility risk that plagues projects like Render or Akash. A GPU operator on Render receives RNDR tokens, which may crash 70% before they can be sold. On Sogni, the revenue is stable fiat. This is a superior incentive structure for risk-averse providers. The 1.58 billion creations metric is not fake—it represents actual computational work. The network has been live for a year, processing millions of requests, which proves technical feasibility. The timing is also perfect: centralized AI platforms are retreating from unlimited plans, creating a market gap. Sogni is filling it with a product that works.
The contrarian view also notes that the team is experienced. Mauvis Ledford built CoinMarketCap's data infrastructure. Mark Ledford has deep AI technical chops. The brother-duo structure reduces internal politics. And the decision to avoid a token launch entirely sidesteps regulatory scrutiny. From a business perspective, this is a well-optimized machine.
But here is the tension: can you have a 'decentralized' network without a token?
Without a token, there is no on-chain governance, no slashing for operator misbehavior, no transparent revenue distribution, no community voting on model additions or fair use rules. The network is owned by the company. Operators are gig workers, not stakeholders. Users are customers, not participants. The 'community' is a marketing term. This is closer to Uber than to Bitcoin. And that is fine—unless the promise was decentralized AI. The bulls point to sustainability, but sustainability is not the same as decentralization. Sogni Unlimited is a centralized platform with decentralized compute sourcing. The compute is sourced from many providers, but the platform is a single point of control.
Takeaway: The New Old Centralization
The forensics reveal a truth markets try to bury: most DePIN projects are not building infrastructure—they are building distribution networks. Sogni Unlimited is a clever business model dressed in crypto clothing. It solves the cost problem of AI inference by outsourcing compute to consumer-grade hardware, and it solves the revenue problem by charging a flat subscription fee. The operators get paid, the users get service, the company takes a cut. That is a viable business. But it is not a revolution. The questions investors and operators should ask are: What happens when Sogni decides to increase its cut from 49% to 60%? What happens when a major AI lab releases a model that requires data center GPUs? What happens when the fair use policy is updated to limit routine users? There are no on-chain mechanisms to stop it. The pattern is as old as 2017: a team builds a platform, crowdsources supply, captures value, and eventually centralizes control. Complexity is simply the veil. Sogni Unlimited is a well-executed case study in how to profit from the AI hype while avoiding the risks of tokens and regulation. That is not a criticism—it is a warning. Patterns emerge only when emotion is stripped away. And the pattern here is clear: Sogni is a centralized SaaS company with a decentralized supply chain. Nothing more, nothing less.
