Sogni Unlimited: The DePIN Subscription That Pays Its Own Way—But Trust Is the Hidden Tax

The subscription economy has a dirty secret: most “unlimited” plans are engineered to fail at scale. Netflix throttles bitrate. OpenAI imposes rate limits. Midjourney caps concurrent jobs. The fine print is a game of incentives—if too many users actually consume, the model breaks. Sogni AI’s Sogni Unlimited is different. It doesn’t hide behind opaque fair-use policies. Instead, it outsources the cost of unlimited generation to a decentralized network of consumer GPUs, paying operators 51% of net subscription revenue in real dollars. No token. No inflation. No Ponzi. Mathematically, it works if the revenue split exceeds the marginal cost of a GPU hour. Empirically, the Supernet has already served 158 million generations in a year. That’s a lot of inference. But math doesn’t care about trust—it cares about constraints. And Sogni Unlimited’s constraints sit behind a single company’s API. Let me walk through the code logic. Not the Solidity—there is none. The smart contract here is the subscription gateway and the scheduler that allocates GPU tasks to operators. The core insight is simple: replace speculative token rewards with stable USD payouts. Operators earn money that can buy electricity, rent, or GPUs. Users pay for a service, not a bet on a token price. This is the cleanest DePIN model I’ve seen since Helium’s early days—before it turned into a hotspot subsidy machine. But cleanliness comes at a price: dependence on a single payment processor. Credit cards are the on-ramp. No wallet, no gas, no slippage. That’s great for mainstream adoption. But it also means the project must comply with Visa, Mastercard, and bank KYC. The team—led by former CoinMarketCap executive Mauvis Ledford and his brother Mark—runs a Singapore-based entity. They control the subscription pricing, the revenue split, the fair-use algorithm, and which models are served. From a systems perspective, this is a centralized hub with decentralized spokes. The GPU operators are distributed, but the rules are set by a 2-person board. If they decide to change the 51% split to 40%, what recourse do operators have? None. No governance token, no DAO, no on-chain voting. That’s a double-edged sword: it avoids the inefficiencies of token voting, but it also eliminates the checks and balances that make DeFi systems resilient. Let’s dissect the fair-use mechanism. Sogni Unlimited promises unlimited generations—but with a scheduler that queues requests under load. The exact thresholds are unpublished. Based on my experience auditing similar resource-constrained systems, the scheduler likely uses a combination of request rate, model size, and GPU availability to prioritize or delay tasks. If too many users hit the network simultaneously, some will wait. The question is: where is the cutoff? If the waiting time exceeds 30 seconds, users will abandon the service. If it exceeds 2 minutes, they’ll switch back to Midjourney. The team claims the network is powered by “consumer GPUs” from independent operators. A single RTX 4090 can generate about 2–4 images per second for a mid-sized model like SDXL. With thousands of GPUs, the aggregate capacity is substantial. But consumer GPUs have failure rates. Operators might drop out when electricity prices spike or when the dollar payout doesn’t cover hardware depreciation. The revenue split must be high enough to retain them. At 51% of net revenue—minus payment processing fees, refunds, and perhaps the project’s own overhead—the actual operator cut could be closer to 40%. That still beats most mining returns, but it’s a thin margin. Now, the contrarian angle: Sogni Unlimited’s real vulnerability isn’t technical—it’s structural. The project is a classic “trust me” system dressed in decentralized clothing. The GPU network is decentralized, but the revenue distribution is not. The model selection is not. The fair-use definition is not. In traditional DePIN projects like Render or Akash, the token serves as both a governance and a penalty mechanism—operators stake tokens, and malicious behavior leads to slashing. Sogni has none of that. An operator could run a free LLM query for themselves and never pass it through the scheduler, effectively stealing compute. There is no zero-knowledge proof verifying that a generation actually happened on their GPU. The trust assumption is that operators are honest or that the project can monitor them via telemetry—which is centralized. This is where my background in zero-knowledge research kicks in: verifying GPU work without revealing the input or output is an open problem. Existing solutions (zkML, zkPoP) are too expensive for real-time image generation. Sogni likely falls back on statistical monitoring—checking that operator uptime and response times match expected patterns. It works for now, but as the network scales, malicious operators could collude to fake work and claim revenue. The only defense is reputation, and reputation is fragile. Privacy is a protocol, not a policy. Sogni Unlimited doesn’t even claim to be privacy-preserving. All generations pass through the project’s API, which could log prompts and outputs. The terms of service likely allow it. For a tool targeting creators, this might be acceptable. But for anyone generating sensitive content—medical data, financial reports, personal art—it’s a red flag. The project should at least offer an opt-in encryption layer or local inference mode. Until then, the “unlimited” tag comes with a privacy leak. Let’s zoom out to the market. Centralized AI platforms are raising prices and capping usage. Midjourney’s base plan is $10/month for 200 generations. OpenAI’s ChatGPT Plus is $20/month but with message limits. Sogni Unlimited undercuts them on price and removes caps. That’s a strong hook. But the user must accept lower-quality open-weight models—no GPT-4, no Midjourney v6. The models listed (Krea 2 Turbo, Flux Alpha, LTX-2.3) are competitive but not state-of-the-art for every task. The value proposition is quantity over quality. For prototyping, it’s perfect. For final output, professionals will still pay for the top-tier models. From a game-theoretic lens, Sogni Unlimited is playing a long-term equilibrium game. The subscription revenue must grow faster than the cost of supporting unlimited usage. If too many power users join, the network may degrade, forcing the team to tighten fair-use or raise prices. That would break the core promise—and likely send users back to centralized platforms. The defense is a diverse user base: light users subsidize heavy users, and the revenue stream smooths out. But the team hasn’t disclosed user distribution. We only know the aggregate generation count. In my experience, the most sustainable DePIN projects are those where the underlying resource has a natural floor price. GPU compute does: the electricity cost plus hardware amortization. If Sogni Unlimited’s subscription revenue can permanently stay above that floor, the network runs profitably. If not, operators leave. The beauty of the model is that it self-corrects—operators join when demand is high and leave when it’s low. The risk is that the team might artificially suppress payouts to increase margins, accelerating the exodus. What about competition? Render Network and Akash Network are the obvious peers. Render focuses on high-end 3D rendering with a token-based economy. Akash offers general cloud compute. Neither has a user-facing subscription product. Sogni Unlimited is the first to wrap a DePIN backend in a simple monthly fee. That’s a first-mover advantage in a niche that could balloon. But both Render and Akash could easily clone the model—they already have the token infrastructure. The difference is that their operators are paid in volatile tokens, which adds friction. Sogni’s USD payouts are more attractive to risk-averse GPU owners. The team’s background gives some confidence. Mauvis Ledford spent years at CoinMarketCap, a data giant that understands both infrastructure and user psychology. Mark Ledford has AI engineering chops. The brothers have been building together for over a decade. That reduces the risk of a rug pull or incompetent execution. But it doesn’t eliminate the risk of mismanagement or regulatory pressure. Regulatory risk is minimal. No token means no SEC classification. Credit card payments are well-regulated. The main exposure is to content liability laws (e.g., generating child sexual abuse material or copyright infringement). All AI platforms face this. Sogni will need robust content moderation and takedown procedures. Failure here could shut them down in major markets like the EU or US. So, what’s the takeaway? Sogni Unlimited is a well-engineered product with a clean economic model. It solves the “unlimited” problem by aligning incentives with operators instead of relying on artificial caps. But the lack of on-chain verification, governance, and privacy features creates critical blind spots. The project is centralized in all the important ways: rulemaking, revenue distribution, and model selection. Over time, these blind spots may grow into vulnerabilities. If the team transitions to a DAO or introduces slashing mechanisms, the risk profile improves. If they don’t, the network’s integrity depends entirely on their goodwill. For developers and creators, Sogni Unlimited is a genuine option—especially for experimental work. For crypto investors, there’s no token to trade. For operators, it’s a chance to earn passive income if you have spare GPU cycles. But read the fine print. The fair-use policy is a black box. The revenue split is subject to change. And trust is a vulnerability, not a virtue. Math doesn’t lie, but contracts do. The numbers say Sogni Unlimited works today. The question is whether the project’s governance can survive its own success.