The center is retreating. Midjourney throttles its unlimited tier. OpenAI quietly caps GPT-4 queries. Adobe Firefly introduces credit walls. The narrative is clear: generative AI, as a service, is moving toward metering, toward restraint, toward pay-per-pixel exhaustion.
Then comes Sogni AI. A counter-punch announced from Singapore, delivered via a threadbare blog post. A $20 monthly plan – dubbed Sogni Unlimited – that promises exactly what the giants are retracting: unfettered access to 100+ open-weight models spanning image, video, and music generation. Running not on centralized data centers, but on a ghost fleet of consumer-grade GPUs, stitched together by the Sogni Supernet.
The crypto news cycle yawned. No token, no airdrop, no speculative frenzy. Just a simple subscription page. But for anyone who has spent years decoding the infrastructure of this industry, this launch is a stress test of a thesis I’ve held since the Terra-Luna collapse: sustainable yield comes from real users, not token emissions. Sogni Unlimited might be the first DePIN product that actually works – or the first to expose the hidden fragility of decentralized compute.
Context: Why Now?
Decentralized Physical Infrastructure Networks (DePIN) have been a hot narrative since 2024, but most projects live on fantasy economics. A token is minted, distributed to GPU operators, traded on exchanges – and the value floats on sentiment, not demand. Sogni’s history is different. Its Supernet, launched a year ago, has already processed over 158 million creations, according to the team. Every render was executed by volunteer-run consumer GPUs, incentivized by a share of revenue – not by inflationary token rewards.
From editorial desk to the bleeding edge of crypto, I’ve watched this pattern before. The early DePIN projects (think Filecoin, Render) relied on heavy speculation to bootstrap supply. Sogni, instead, built a mill before announcing the product. The Supernet was humming with traffic – image-to-image, text-to-video, music synthesis – before Sogni Unlimited was even a pricing page.
Now, the product wraps that mill in a clean subscription layer. No wallet required. Credit card. Email. Start generating. The operator gets 51% of net revenue. The user gets infinite, fair-use access. The company takes the rest and maintains the pipeline. It’s brutally simple. And that simplicity is both the beauty and the risk.
Core: The Mechanics of an Infinite Spigot
Let’s dissect the model. The plan costs $20/month (or $199 annually). For that, a user can generate images, videos, and music using models like Krea 2 Turbo, LTX-2.3 video, Stable Diffusion variants, Flux, and music generators. The catalog updates weekly, tracking the open-source frontier. If a new model drops – say a more efficient transformer for video – Sogni adds it, often within days.
But infinite is never infinite. The fine print is a “fair use” scheduler. Under heavy load, bursts are queued. Users generating 10,000 requests a day might be throttled. The team behind Sogni (CEO Mauvis Ledford, former CoinMarketCap; CTO Mark Ledford, open-source AI veteran) frames this as a self-balancing system: more subscribers → more revenue → more operator incentive → more compute → more capacity. A virtuous cycle on paper.
Operators (those running the GPUs) earn 51% of net subscription revenue. Net means after payment processing fees, VAT, and refunds. No tokens, no swap-to-stablecoin dance. Straight USD. This is critical. In an industry where DePIN operators often hold depreciating governance tokens, Sogni offers fiat income. The barrier to entry is low: a consumer GPU, a decent internet connection, and the will to run a Docker container.
Decoding the heuristic break in 2021 NFT metadata showed me how centralized gateways could ruin supposed permanence. Here, the risk is reversed: the network relies on thousands of independent operators. If a critical mass leaves – say, because NVIDIA releases a new generation that makes older GPUs unprofitable – the spigot may run dry. But the 51% split is generous compared to cloud giants like AWS (which takes 80%+ margin). Early operators on forums report earnings of $80-$150 per month per RTX 4090, depending on usage. That covers electricity and device depreciation, with some profit.
Still, the real core insight is this: Sogni has inverted the typical crypto incentive structure. Instead of paying operators with tokens that need liquidity, it pays them from actual subscription revenue. This removes the Ponzinomic risk that plagues 90% of DePIN projects. No bootstrapping with treasury. No artificial APR. Just a direct link between user demand and supply reward.
Contrarian: The Centralization of Decentralized Compute
Here’s the angle that most coverage misses: Sogni Unlimited is not decentralized. Not really.
The Supernet may use distributed GPUs, but control remains tightly held. The company decides model additions, pricing changes, fair-use thresholds, and – most importantly – the net subscription revenue calculation. Operators trust the firm to accurately deduct fees and distribute 51%. Users trust that their request queue is prioritized fairly. There is no chain-based governance, no on-chain proof of settlement, no slashing for malicious behavior.
This is a trust-minimized-but-not-trustless system. Sogni essentially runs a centralized scheduler on top of a distributed compute layer. The same critique applies to most early DePIN projects: they are federated, not truly decentralized.
But does it matter? For the user, probably not. They get unlimited AI generation for a low monthly fee. For the operator, it depends on whether the platform remains honest. If Mauvis and Mark decide one day to reduce the operator share to 30%, or to introduce a new fee category, the operators have no on-chain recourse. They can only exit. And if they exit en masse, the service degrades.
This is a classic fork in the road for DePIN: does the network need blockchain-level transparency to survive, or is a reputation-based model sufficient? My experience executing flash loan arbitrage in 2020 taught me that trust in centralized oracles can collapse in milliseconds when incentives misalign. Sogni’s risk isn’t an oracle update – it’s a Terms of Service change.
Furthermore, the reliance on consumer GPUs introduces a quality consistency problem. A GTX 3060 serves a Stable Diffusion prompt slower than an RTX 4090. The user doesn’t care about the node – they want speed. Sogni’s scheduler must route tasks intelligently, potentially favoring fast nodes, leaving slower ones underutilized and underpaid. This could create a tragedy of the commons: only the fastest GPUs join, making the network effectively a small fleet of high-end cards, while the slogan “community-powered” rings hollow.
Takeaway: The Proving Ground for DePIN’s Next Chapter
Sogni Unlimited is a milestone. It proves that a DePIN network can generate real, fiat-based revenue from real users without token speculation. It also reveals the limitations: centralized control, quality variance, and the fragility of operator trust.
The next watch is user growth. If Sogni can cross 100,000 subscribers within a year, the model becomes self-sustaining. If not, the network may shrink. I’ll be monitoring two metrics: average GPU hours utilized per week (to gauge demand realness) and operator churn rate (to gauge satisfaction).
One thing is clear: the era of token-guzzling DePIN is over. The winners will be those who deliver services people actually pay for. Sogni Unlimited is the first real test of that thesis. The center may be retreating, but on the edges, a new kind of compute economy is being stress-tested – and this time, the code is just a subscription page.