Sourya Majumder has a problem with your computer. Not the hardware, exactly—the entire paradigm. One cursor. One keyboard. One workflow at a time. For a founder who believes AI agents and humans should be able to spin up unlimited parallel processes without thinking twice, the traditional desktop feels like a relic.
His answer is Infinity, a cloud-native operating system that runs entirely in a browser and streams Windows and Linux applications side-by-side as if they were YouTube videos. No virtual machines, no bloated installations—just a tab, a login, and access to Blender, Unreal Engine, AutoCAD, or whatever else you need, all rendered in what ProjectX, Majumder's six-person startup, describes as "one clean UI."
It's an audacious pitch. Whether it's a viable one remains to be seen.
ProjectX is part of the Y Combinator Spring 2026 batch, and the early reception has generated significant attention. A demo video posted at Founders Inc. generated over 850,000 views within 24 hours—enough to briefly crash the company's servers, according to a founder post on LinkedIn from around that period. Thousands of signups flooded in overnight. The hype was real, at least in the moment.
But hype and infrastructure are different beasts.
Streaming Software Like It's Netflix
At its core, Infinity is a bet on cloud delivery of compute-intensive applications. The system uses containerized microservices for each app, streamed via AV1 codec over WebRTC. You open a browser, sign in through Google or GitHub at infinity.projectx.cloud, and launch whatever you need. The company claims cold starts under three seconds and seamless cross-OS compatibility—mixing Windows tools, Linux environments, and web apps in a single session without the usual friction.
Need to run robotics simulations in ROS while editing 4K footage in DaVinci Resolve? Infinity supports Windows, Linux, and web apps in a single session, though specific workflow claims remain unverified by independent testing.
The technical backbone involves NVIDIA A100 and H100 GPUs, dynamically sliced using Multi-Instance GPU (MIG) technology and allocated per application in real time. ProjectX claims this delivers GPU utilization rates above 95%, compared to the 40-60% typical of legacy virtual desktop infrastructure. That's a meaningful efficiency gain if it holds up under production loads. The word "if" is doing some work here.
No independent benchmarks have surfaced yet to validate those numbers.
The Agentic Computing Bet
The company's YC profile emphasizes something broader than remote desktop replacement: "the first OS where humans and agents work with no limits." It's a framing that taps into the growing conversation around agentic AI—systems where autonomous software agents operate alongside human users, executing tasks, managing workflows, and making decisions in parallel.
Microsoft released an Agent Governance Toolkit earlier this year that explored policy layers and execution rings for what it called "Agent OS" architectures. Infinity is positioning itself as the infrastructure where that kind of orchestration actually runs, though the practical contours of human-agent collaboration in Infinity remain largely theoretical for now. There are no public customer case studies, no enterprise deployments on the record.
Still, the timing is interesting. If agentic workflows become the norm—if AI systems start requiring the ability to spin up dozens of simultaneous processes without waiting for a human to click through—then the single-cursor model Majumder critiques starts to look less like tradition and more like a constraint.
Competitive Landscape and Cost Claims
ProjectX isn't subtle about its targets. The company claims Infinity is "70% cheaper and 10x smoother" than AWS WorkSpaces or Citrix DaaS, the virtual desktop platforms that dominate enterprise remote access. Those are vendor claims that remain unverified by independent third parties.
The competitive set includes Amazon AppStream 2.0, which streams Windows apps over HTTPS, and Kasm Workspaces, which offers browser-based containerized desktops. Infinity differentiates on GPU-intensive use cases, cross-OS flexibility, and—crucially—its explicit focus on agentic workflows rather than simple desktop virtualization.
The security pitch involves zero-trust architecture with eBPF-level sandboxing and enterprise-grade encryption. No third-party audits have been disclosed, and no enterprise customers have publicly confirmed deployments, which leaves some of the operational details in the realm of promise rather than proof.
Traction, Timing, and What's Actually Known
Infinity appears to have launched publicly sometime in the March or April window this year, based on founder posts and a Hacker News submission with the cheeky title "Download More RAM (For Real)." The product is accessible via web login, with beta and university adoption programs channeling through a Linktree registration page. A promotional push offered 100 free credits during a limited-time weekend signup window.
Formal pricing details remain behind a JavaScript-gated page that hasn't been publicly accessible to web crawlers, which makes it difficult to assess the business model in full.
On the funding front, details are patchwork. ProjectX secured $200,000 in Google Cloud infrastructure credits in January 2024. More recently, a data provider called Signalbase reported a $500,000 pre-seed round, though investors were not disclosed and no formal press release or independent confirmation has verified the specifics. CB Insights lists Y Combinator, Pontaq, and Google Cloud as investors, but the round structures and exact dates remain murky.
It's the kind of early-stage opacity that's common for startups fresh out of accelerators, though it does complicate efforts to assess momentum.
The Open Question
For now, Infinity is a compelling concept with impressive technical ambitions and no extended track record. The viral demo and early developer interest suggest the pitch is landing—at least with a certain audience. AI and ML engineers, game developers, creative studios, robotics teams: these are users who need GPU firepower but don't necessarily want to manage local workstations or wrestle with cloud configuration.
ProjectX is exhibiting at GITEX and appears to be leaning into that exact demographic. If the architecture delivers on GPU efficiency, cost savings, and the seamless human-agent collaboration it promises, Infinity could carve out a meaningful position in the cloud computing stack—a category that's crowded but still evolving quickly.
If it doesn't, well. The history of ambitious infrastructure plays that looked brilliant on stage but stumbled in production is long and littered with cautionary tales. For a six-person team, the margin for error is thin.
The next chapter will be written not in demo videos, but in production uptime, customer renewals, and whether the workflows Majumder envisions—unlimited, parallel, unbounded—actually work when the servers are under real load.
