Six months can be a lifetime in the startup world—or, in Applied Compute's case, the exact interval between going public and raising $80 million at a valuation that would make most founders' heads spin.
The enterprise AI startup, founded by a trio of former OpenAI researchers, just closed that round at a $1.3 billion post-money valuation. Kleiner Perkins led the investment, with Elad Gil, Lux Capital, Greenoaks, Neo, and Hanabi joining. It brings Applied Compute's total haul to $160 million, a sum that reflects both the founders' pedigree and perhaps something more: investor hunger for AI tools that don't treat every company like an interchangeable tenant.
Yash Patil, Rhythm Garg, and Linden Li left OpenAI—though not before spending time on agent development, reinforcement learning for reasoning models, and the infrastructure that makes machine learning run at scale. Their San Francisco-based firm now asks a provocative question: What if enterprises stopped renting intelligence from the same models everyone else uses and started building AI that actually learns their business?
The Case Against One-Size-Fits-All AI
Applied Compute calls its approach "Specific Intelligence," and the name does some heavy lifting. Rather than deploying generic foundation models that treat a logistics giant the same way they treat a biotech lab, the company builds what it describes as proprietary agent workforces. These systems ingest a company's institutional knowledge, then specialize using memory and reinforcement learning—all while running inside the customer's own virtual private cloud.
Everything sits behind role-based access controls. SOC 2 certification comes standard. The pitch, articulated bluntly in the company's April announcement, centers on ownership: enterprises need AI that becomes "the advantage you own," not shared intelligence accessible to competitors on equal terms.
It's a compelling narrative. Whether it holds up under scrutiny depends on whether customers see measurable returns—and whether they're willing to pay a premium for bespoke models when off-the-shelf options keep getting cheaper.
Some Early Proof Points
Applied Compute has published case studies that suggest the approach can deliver. DoorDash, for instance, deployed a proprietary agent trained on the company's quality standards across U.S. menu traffic. The result, according to a February case study citing co-founder Andy Fang, was roughly a 30 percent drop in critical menu errors. That's the kind of operational improvement CFOs notice.
Another engagement with Mercor produced a smaller model that ranked first on APEX-Agents for corporate law tasks and fourth overall on Mercor's benchmark—at what the company describes as a fraction of the cost of large models. Cognition publicly endorsed the collaboration at launch. The March announcement name-checked customers spanning financial services, healthcare, logistics, biotech, and cloud hyperscalers, though details on those deployments remain sparse.
Still, two detailed case studies in a market this noisy counts for something. Most enterprise AI startups wave around pilot projects and "letters of intent." Actual performance data, even from a handful of customers, matters.
Valuation Velocity in a Choppy Market
The numbers tell their own story. Applied Compute's valuation climbed from roughly $100 million at its June 2025 seed round—a $20 million raise led by Benchmark's Victor Lazarte—to $1.3 billion by the time of its October 2025 public launch. That launch brought $80 million in backing from Benchmark, Sequoia, Lux, Hanabi, Neo, Definition, and angels including Gil and Omri Casspi's Swish Ventures (yes, the former NBA player).
The latest round maintains that $1.3 billion valuation while adding Kleiner Perkins as lead investor. In a venture landscape where late-stage rounds have grown scarce and down-rounds more common, holding valuation signals continued conviction. As of March 5, 2026, the company had scaled to 20 employees—a lean team for a billion-dollar enterprise.
A Crowded Field, A Distinct Wedge
Applied Compute enters a market thick with enterprise AI agent builders. Sierra has reportedly pursued additional funding at a $10 billion valuation. Harvey raised $200 million at $11 billion in March 2026. Dozens of well-funded competitors are chasing the same Fortune 500 CIOs.
Applied Compute's wedge is architectural: the company bets that proprietary intelligence, continuously refined on a customer's own data and success metrics, will command a premium over generic models that never truly belong to anyone. It's the difference between renting an apartment and owning a home—except the home learns your habits and gets better at predicting what you need.
The startup plans to use the fresh capital to expand its agent development platform and pursue Fortune 500 deployments across new verticals. It has published research on accelerating reinforcement learning with high-leverage samples and continues hiring across research, infrastructure, and engineering.
Whether enterprises will indeed pay for that ownership model—and whether Applied Compute can deliver it at scale—remains the open question. But eight months in, with $160 million raised and a roster of recognizable customers, the company has earned the right to try answering it.
