When enterprise AI gets messy—and lately, it always does—there's money to be made in the middleware. Portkey, a San Francisco outfit barely past its second birthday, closed a $15 million Series A round in February 2026 betting that companies drowning in multi-model deployments will pay handsomely for someone to impose order on the chaos.
Elevation Capital led the round, with Lightspeed—Portkey's original seed backer—participating again. The startup, which counts Postman and Snorkel AI among its customers, says it now processes north of 500 billion LLM tokens daily across more than 24,000 organizations. That translates, the company claims, to roughly $180 million in annualized AI spend flowing through its infrastructure.
The pitch is straightforward enough: enterprises need a control layer. Something between their applications and the dizzying array of AI providers that have proliferated over the past two years.
The Plumbing Problem
Portkey handles 125 million requests per day, routing over $500,000 in daily AI spending across what it says are more than 60 providers and some 1,600 models. For perspective, that's the kind of sprawl that gives finance teams nightmares and engineering teams whiplash.
Founders Rohit Agarwal and Ayush Garg—both veterans of Freshworks and Pepper Content—raised a $3 million seed from Lightspeed on August 23, 2023. Eighteen months is a quick turnaround to Series A by any standard; perhaps quicker than even they anticipated when the company was founded in 2023. Then again, enterprise AI infrastructure has been on a tear, and Portkey caught the wave early.
The platform itself functions as what the industry calls a "gateway"—sitting in the path of AI traffic to manage routing, observability, governance, and cost controls. Snorkel AI, which builds tools for data labeling and model training, uses Portkey's observability stack to evaluate model performance. That's a validation of sorts: selling infrastructure tooling to companies that themselves sell infrastructure tooling.
Governance Gets Real
Much of the fresh capital will fund what Portkey calls "governance for agentic workloads." Translation: as companies deploy autonomous AI agents that act on their own, someone needs to set guardrails. Permissions, identity boundaries, access controls, budget caps. The kinds of controls that sound boring until an agent starts racking up six-figure API bills or accessing data it shouldn't.
The company is also investing in go-to-market—which is startup-speak for hiring salespeople—and building out faster, lower-latency infrastructure for production use cases. Speed matters when you're sitting in-path of every AI request a customer makes.
One move stands out. Portkey is making its core enterprise gateway free, betting it can acquire users at scale and monetize through advanced features: observability exports, compliance tooling, private cloud deployments. It's a familiar SaaS playbook, though not without risk. Free tiers can attract tire-kickers as easily as serious customers.
A Crowded Field
Portkey isn't operating in a vacuum. The AI infrastructure layer has gotten crowded fast. Cloudflare offers an AI Gateway. LiteLLM and Helicone are in the mix. OpenRouter competes on similar turf. Portkey's argument—rightly or not—is that it's a unified platform rather than duct-taped point solutions.
The pricing structure reflects standard enterprise software tiers: free for developers, $49 monthly for production workloads with usage overages, and enterprise contracts that bundle SOC 2 compliance, HIPAA support, and custom business associate agreements. It's the enterprise tier where real revenue lives, naturally.
With total funding now at $18 million and a team LinkedIn pegs between 11 and 50 employees, Portkey is still in the early buildout phase. But the raise signals investor confidence that managing multi-model AI complexity isn't a fleeting problem. As enterprises grapple with the operational realities of production AI—tracking costs, preventing runaway spend, imposing governance on systems that increasingly act autonomously—the need for infrastructure that sits between application and model layer looks less like a nice-to-have and more like table stakes.
Whether Portkey becomes the default control plane or just another vendor in a fragmented market remains to be seen. For now, though, the company has capital and momentum. And in infrastructure, timing matters almost as much as technology.
