Gartner has forecast that companies will cancel more than 40 percent of their agentic AI projects by the end of 2027. Interestingly, the reason is not that the technology fails to work. Instead, it usually costs more to run than it actually delivers. In fact, that is arguably the single most expensive problem in enterprise AI right now. So it makes sense that Sapiom AI is betting real money on fixing exactly that. This week, the company closed a 35 million dollar funding round, and it is worth paying attention to, especially if you are building, buying, or writing about enterprise AI.
Who Sapiom is, and why this raise matters
Sapiom AI is formally registered as Sapiom, Inc. Company databases generally describe it simply as a software company. This week, it announced a 35 million dollar Series A. Dragonfly led the round. Accel, Gradient, Coinbase Ventures, Operator Collective, Formus Capital, and VanEck Ventures also joined in. Meanwhile, several existing investors returned too, including Okta Ventures, Menlo Ventures, Array Ventures, and, notably, Anthropic.
This new round arrives just six months after Sapiom’s 15 million dollar seed round. It also comes only eleven months after the company was founded. As a result, total funding now stands at 50 million dollars in under a year.
So why raise this much, this fast? According to founder and CEO Ilan Zerbib, it comes down to a problem most AI teams eventually hit. Teams can build an impressive AI agent demo in days, he explains. However, turning that demo into something that runs continuously, takes real action, and stays affordable is still brutally hard. In other words, the real bottleneck in enterprise AI has quietly shifted. It is no longer about whether you can build the thing. Increasingly, it is about whether you can afford to keep running it.
What Sapiom actually does
To tackle that exact problem, Sapiom launched three products alongside the funding news. Each one targets cost and reliability directly.
First, Sapiom Router sits between an AI agent and the models it calls. It automatically matches each individual call to the cheapest model that can actually handle the task. That way, teams stop defaulting to an expensive frontier model purely out of convenience.
Second, Sapiom Agent Studio gives engineering teams a local environment. There, they can build, test, inspect, and deploy agents directly against their own existing codebase.
Finally, Sapiom Runtime provides the managed production infrastructure where those agents actually run. This covers access control, routing, error recovery, and step level visibility. In short, it is the operational layer needed to run agents at real scale, not just as a one-off prototype.
So far, these three products appear to be working. Since launching six months ago, Sapiom has processed more than 270 million transactions. It now powers over 100,000 agent runs every single day. For example, one customer cut inference costs by 75 percent simply by switching onto the platform. Even more strikingly, a separate customer, an AI-run company called Polsia, saw its projected revenue jump from 100,000 dollars to 10 million dollars in a single year, according to Semafor. Notably, Polsia operates with no human employees at all, running swarms of agents to handle its operations instead.
The awkward part worth naming honestly
That said, there is a genuine tension sitting inside this story. It is worth naming plainly rather than glossing over. Specifically, Anthropic is one of Sapiom’s returning investors in this Series A. In other words, a major AI model maker is directly funding a startup built to help companies spend less on AI model makers.
However, Zerbib has addressed this directly. He frames the relationship as aligned, not adversarial. In his view, cheaper inference simply means companies can afford to build more agents overall. Some of that expanded usage, he argues, will still require the most powerful and expensive frontier models, specifically for tasks that genuinely need them. Even so, it remains worth watching whether that framing holds up as routing technology matures and gets better at avoiding expensive models altogether.
Why this is landing now
To understand the timing, it helps to zoom out first. This raise did not happen in a vacuum. According to PYMNTS, companies spent roughly two years in what one report bluntly called tokenmaxxing. In practice, that meant defaulting employees to the biggest, most expensive AI models with little scrutiny. Now, corporate AI budgets are finally getting their first real, hard audit.
Consistent with that shift, a KPMG survey of more than 2,000 business leaders found something telling. Conducted in April and May 2026, it found that only 7 percent of leaders could point to established, measurable returns from their AI investments so far. Unsurprisingly, some companies have already started capping what individual staff can spend on AI tools.
Given that backdrop, the Gartner statistic from the top of this post stops looking abstract. Instead, it starts looking like the exact business problem Sapiom is betting its funding on solving. After all, if agentic AI projects really are getting cancelled mostly over cost, not capability, then a company built to make agents cheaper, without making them less capable, sits directly at the center of that problem. It is not just adjacent to it.
What this means for anyone building or evaluating AI agents
So what should you actually take from this if you are evaluating agentic AI yourself? First, treat it as a useful, concrete data point. It suggests the harder problem right now is not proving a model can do the work. Rather, it is proving the economics survive contact with real, sustained production usage.
Therefore, before committing budget to an agentic AI project, ask one specific question early. How exactly does the vendor or internal team plan to manage per-task model costs at scale? Do not simply assume a compelling demo will translate into an affordable, durable production system. Often, it will not.
The larger pattern
Taken together, both stories point to the same underlying reality about agentic AI in 2026. The technology itself is genuinely capable. However, the infrastructure needed to run it safely, reliably, and affordably is still being built in real time, often by companies that did not even exist a year ago.
Ultimately, anyone researching Sapiom funding, Sapiom careers, or following Sapiom on LinkedIn right now is watching exactly that story play out. This is a company racing to build the missing infrastructure layer before the wave of project cancellations Gartner is forecasting actually arrives.
This post is part of our ongoing coverage of AI applications in business. For related reading, see our guide to agentic AI in healthcare and the accountability risks it raises. Also see our broader breakdown of artificial intelligence in business.
Frequently asked questions
What does Sapiom AI actually do? Simply put, Sapiom provides infrastructure for building, running, and scaling AI agents in production. Its core product, Sapiom Router, automatically routes each AI model call to the cheapest model capable of handling that specific task, rather than defaulting to the most expensive available model.
Who founded Sapiom and who is its CEO? Sapiom was founded by Ilan Zerbib, who also serves as its CEO. Notably, the company was founded roughly eleven months before its Series A funding round.
How much funding has Sapiom raised? So far, Sapiom has raised 50 million dollars in total. This includes a 15 million dollar seed round and a 35 million dollar Series A led by Dragonfly, with participation from Accel, Gradient, Coinbase Ventures, and others, including Anthropic.
Why are companies cancelling agentic AI projects? According to Gartner, more than 40 percent of agentic AI projects will be cancelled by the end of 2027. The primary reason is escalating operational cost, not the technology failing to work. Supporting that trend, a KPMG survey found only 7 percent of executives could identify clear, established returns from their AI investments so far.
Is Anthropic involved with Sapiom? Yes. Anthropic is listed among Sapiom’s investors. It participated in both earlier funding rounds and the new Series A, alongside firms including Okta Ventures, Menlo Ventures, and Array Ventures.
