Axiom Partners' $52M Fund Plans for Half Its Bets to Fail

Sandhya Venkatachalam expects roughly half of the 35 investments her $52 million fund will make to fail — and she has built Axiom Partners' entire model around that assumption.

"We fully expect about half of them to fail, whether that means a company shuts down or simply never reaches the growth trajectory we're looking for," Venkatachalam, founder and managing partner of Axiom Partners, told Crunchbase News. "The model depends on finding an exceptional outcome. We need one outstanding investment to return the fund."

That arithmetic is standard power-law venture investing, but Venkatachalam brings an unusual pedigree to it. She spent the first half of her career building technology companies: she led product at an early data center hardware company sold to Cisco, then worked as a product executive at Skype before its sale to Microsoft. Those roles pulled her into data and machine learning years before AI became venture capital's dominant theme.

She later became a general partner at Social Capital, where she led early institutional investments in AI chipmaker Groq, and then invested at Khosla Ventures. Now she runs Axiom, a fund backing startups that use AI to do work in industries such as construction, industrials and insurance.

Against the narrow founder profile

Venkatachalam credits Vinod Khosla with shaping how she thinks about founders. "Silicon Valley has gravitated toward a fairly narrow idea of who can build the next great AI company: Someone with a Stanford computer science or machine learning background, or experience at OpenAI," she said. "We're looking for more nonobvious founders, particularly in nonobvious industries."

She also took a different approach to risk from her time at Khosla Ventures. Rather than asking "every conceivable diligence question," Axiom focuses on the risks that matter for a company's next set of milestones. Two questions drive the decision: Can this team do what it says it will do? And if it can, could the result be massive?

"I think that's what my investors are backing me to do: identify categories of the future, rather than participate in the categories everyone already recognizes," she said.

Axiom's structure is another departure. The firm is built around people who are actively working with AI in their day jobs — part-time partners with dedicated time at Axiom who receive carry in the fund.

"They're partners in the work, rather than people whose names appear on an adviser list," Venkatachalam said. "Some of the best angel investors are people who are still operating and building. I don't need these people full time. In fact, they would be less valuable to Axiom if they left the work that keeps them close to the market."

The firm also applies AI internally. It has built what it calls the Axiom Brain to help track market trends, identify interesting people and companies, and accelerate diligence. "For me, the value is the ability to act quickly," Venkatachalam said.

Selling work, not software

Axiom markets itself as investing in "AI for the real world." In practice, that means AI should "benefit a much broader population than the early adopters who are already using it," Venkatachalam said. The fund targets industries underserved by technology, where AI can produce an outcome rather than deliver another software tool.

Some portfolio companies involve hardware, sensors or robotics; others are entirely software-based. "We generally don't invest in products that look like conventional enterprise software tools," she said. "We want to see AI delivering a result."

That thesis is already showing up in contract sizes. Even at alpha or design-partner stage, Axiom diligence examines whether customers will buy AI from labor budgets rather than software budgets. "We're often seeing contract values in the hundreds of thousands of dollars, rather than the much smaller contracts you might expect for a midmarket software tool," Venkatachalam said. "We've seen that buying behavior play out across the majority of the portfolio companies we've invested in."

On durability, her argument is about depth of integration. "If you're doing important work inside a customer's business, and that work is worth a lot of money, you become difficult to replace," she said. In industrial settings, that means understanding a customer's data, training on it, learning the workflows that matter, and standing behind the result — "more than putting an interface on top of a model." Those relationships, she argued, are hard for another startup to replicate and involve work that the large AI model companies may have little interest in doing themselves.

The Groq lesson

Venkatachalam's early Groq investment began with a hardware question: why Google was building its own networking switches when it could buy them from existing suppliers. That inquiry led her to Jonathan Ross, who had worked on Google's chip efforts and left to found Groq. Ross made the case that the much larger future market would be inference, not training.

"I'll be honest: In 2016, I barely understood inference," Venkatachalam said. "But if you believed these models would spread, it made sense that people would build on top of them and need the infrastructure to support that. That insight drove my investment."

The experience taught her "the value of being a little early and having some patience." Axiom's thesis, she said, has not really changed: "We're still asking what will be built on top of AI infrastructure and models. We want to invest while the answer is emerging, before there's a consensus."

For Axiom's limited partners, that means accepting a fund designed to absorb losses in exchange for a shot at the outlier. One exceptional outcome, Venkatachalam said, can offset many bets that didn't work — and if Axiom repeats anything like its Groq timing, the math will take care of itself.