Early Groq Backer Anticipates Half Of Her Investments To Fail

Sandhya Venkatachalam spent the initial phase of her career developing technology companies. She served as head of product for an early data center hardware firm that was acquired by Cisco, and subsequently worked as a product executive at Skype prior to its acquisition by Microsoft. These positions immersed her in data and machine learning long before artificial intelligence became the primary focus of venture capital.

Later, she joined Social Capital as a general partner, heading early institutional investments in the AI chip manufacturer Groq, and subsequently invested at Khosla Ventures. Presently, she operates as the founder and managing partner of Axiom Partners, a $52 million fund that finances startups utilizing AI to execute operations across sectors such as construction, industrials, and insurance.

During an interview with Crunchbase News, Venkatachalam explores her reasons for looking past traditional founder backgrounds, the attributes that create durability in an AI enterprise, and the ways in which her early investment in Groq influenced her methodology.

This interview has been edited for clarity and brevity.

You invested at Khosla Ventures before starting Axiom. What did you take from that experience, and what did you want to do differently?

Venkatachalam: One takeaway was Vinod Khosla’s expansive perspective regarding the origins of exceptional founders. Silicon Valley has typically favored a relatively restricted profile for individuals capable of creating the next major AI enterprise: someone possessing a Stanford computer science or machine learning pedigree, or background experience at OpenAI. We actively search for nonobvious founders, specifically within nonobvious sectors.

Another lesson involved how we approach risk assessment. Instead of subjecting a company to every conceivable diligence inquiry, we concentrate strictly on the risks critical to hitting its next set of milestones. Can this team execute its stated plans? And if so, could the ultimate impact be massive?

That requires accepting that numerous wagers will fail while striving for outliers. I believe my investors back me precisely for this reason: to identify future categories rather than merely participate in the sectors everyone already recognizes.

At Axiom, the fundamental distinction is that we structured the firm around individuals who are actively engaged with AI. If you are not continuously building, productizing, pricing, or bringing AI to market, maintaining relevance is extremely difficult. Our roster includes professionals doing precisely that in their separate roles. They help keep our investment perspective current, and founders appreciate working with them because they have navigated many of the exact same hurdles.

Additionally, we incorporate AI across the entire firm. We developed an internal system called the Axiom Brain to interpret market trends, discover interesting talent and companies, and accelerate due diligence and other operations. For me, the true value lies in the capacity to act rapidly.

You mentioned that some of those AI practitioners have other jobs. How does their role at Axiom work?

Venkatachalam: They dedicate specific time to Axiom on a part-time basis and also receive carried interest in the fund. They function as active partners in the work rather than just names listed on an advisory roster.

Their external occupations form the core of this model. Some of the finest angel investors are individuals actively operating and building in the field. I do not require these professionals full-time; in fact, they would offer less value to Axiom if they abandoned the daily work that keeps them close to the market.

Axiom says it invests in “AI for the real world.” What does that mean when you’re evaluating a startup?

Venkatachalam: Our perspective dictates that AI should benefit a much wider population than simply the early adopters already utilizing it. We examine industries underserved by modern technology, where AI can deliver concrete outcomes rather than merely supplying another software tool.

That pursuit may direct us toward construction, industrials, or insurance. Certain companies in this space incorporate hardware, sensors, or robotics, while others remain entirely software-based. The common thread among them is that they execute work critical to customers within real-world industries.

Generally, we steer clear of products resembling conventional enterprise software tools. We want evidence of AI delivering tangible results.

You’ve described a shift from software people use to digital workers that perform jobs. Are customers actually paying for AI from labor budgets?

Venkatachalam: Yes, and that serves as one of our core investment criteria. Even when a portfolio company operates at an alpha or design-partner phase, we conduct due diligence to confirm whether customers are prepared to purchase it under those terms. We frequently observe contract values reaching the hundreds of thousands of dollars, eclipsing the much smaller agreements typical of midmarket software tools.

We have witnessed this purchasing behavior manifest across the majority of the portfolio companies we have backed.

AI products are becoming faster to build and easier to imitate. What makes one durable enough to become a large company?

Venkatachalam: If you are executing critical work inside a customer’s organization, and that operational work holds substantial financial value, you become difficult to displace. You are managing what we designate as the last mile of the assignment.

Within industrial contexts, for instance, delivering an outcome requires deep integration with the customer’s existing systems. You must comprehend their data, train models on it, master essential workflows, and stand behind the output. Accomplishing that demands far more than merely applying an interface on top of a foundational model.

Such relationships and capabilities prove challenging for competing startups to duplicate. They also involve labor that major AI model developers may have little desire to handle themselves.

Before Axiom, you backed Groq when AI inference was far from an obvious investment category. What led you to it?

Venkatachalam: My background spans both hardware and software. At a certain point, I grew curious about why Google manufactured its own networking switches instead of purchasing them from established vendors. Investigating that inquiry revealed that the company was developing its own proprietary chips as well.

That discovery led me to Jonathan Ross, who had participated in those initiatives before departing to establish Groq. I began exploring why major tech corporations were building chips specifically to train models. Subsequently, Jonathan presented the argument that the much larger future market would center on inference.

To be entirely transparent: back in 2016, I scarcely understood inference. However, if one believed these models would proliferate, it stood to reason that individuals would build applications on top of them and require the underlying infrastructure to support that growth. That insight motivated my investment.

How did that experience shape what you look for now?

Venkatachalam: It demonstrated the value of arriving slightly early and exercising patience. You do not need to be wildly contrarian, but you must recognize an opportunity before it becomes apparent to everyone else.

In a sense, our foundational thesis remains unchanged. We continuously question what will be constructed on top of AI infrastructure and models. We seek to invest while the answer is still emerging, prior to the establishment of a consensus.

What happens when one of those early bets doesn’t work out?

Venkatachalam: We factor that into our planning. Operating a $52 million fund, we anticipate making approximately 35 investments. We fully expect roughly half of those bets to fail, whether that entails a company shutting its doors or simply failing to achieve the growth trajectory we seek.

Our model relies upon uncovering an exceptional outcome. We require one outstanding investment to return the entire fund. Investing early enough to catch a company that expands significantly can easily compensate for numerous wagers that fail to pan out.

That willingness to tolerate losses forms an essential component of backing opportunities before they become obvious. It is inherently integrated into how we manage the fund.

Related Crunchbase query:

• Global Venture Funding To AI Startups In 2026

Illustration: Dom Guzman

?Frequently Asked Questions

01Who is Sandhya Venkatachalam?

Sandhya Venkatachalam is the founder and managing partner of Axiom Partners, a $52 million venture fund. She is a former general partner at Social Capital and Khosla Ventures, and an early investor in AI chipmaker Groq.

02What is Axiom Partners?

Axiom Partners is a $52 million venture capital fund that invests in startups utilizing artificial intelligence to execute work in traditionally underserved, real-world industries such as construction, industrials, and insurance.

03What percentage of Axiom Partners’ investments are expected to fail?

Venkatachalam notes that out of an anticipated 35 investments from their $52 million fund, they fully expect about half to fail or shut down, relying on exceptional outliers to return the fund.

---