Featured Investor | February 2025 - Jake Epstein of Canaan Partners
Written by
Isaac Snitkoff
Jake Epstein is an investor at Canaan, a 37-year-old early-stage venture capital with over $7B in assets under management investing in companies from seed to series B. He focuses on enterprise software and deep tech investments, splitting time between, cybersecurity, infrastructure, AI apps, and various deep tech categories (robotics, energy, space, etc.).
Prior to this, he was a product manager at Uber, most recently leading a team building Uber experiences & marketplace tech for emerging markets. Jake also spent time in product at legacy public enterprise technology companies, working at Pegasystems on low code / no code business process management systems and at Juniper Networks on internal supply chain management tooling. He graduated from Dartmouth where he majored in Computer Science, Engineering, and Economics.
EVCA: Describe a defining moment in your career and how it shaped where you are today.
Jake: One of the most defining - and humbling - moments of my career taught me a lesson that continues to shape how I think about building and investing in great products: deep customer empathy often outweighs technological sophistication.
I learned this lesson early in my career, only a year into working in product at Uber. As a member of Uber's Associate Product Manager (APM) program, I spent three, half year rotations working on various products across the company. My final stint was on a team focused on building the Uber mobility experience for the developing world. After completing the APM program, I stayed on to lead this effort.
This team was largely formed in response to an early-stage Uber competitor called InDrive, which had taken the Latin America and Sub-Saharan Africa rideshare market by storm. Founded in 2012 in Yakutsk, Siberia - primarily known as the coldest city in the world - it began as a local social group for students to find and offer rides to neighboring cities for cheaper prices than taxis. Over the years this was productized into a standalone app with a radically different approach from Uber. Instead of using machine learning to generate prices, InDrive let riders and drivers negotiate fares directly, taking only a small commission.
The InDrive product and business model starkly contrasted with that of Uber, which used an advanced ML-driven system to provide an upfront price to riders and drivers. The underlying ML models were maintained by an army of machine learning engineers and data scientists, constantly tuning them to ensure that when a ride was requested, a driver accepted with a reasonable pickup ETA. From a purely technical standpoint, Uber’s approach was vastly more sophisticated.
Yet in much of the developing world, InDrive was rapidly overtaking Uber (particularly in tier 2 and tier 3 cities). To the product and engineering teams in San Francisco, this was baffling—why was a seemingly rudimentary app, which reintroduced the friction of price negotiation, outperforming Uber?
I travelled to LatAm to conduct user research with riders and drivers and better understand. A quote from a Colombian grandma and frequent InDrive user summed up the resulting learnings well: "I will never use Uber. Not even if it is cheaper, if the cars are nicer, or if I can negotiate. InDrive is a Colombian company".
Of course, InDrive was originally a Siberian (now American) company, not a Colombian one. But they deeply empathized with their users in each unique sub-market they operated in - camouflaging themselves with excellent marketing, local operations folks on the ground, and with an experience that their customers actually enjoyed. Uber's product, based on what was most efficient and suited for consumer taste and preferences in the U.S., overfit markets where upfront pricing was hard to do reliably and there was a cultural preference to negotiate directly.
This lesson has shaped my perception of company building and investing. Successful companies are built on good product, and good product is built with deep knowledge and empathy of one's customers. Great products can often beat out incumbents, even with comparatively little capital and when founded outside Silicon Valley, by getting this right from the start. Product-market fit isn’t just about solving a problem; it’s about solving it in a way that aligns with local behaviors, expectations, and values.
Today, whether I’m building products or evaluating startups, this remains my guiding principle. The companies that win often aren’t the ones with the most advanced technology, but the ones that get customer empathy right from day one.
EVCA: What is an emerging technology trend that will have a significant impact on the world in the next decade?
Jake: We often discuss AI’s end applications, such as autonomous agents and “service-as-software” products that replace white-collar labor, but we overlook the critical resource that fuels these systems: data.
For years, AI models have been drinking from a deep well of freely available internet text, improving their performance as they consumed more. But that well is drying up. High-quality data is finite, and much of what remains is polluted - redundant content, SEO spam, and synthetic noise. Despite rising training costs, AI’s reasoning improvements are plateauing because the supply of clean, usable data is running out.
Some believe synthetic data will replenish this well, but it’s more like desalination - costly, imperfect, and unlikely to quench AI’s growing thirst. Training models on AI-generated outputs risks reinforcing biases and may not meaningfully improve reasoning.
More likely, as AI models consume fresh data faster than humans create it, access to proprietary, high-quality datasets will become the new competitive advantage. Just as water scarcity has driven geopolitical conflicts, we are heading toward data wars, where companies will race to acquire and control these vital resources. Whoever holds the rights to proprietary datasets - or the means to extract and refine them - will dictate the future of AI.
Much of this critical data exists behind corporate walls, on billions of consumer devices, and in human expertise. Unlocking it will require new methods - data pipelines that tap into novel sources, filtration systems that remove noise, and infrastructure that ensures a steady, reliable flow. Techniques like expert marketplaces, federated learning, and reinforcement learning from human feedback will become essential tools for extracting and refining the next generation of AI’s most valuable resource.
The AI boom has been fueled by an era of abundant, freely available data. That era is ending. The companies that survive the coming data drought will be those that control, purify, and distribute the cleanest water - because in AI, as in history, power belongs to those who control the well.
