Manu
Sharma

Manu Sharma grew up in industrial towns across northern India, the only child of a textile engineer father who moonlighted as a DIY scientist. He built things in their home workshop. He took apart machines. He was fifteen when he got his first computer - late enough that he remembers the world without one, early enough that software became a second language before he had a job.

He left Roorkee for Embry-Riddle Aeronautical University in the United States, then Stanford for a master's in aerospace engineering. At some point between differential equations and aerodynamics labs, he launched a camera to the edge of space using a weather balloon and less than $200 in parts. The kind of move that tells you more about a person than a resume does.

Before graduation he had already co-founded two companies. Ardulab - later Infinity Aerospace - created an open-source research platform for conducting science experiments aboard the International Space Station. Nuovo Wind, a renewable energy startup, was named one of the world's 50 most promising startups within a year of its founding. Both ventures dissolved, as student ventures do. The pattern they established - identify an unsexy infrastructure problem, build a platform, attract a community - did not.

After Stanford, Manu joined DroneDeploy in 2013 as one of its earliest engineers. His signature project there: CoPilot, a system that let pilots fly drones over the internet using LTE connectivity. Controlling an aircraft through a cellular network was not a solved problem in 2013. He solved it, then moved into product management, where he co-created the DroneDeploy App Marketplace alongside Daniel Rasmuson - a colleague who would later co-found Labelbox with him.

In 2017, he moved to Planet Labs to lead its data analytics platform. Planet was doing something outrageous: imaging Earth's entire landmass daily using hundreds of satellites in low Earth orbit. Manu's job was to make that raw imagery useful - to build computer vision and analytics products that could answer questions from it. Agriculture, insurance, defense, intelligence. The problem was always the same: raw data is worthless without a system for labeling, contextualizing, and iterating on it.

That insight - that data quality is the actual bottleneck, not compute or model architecture - became the founding thesis of Labelbox. In March 2018, Manu co-founded the company with Brian Rieger and Daniel Rasmuson, both former DroneDeploy colleagues. All three had lived inside the friction of building ML pipelines without proper tooling. They knew exactly which problem they were solving because they had scraped their knees on it.

The world is not short of companies claiming to be the picks-and-shovels play in the AI gold rush. Labelbox is one of the few with a genuine claim. The platform started as annotation tooling - software that makes it easier for human reviewers to label images, video, text, and audio for machine learning training sets. That is the product people know.

But annotation is now the least interesting part of what Labelbox does. As foundation models displaced the need for training from scratch, Manu recognized the shift early and repositioned the company accordingly. The real problem - the one enterprises and frontier labs alike are willing to pay for - is post-training data: RLHF datasets, synthetic data generation for frontier models, expert human evaluation pipelines, and model benchmarking infrastructure.

The thesis Manu has been articulating since 2023 is about the bifurcation of the AI ecosystem. A small number of organizations build base models. Everyone else rents intelligence from those models and adds layers on top. Both groups need Labelbox: the builders need massive amounts of precisely structured training and evaluation data; the renters need tooling for fine-tuning, alignment, and quality assurance. The market is, in his words, the $100 billion post-training race.

The aerospace thread in Manu's biography is not ornamental. He holds a private pilot's license, earned at Mojave Spaceport in California - a working commercial spaceport where Burt Rutan once built experimental aircraft and Dick Rutan trained for the Voyager circumnavigation.

Rutan himself - the man who co-piloted Voyager on the first non-stop, non-refueled circumnavigation of the Earth in 1986 - mentored Manu in flight training. It is an unusual credential for a software CEO. It is also precisely the kind of detail that explains why a person builds infrastructure companies: you become a pilot because you want to understand the whole system, not just the part you control.

The angel investment portfolio follows the same logic. Manu has backed Replicate, LangChain, and Together.ai - three companies that collectively represent the tooling, orchestration, and compute layers of the AI stack that Labelbox sits inside. He is not hedging. He is mapping the system.

The Labelbox cap table is a useful mirror for the company's trajectory. The earliest institutional money came from Andreessen Horowitz, then one of the few major firms willing to back infrastructure plays in the pre-ChatGPT AI world. B Capital Group followed in subsequent rounds.

The Series D is the interesting one: $110M led by SoftBank Vision Fund 2 in January 2022, joined by Snowpoint Ventures, Databricks Ventures, and Cathie Wood's ARK Invest. Databricks' participation - investing directly in the data infrastructure company whose users also use Databricks - is the clearest statement of strategic alignment in the round. SoftBank's involvement places Labelbox in a portfolio of companies SoftBank believes will define their respective infrastructure categories globally.

ARK's participation from Cathie Wood's fund is, as ever, a bet on transformative technology platforms. Labelbox qualifies as exactly that: a company whose value grows with the overall pace of AI adoption.