Vipul
Ved Prakash

The Razor's Edge

Before most people had heard the word "spam," Vipul Ved Prakash had already written the antidote. In the early 2000s, from a cramped apartment somewhere between Napster's implosion and the dot-com wreckage, he built Vipul's Razor - a collaborative, distributed spam-filtering system that used network intelligence instead of keyword rules. The internet had a poison problem. He wrote the immunization.

The open-source community liked it so much they named it after him. By 2003, MIT Technology Review had named Prakash to their annual Top 100 Young Innovators Under 35. He was 25. That honor, historically, goes to people who have spent years assembling the right resume. Prakash earned it by solving an actual problem that was breaking actual inboxes.

What followed from Vipul's Razor was Cloudmark - a company he co-founded with Jordan Ritter, Napster's co-founder, to commercialize the anti-spam technology. It was a telling choice of collaborator: Ritter had just helped build the most disruptive file-sharing network on the planet, and Prakash had been an engineer there during its peak controversy. These were people who understood networks as weapons, tools, and communities all at once.

A Table Tennis Player Who Dropped Out of One of India's Best Colleges

Born in New Delhi in 1977, Prakash came up in an India that was in the middle of economic liberalization and a technology awakening. He studied mathematics, physics, and computer science at St. Stephen's College, Delhi - one of the country's most prestigious institutions - and then dropped out. Not from lack of ability; the lectures were simply slower than what he could build.

Before that, he had been a competitive professional table tennis player for the Food Corporation of India. Precision under pressure, pattern recognition, fast reflexes. The sport left traces. His career - a sequence of early and decisive bets on what the internet actually needed - suggests someone who reads the game several moves ahead and moves before the moment is obvious.

After leaving college, he co-founded Sense/Net with Ashish Gulhati in the late 1990s, an internet privacy company, and wrote a technology column for PC World Magazine India called "Net Zeppelin." He was also a cypherpunk - part of a community of cryptographers and privacy activists who believed that strong cryptography was a civil right. His most famous act of civil disobedience in this arena: he encoded the RSA encryption algorithm as a dolphin-shaped Perl script, which was printed on a T-shirt and sold by ThinkGeek. The T-shirt was technically export-controlled software. That was the point.

Topsy: Real-Time Social Search Before Google Had the Answer

In 2007, Prakash founded Topsy. The premise: the real-time web - Twitter, blogs, forums - was producing information faster than any index could process it. Google was built for a world where pages were static and links were votes. Topsy was built for a world where a tweet from three minutes ago was more relevant than an article from three years ago.

The company became the definitive real-time social search engine. It indexed every public tweet - billions of them - and built tools that let analysts, marketers, and journalists understand what was happening on the internet right now. In 2013, Apple acquired Topsy for over $200 million.

Prakash stayed at Apple for five years, becoming Senior Director of Engineering with responsibility for search technology across Spotlight and Siri. In 2015, he walked on stage at WWDC to introduce iOS Search APIs to developers - a rare public moment for someone who otherwise operates most productively out of view. By 2018, he was done. The infrastructure problems he wanted to solve were elsewhere.

Together AI: The Open-Source AI Factory

On June 11, 2022, Prakash co-founded Together Computer Inc. - now known as Together AI - alongside four Stanford and CMU researchers: Ce Zhang (CTO), Chris Re, Percy Liang, and Tri Dao. The founding lineup is worth pausing on: Tri Dao is the author of FlashAttention, one of the most important algorithmic advances in transformer efficiency. Percy Liang runs Stanford's Center for Research on Foundation Models. Chris Re is a MacArthur Fellow and professor who has shaped modern machine learning infrastructure. This was not a team assembled by a recruiter. It was a thesis.

The thesis: that the AI era would be defined not by whoever built the biggest proprietary model, but by whoever built the best open-source infrastructure for running, fine-tuning, and deploying foundation models at scale. Together AI would be the cloud for open AI - the place where a three-person startup could access the same quality of GPU infrastructure and model acceleration as a hyperscaler, without the budget or the waiting list.

It worked faster than almost anyone expected. By February 2025, Together AI had raised $305 million in a Series B led by General Catalyst, co-led by Prosperity7, with participation from NVIDIA, Salesforce Ventures, Kleiner Perkins, Coatue, and more than a dozen other institutional investors. The valuation: $3.3 billion. Total funding across all rounds: $533.5 million. Annualized revenue: $100 million, crossed right before the round closed.

The client list tells the story: Salesforce, Zoom, Zomato, ElevenLabs, Hedra, Cartesia. These are not experimenters. They are production-scale AI companies running revenue-critical workloads on Together's infrastructure. The platform now supports custom training, fine-tuning, batch inference (at 50% cost reduction versus standard), and GPU cluster services - all built on top of research innovations including FlashAttention-4 and the ATLAS runtime, which delivers up to 4x faster LLM inference.

What He Keeps Building Toward

There is a through-line in Prakash's career that goes beyond the exits and the funding rounds. Vipul's Razor was open-source. Topsy was built on public data. Together AI is defined by its commitment to open-source models and open research. The pattern is a particular kind of idealism - the belief that the most valuable things on the internet become more valuable when they are accessible to everyone, not locked behind a corporate API key.

"If you look at the next 10 years or the next 20 years," he said in a 2025 interview, "we are doing maybe 0.1 percent of the AI that we'll be doing 10 years from now." For someone sitting at a $3.3 billion company with $100M in annual revenue, that's a remarkably humble framing. Or maybe not: a man who built the internet's spam filter when nobody was paying attention is exactly the kind of person who knows that the interesting part of a curve is always what comes next.

In May 2026, Prakash appeared at the Milken Institute Global Conference in Los Angeles to speak on "AI Inside the Enterprise: Redefining Competitive Advantage." The conversations have changed - from startups to sovereign nations, from developers to Fortune 500 boardrooms. The underlying argument has not: open infrastructure beats closed systems, distributed beats concentrated, and the future of AI does not belong to anyone who hoards it.

He is building a company that means it.