October 14, 2012. Felix Baumgartner stands at the edge of a capsule 39 kilometres above New Mexico
and steps off. Below him, at YouTube headquarters in San Bruno, Varun Talwar watches the traffic monitors.
Eight percent of all internet traffic is watching this jump. Not 8% of YouTube.
Eight percent of the entire internet. The infrastructure doesn't blink.
That moment didn't make Talwar famous. It made him curious. What happens when 8% of the internet
tries to flow through a pipe that was never designed for it? What does that reveal about the
gap between how we build distributed systems and how distributed systems actually behave under pressure?
He spent the next four years finding out. First by building gRPC at Google Cloud -
the open-source remote procedure call framework that now handles service-to-service communication
at Google, Netflix, Square, and thousands of companies you've never heard of but whose products
you use every day. Then by co-creating Istio, the service mesh that became the de facto standard
for Kubernetes networking.
"Solve a hard problem and make it easier for people to adopt the technology."
- Varun Talwar
The philosophy is deceptively simple. Hard problems, in infrastructure, are always the same problem
wearing different clothes: something critical is invisible, and the cost of it being invisible
only becomes obvious after something breaks. gRPC made inter-service communication predictable.
Istio made the space between services observable, manageable, and secure.
Both projects were, at their core, bets on making the hard invisible into the easily auditable.
In 2018, Talwar and co-founder JJ Jeyakeerthi - who had led Twitter's Cloud Infrastructure
Management platform - left to build Tetrate. The thesis: enterprises needed what Google had,
and they needed it without building a decade of internal expertise first.
Service mesh was Google's secret infrastructure advantage,
handed to the world as open source. Tetrate would be the company that made enterprises
capable of actually using it.
Four years and $52.5 million later, Tetrate runs inside the US Air Force's wartime communications
infrastructure. The path from a teenager winning 80% of Delhi programming contests to
securing military communications is, in retrospect, a straight line - even if it ran through
Singapore, MIT, YouTube, and two open-source revolutions to get there.
Now in 2025, Tetrate has pivoted again, this time into AI. The Envoy AI Gateway - co-released
with Bloomberg and the first CNCF-backed AI gateway - brings the same philosophy to the era
of large language models: make the hard invisible auditable. Where are the token costs flowing?
Which model is handling which request? If an AI agent calls another AI agent calls an API
that calls a database, who authorized that chain?
"Envoy is rapidly becoming the community of choice for AI innovation."
- Varun Talwar, February 2025
Talwar's career has always been about the layer no one sees until it fails.
In 2012 that was livestreaming infrastructure. In 2016 it was service mesh.
In 2025 it's AI governance. The bets get bigger each time, but the instinct is the same:
find the invisible, make it visible, and ship it before the bill comes due.
