CosavuAbout Cosavu
Making AI more efficient
for companies,
sustainable for the world.
Vision
The cost of
scale.
One AI query — from prompt to output — can cost around $0.18 on average in enterprise workflows. ChatGPT alone is estimated to run over 2.5 billion queries per day. Even if each query uses only about 0.3 ml of water and a small amount of compute behind the scenes, at global scale the impact becomes massive.
A daily AI conversation may look simple to the user, but underneath it, data centers are using electricity, compute, and fresh water to keep the system running.
AI adoption is still early. If adoption grows from where it is today to much wider usage across enterprises, consumers, and governments, the world will need more data centers, more compute, more energy, and more water.
Cosavu was built with this problem in mind. Our vision is to make AI adoption more efficient at scale — by reducing unnecessary tokens, wasted inference, and repeated compute, we help companies use AI with less cost, less infrastructure pressure, and lower resource waste.
Mission
From problem
to infrastructure.
The idea for Cosavu was introduced by Arun Teja in 2025. Arun met Thishyaketh in 2024 and was inspired by his journey — starting TNSA at the age of 13, building foundational LLMs, and working on AI service development. Arun later introduced Thishyaketh to Satya Komal, who had a strong enterprise client network and helped bring real AI development opportunities to TNSA.
Arun and Thishyaketh started managing operations and working closely with enterprise AI projects. During this time, Arun was involved in customer success engineering and started noticing the same bottlenecks again and again in production AI workflows: context, data security, cost, and latency.
The demos worked. But production was different.
When AI workflows touched real enterprise data, changing user needs, permissions, and business context, the systems became harder to scale. The problem was not just the model. It was the context around the model.
Satya Komal then invested in starting deeper research around this problem. Arun and Thishyaketh began working on a solution at the architecture level, which eventually became STAN.
After building STAN, the team continued validating it with enterprise workflows and customer conversations. The conclusion became clear: context control was not just a feature. It was infrastructure needed for enterprise AI to scale in production.
Cosavu was created from that realization. STAN and Cosavu were built with high-scale inference, token efficiency, compute cost, and production AI workflows in mind. The three of us identified this problem through real enterprise work, validated it with industry professionals, and then built Cosavu to solve it.
Foundation Team
The foundation team for Cosavu.
Arun Teja
Co-founder
Arun brings enterprise sales and GTM experience across security SaaS and BFSI. He has sold security products to Fortune 500 clients and helped scale revenue in the US market to a million within the first two quarters. At Cosavu, Arun shaped the original problem statement after seeing production AI bottlenecks across customer workflows, especially around context, cost, security, and latency.
Thishyaketh
Co-founder
Thishyaketh founded TNSA and has worked deeply on pretraining LLMs, attention mechanisms, foundational models, and open-source AI systems. He is also a quantum researcher working across AI and mathematics. He built and open-sourced foundational LLM work, created the OpenArchX ML framework which received contributions from Google, and published research on attention mechanisms at NeurIPS and PNAS. After Arun shared the problem statement and the team validated it through multiple client conversations, Thishyaketh started the research and built STAN.
Satya Komal
Founding Investor
Satya Komal is an investor and entrepreneur with over 20 years of experience in building and scaling technology ventures. He has invested in more than 15 startups, including a unicorn, and has cultivated a strong network across the IIT ecosystem and Silicon Valley. For Cosavu, Satya played a pivotal role in connecting the team with enterprise customers and enabling them to work on production-grade AI workflows through TNSA. He funded the early research that evolved into Cosavu and supported the development of STAN, with a strong focus on enterprise-grade security, compliance, and scalable deployment for large organizations.
The name
What does
Cosavu mean?
The name Cosavu came from Arun.
“Co” stands for conserve.
“Savu” is a Finnish word for smoke.
For us, Cosavu means conserving the tokens, compute, water, and energy behind every AI workflow.
Join us
Want to build the
infrastructure with us?
Today, Arun and Thishyaketh bring both market and technical experience from the enterprise clients and projects made possible through Satya's network.