Sarvam AI will host Epoch, its public AI conference, on July 30–31 in Bengaluru. The company says the event will feature model updates and new product launches. A teaser posted on X frames Epoch as a technical gathering — “a full pass on the data,” in machine‑learning parlance, according to Caleb Friesen, who produced the video.
Format and anticipated announcements
Epoch is structured as two distinct tracks: a Builder Edition for developers and partners, and an Enterprise Edition focused on enterprise adoption. Sarvam says the Builder Edition will be the forum for new model and product announcements, research presentations, and technical exercises. The program will include hands‑on workshops, technical deep dives, and live demos, while the Enterprise Edition will show how to integrate these tools into organizational workflows.
Models and performance highlights
Sarvam already publishes several models, including Sarvam‑105B and Sarvam‑30B, and a Vision model that has performed well on document reading for Indian languages. In May, the company cut the price of its Vision model by 67% to encourage developer use; Sarvam reports that the platform has helped digitize more than 35 million pages.
Funding, compute costs, and public‑private partnership
Epoch follows a $234 million funding round that valued Sarvam at $1.5 billion, led by HCLTech, which committed $150 million as a strategic investor. On infrastructure, co‑founder and CEO Pratyush Kumar said the company spent “about $20 million” on compute across three training cycles — including for Sarvam‑105B — clarifying this is a cumulative figure rather than a single run. He added that this level of spending is “well below” that of some Chinese competitors, comments made on an NPCI podcast.
A significant portion of the compute budget relied on a public‑private arrangement: Sarvam was selected by India’s sovereign AI mission, which provided access to 4,096 Nvidia H100 GPUs for six months in exchange for an equity stake.
Usage and large‑scale deployments
Sarvam cites meaningful usage metrics: more than 2 million daily interactions, roughly 10 million API requests per day, and over 500,000 hours of audio transcribed monthly. Notable deployments include multilingual voice agents used in a government program to collect information from 17 million farmers, and a national insurance campaign that reached 45 million policyholders.
What to expect from Epoch
Epoch is an opportunity for Sarvam to demonstrate technical progress, detail product updates, and present concrete enterprise and public‑sector use cases. The Builder Edition is aimed at operational announcements for developers and partners; the Enterprise Edition is designed to translate those technologies into tangible workflow gains for decision makers.
Bottom line
Sarvam Epoch will try to accomplish two goals: show measurable technical advances and prove the startup can industrialize AI at scale — an especially salient claim after its recent funding round and privileged access to national compute resources.
