Sarvam AI has hired Sashikumar Sreedharan as its president and chief business officer, according to reports . Until recently, Sashikumar led Google Cloud’s business in India. Before that, he served as managing director of Microsoft India’s enterprise business, giving him experience selling to Indian CIOs on behalf of two of the three largest cloud providers. The appointment suggests that one of India’s best-funded foundation-model startups increasingly sees commercialization and enterprise distribution as a critical constraint.

Over the past year, Sarvam has shown it can train foundation models in India. It remains to be proven that enterprises and governments will pay for them at a scale that matches their valuation. Closing that gap will be Sashikumar’s key priority, and he arrives at a time when Sarvam is quietly redefining what it sells.

In June, Sarvam announced a $234 million first close of a planned $300 million Series B at a $1.5 billion post-money valuation. HCLTech committed $150 million as the lead strategic investor.

The revenue base is far smaller. Inc42 reported that Sarvam booked ₹45.1 crore, roughly $5 million, in FY26 revenue. That puts the valuation at nearly 300 times its latest annual revenue. It’s a crude comparison, but it still shows how much future commercialization investors have already priced in.

Sarvam unveiled Sarvam 30B and Sarvam 105B at the India AI Impact Summit in February and open-sourced both mixture-of-experts models under the Apache 2.0 license on March 6. As I noted at the time, much of the benchmark evidence behind them came from evaluations Sarvam designed and ran itself.

The commercial traction comes from less visible products. The company disclosed that its conversational platform handles more than 2 million interactions a day. Sarvam Vision, its document model, is digitizing more than 35 million pages of insurance forms and land records. These are the workloads where Indian languages, handwriting, and noisy audio give a local vendor an edge.

At the same time, Sarvam is chasing frontier scale. In July, it added Devendra Singh Chaplot, a Mistral AI founding team member and former xAI pre-training lead, as an advisor and outlined plans for a trillion-parameter model and a San Francisco office.

Sarvam Is Also Becoming An Inference Provider

Sarvam’s developer documentation reveals a second shift. Its API now offers third-party open-weight models in beta , including Zhipu’s GLM-5.3, DeepSeek V4 Flash and Google’s Gemma 4 31B. The company says these models are served on Sarvam infrastructure, and developers reach them with the same credentials they use for Sarvam-105B.

The pricing makes the strategic tension explicit. Sarvam charges ₹29.28 per million input tokens for Sarvam-105B and ₹19.80 per million for DeepSeek V4 Flash , with a similar gap for output tokens. Sarvam is hosting a Chinese model and selling it below its own flagship, which raises an early question for Sashikumar about whether the company is optimizing for the share of its own models or the share of its platform.

The third-party model catalog also puts Sarvam in competition with Indian GPU clouds such as Yotta, E2E Networks and Neysa, and with global inference platforms like Together AI and Fireworks AI.

When an Indian bank runs a Chinese model through Sarvam, sovereignty shifts from model provenance toward deployment sovereignty, meaning where inference occurs, who controls the infrastructure and weights and which jurisdiction governs the service. Sarvam’s public documentation does not specify the exact region in which each beta model runs, and regulated buyers will need to know that. This shift makes Sarvam’s differentiation from domestic GPU clouds depend more on its software, compliance and enterprise services.

How Mistral, DeepSeek And Zhipu Influence Sarvam’s Options

Labs outside the US frontier shape Sarvam’s choices as role models, suppliers and rivals, though none maps neatly onto its mix of model lab, application vendor and inference platform.

DeepSeek built influence through open weights, low prices and research credibility, releasing its R1 reasoning model under the MIT license in January 2025 . Its parent, the hedge fund High-Flyer, funds a research-first posture that Sarvam cannot replicate.

Mistral also ships open weights and sells APIs to developers. It raised €1.7 billion in September 2025 in a round led by ASML, and in January, France’s Ministry of the Armed Forces awarded it a framework agreement under which its models run on French infrastructure. Research firm Sacra estimates Mistral’s annualized revenue at about $400 million.

For Sarvam, the most relevant difference is distribution, because Mistral turned sovereignty into a procurement requirement for European defense agencies and regulated banks. Sashikumar’s background fits that motion well. Before leading Google Cloud India, he was its chief operating officer for Asia Pacific.

Zhipu AI shows both the promise and the cost of selling sovereignty. The Beijing-based lab listed in Hong Kong in January and leads an alliance that helps ASEAN and Belt and Road countries build national models. It reported first-half 2026 revenue of 954 million yuan, about $142 million and five times the prior year, while losing 2.07 billion yuan. Sovereign and government-linked demand can generate fast revenue growth without approaching profitability.

Zhipu also carries a burden that Sarvam does not, since the US Commerce Department added it to the Entity List in January 2025.

The hire also carries an irony I flagged in February, when I noted that India’s sovereign AI ambitions run largely on American cloud. Sarvam has now recruited the executive who sold that cloud to Indian enterprises.

Why Better Selling Alone Will Not Fix Sarvam

Sashikumar inherits three problems that a stronger sales organization alone will not solve. Sarvam lacks independent evidence that its models match GPT, Gemini and Claude on Indian-language tasks, and a bank CIO will not sign a multi-year contract on national pride.

A trillion-parameter training program, a US expansion, coding and cybersecurity models and a third-party inference business also compete for capital and management attention. HCLTech partners with Microsoft, Google and AWS on AI services for the same clients it could bring to Sarvam, which creates a credible risk of channel conflict.

Sashikumar’s First Three Priorities

The first priority is to decide what Sarvam sells. The company now presents itself as a model builder, an application vendor and an inference platform, and pricing DeepSeek below Sarvam-105B sends buyers a mixed signal. Sashikumar should lead with the voice and document products, where Sarvam’s traction is strongest and where state governments, banks and insurers already have budgets, and position the third-party catalog as an on-ramp that routes Indian-language tasks to Sarvam’s own models.

The second priority is to replace self-reported benchmarks with customer proof. Sarvam should commission third-party evaluations of regulated workloads, such as claims processing in Indian languages, and publish error rates and cost per transaction. Two or three named reference customers in banking or government would do more for its credibility with CIOs than another leaderboard result.

The third priority is to turn HCLTech’s $150 million stake into a working sales channel. That means clear rules on when HCLTech leads with Sarvam and when it leads with a hyperscaler, along with a disclosed hosting region for every model so that regulated clients can clear data-residency reviews. Building that kind of partner motion is the work Sashikumar did at Google Cloud India.

With Sashikumar, Sarvam is signaling that enterprise sales and distribution matter at least as much as another benchmark win. Its inference business complicates any Mistral comparison, since Sarvam now sells its own models alongside Chinese and American-designed alternatives, including DeepSeek at a lower token price than Sarvam-105B. Next year will reveal whether Sarvam is building India’s champion foundation model or its sovereign AI distribution layer. If customers continue choosing cheaper third-party models on the platform, distribution will become the bigger business.