Headline & intro
India’s newest AI chat app is not just another interface to talk to a large language model. With Indus, Sarvam is quietly testing whether India can own a meaningful slice of the AI stack instead of renting it from Silicon Valley. That should matter far beyond Bengaluru.
If India succeeds in building competitive, local-language models and distributing them at massive scale, the balance of power in AI could tilt away from the current US–China duopoly. For Europe, which is also wrestling with questions of digital sovereignty and AI regulation, Indus is an early glimpse of what a “third pole” in AI might look like — and of the trade-offs that come with going local.
The news in brief
According to reporting by TechCrunch, Indian startup Sarvam has released Indus, a new AI chat app for web, iOS and Android, currently in beta and apparently limited to users in India. Indus serves as the primary interface for Sarvam 105B, the company’s newly announced 105‑billion‑parameter large language model, unveiled alongside a 30B model at the India AI Impact Summit in New Delhi earlier in the week.
Indus supports text and voice input and can reply in both text and audio. Users can sign in with a phone number or via Google or Apple accounts. The app has several constraints at launch: users cannot selectively delete chat history (only by deleting their account), there is no option to disable a reasoning mode that can slow responses, and access may be throttled due to limited compute capacity, with a waitlist for some users.
Founded in 2023, Sarvam has raised about $41 million from investors including Lightspeed, Peak XV and Khosla Ventures to build models tailored to Indian languages and use cases, TechCrunch reports. The company has also announced enterprise initiatives, hardware ambitions and partnerships with HMD (for Nokia feature phones) and Bosch (for automotive AI applications.
Why this matters
Indus is interesting not because the world needs yet another chat interface, but because it signals three strategic bets.
First, India wants its own foundation models. Sarvam 105B is explicitly positioned as an India‑centric LLM: tuned for local languages, local contexts and local price sensitivities. That’s a direct challenge to the assumption that OpenAI, Anthropic and Google will simply absorb India as a gigantic downstream market. If India can run high‑quality models on its own infrastructure, it gains leverage over pricing, data flows and policy.
Second, this is AI for the “next billion” in practice, not in pitch decks. The partnership with HMD to bring AI to Nokia feature phones is easy to overlook but strategically huge. India still has hundreds of millions of users on basic devices and patchy connectivity. If Sarvam can deliver meaningful AI assistance over low‑end hardware, it will expand the addressable market for generative AI far beyond premium smartphones and desktops — something Western players have only talked about.
Third, it accelerates competition on local-language performance. OpenAI and Anthropic are strong in English and major world languages, but coverage for India’s linguistic diversity has been uneven. A model built from day one to handle multiple Indian languages can quickly become the default for vernacular use, even if it lags the global leaders slightly on frontier capabilities.
The losers, at least in the short term, are the global platforms that have treated India primarily as a distribution market, not a source of competing core technology. Indus won’t dent ChatGPT usage tomorrow — OpenAI and Anthropic both report strong adoption in India — but it changes the negotiating posture: Indian regulators and enterprises now have a plausible domestic alternative to point to.
The bigger picture
Indus slots into a broader trend: AI is fragmenting along national and regional lines.
On one side, US firms like OpenAI, Anthropic, Google and Meta are racing to build ever‑larger general‑purpose models and distribute them via global platforms and APIs. On the other, major economies are asking whether relying entirely on foreign, mostly US‑controlled models is acceptable for critical infrastructure, culture and security.
India is moving from asking that question to building answers. Sarvam joins a growing set of domestic model builders, often backed by deep‑pocketed investors and tech veterans, all arguing that India’s scale and linguistic complexity justify a local stack.
We’ve seen earlier versions of this playbook. China erected strong regulatory and commercial barriers and incubated its own AI giants. Europe pushed hard on regulation — GDPR, now the EU AI Act — but has struggled to match that with equally strong homegrown AI platforms. India appears to be attempting a hybrid: relatively open to foreign players, but with explicit political and commercial support for domestic models.
This also reflects a subtle shift in AI competition. The frontier race — who has the biggest, smartest model — still matters, but it’s increasingly complemented by a fit‑to‑context race: who best understands and serves the nuances of local languages, legal systems and cultural expectations.
Indus is an early test of whether that fit‑to‑context strategy can carve out defensible market share before global players close the gap on local language support, and before open‑source models become “good enough” everywhere.
The European angle
For European readers, Indus may feel distant — it’s India‑only for now — but the underlying dynamics look familiar.
The EU talks constantly about digital sovereignty, yet most of the foundation models Europeans use today are American. Europe is trying to rebalance this with the EU AI Act, which will impose obligations on providers of general‑purpose AI models, especially those deemed to pose systemic risks.
If Sarvam eventually targets European customers, it will have to navigate this regulatory maze: transparency around training data, evaluations, copyright, risk management and downstream usage. That’s non‑trivial for a young startup, especially one originally optimised for Indian data and norms.
At the same time, Europe has its own small‑language challenge. From Slovene to Welsh, many European languages suffer from the same data scarcity and tooling gaps that plague India’s regional languages. A company whose core competence is building models for low‑resource, morphologically complex languages may actually be a useful partner — or competitor — for European AI builders wrestling with similar problems.
There’s also a geopolitical undercurrent. European policymakers quietly worry about an over‑reliance on US (and, to a lesser degree, Chinese) AI infrastructure. An emerging Indian AI ecosystem offers a potential third supplier base that is politically less sensitive, at least on paper. Whether that translates into real procurement diversification will depend on performance, price and compliance.
For European enterprises, the message is simple: the future AI stack will likely include models from multiple regions, each optimised for different contexts. Ignoring developments like Indus because they launch “far away” is a mistake.
Looking ahead
The immediate questions around Indus are tactical.
Can Sarvam scale compute capacity fast enough to remove waitlists and latency issues without blowing up its cost structure? Limited capacity at launch is understandable, but Indian users are notoriously unforgiving of unreliable apps, and alternatives are one tap away.
Can the company harden its product around privacy and control? The current inability to delete individual chat histories, and the lack of granular settings (for example, to toggle computationally heavy reasoning), are acceptable in a beta, but they won’t fly with enterprises — and they’d be a non‑starter under EU‑style data protection norms.
More strategically, Sarvam will have to answer three bigger questions over the next 12–24 months:
- Distribution: Will Indus remain a standalone app, or become an engine embedded into telco bundles, government services, OEM deals and super‑apps? In India, distribution partnerships often matter more than brand.
- Positioning: Does Sarvam want to be “India’s ChatGPT” for consumers, or a B2B infrastructure provider selling APIs and on‑prem deployment to banks, IT outsourcers and governments? Doing both well is hard.
- Governance: How will Sarvam handle safety, misinformation and political content in an election‑intense, multilingual democracy? That’s not just a technical challenge but a policy and reputational minefield.
For global players — including European startups — the opportunity is to learn. Indus will be a live case study in whether aggressively local, language‑first model strategies can generate defensible value against the gravitational pull of US hyperscalers.
The bottom line
Indus is not yet a threat to ChatGPT or Claude, but it is a clear signal that India does not intend to be just a consumer market for other people’s models. If Sarvam can turn linguistic and cultural proximity into real product advantage at scale, it will strengthen the argument for regional AI stacks from Europe to Latin America. The open question for European readers is whether you want to rely on US models, build your own — or one day run an Indian engine under the hood.
