Mint Explainer | Anthropic IPO proves profitable gen-AI is hard. India on the same boat too?

Summary

Most generative AI companies incur incredibly high operating costs for servers in data centres, while adoption of their products remains either heavily subsidised or in its early stages.

On 26 October, Anthropic PBC, one of the world's largest generative artificial intelligence (AI) companies, said in its initial public offering (IPO) filing that it might not have a clear timeline for turning profitable. Despite reporting its first operating profit in the June quarter, it posted a net loss of $80.8 billion in the first half of 2026 on revenue of $16.3 billion.

Mint examines why turning the technology into a profitable business remains a key challenge.

What explains Anthropic's dire profitability warning?

On page 63 of its S-1 regulatory filing in the US for a public listing, Anthropic said that "if we fail to effectively manage the growth of our business and operations, we may be unable to maintain effective internal controls, reporting systems, and compliance processes, may face increased regulatory scrutiny, and may suffer harm to our brand and our ability to attract and retain customers".

"As a result, our business, financial condition, and results of operations could be adversely affected. We have a history of net losses and may not generate sufficient revenue to achieve or sustain profitability in the future."

This highlights all the issues around modern-day generative AI, despite the likes of OpenAI and Anthropic setting all-time records of valuation growth. Most of these companies incur incredibly high operating costs in terms of servers in data centres, while adoption of their products remains either heavily subsidized or in early stages. Additionally, Anthropic acknowledged that AI is a fast-moving field, and its foundational products can be disrupted in a very short time, further hurting its financials.

At the heart of it, however, lies the fact that the bulk of Anthropic's costs come not just from building AI algorithms, but from keeping its Claude chatbot and underlying technologies running. At the consumer end, its usage remains subsidized. Companies, meanwhile, have been very cautious about adopting AI due to its risks. All of this combined has led to the grave warning on Anthropic's books right now.

Are other AI entities also equally at risk of losses?

Yes, and that is what makes the modern-day AI industry particularly difficult for investors to bet on. The risks highlighted by Anthropic will also apply, in varying degrees, to OpenAI. Google's Gemini and Microsoft's Copilot face some of the same challenges, but they benefit from being part of established ecosystems: Google's Android and Microsoft's Office suite, both of which have been successfully monetized. However, the broader challenges Anthropic has highlighted will apply to all companies developing foundational AI models.

Which are Anthropic's equivalents in India?

The closest Indian parallels to Anthropic and OpenAI are a handful of government-funded startups -- Soket AI, BharatGen, a non-profit entity incubated at IIT Bombay, and others -- that are still in their early stages. Only Sarvam AI, which has reached a valuation of $1 billion or more, has launched a consumer-facing product. None has unveiled enterprise offerings yet.

Two other entities worth noting are publicly listed Fractal Analytics and Sharad Sanghi-backed Neysa AI, which operate in AI infrastructure and enterprise AI, respectively. However, neither is a pure-play generative AI model builder.

Do Sarvam and BharatGen face the same risks as the US giants?

In India, the challenges of monetizing AI are slightly different. India has not spent the billions that Anthropic and OpenAI have spent to build large language models. Instead, India's AI innovations involve building smaller, more targeted AI models that cater to specific enterprise use cases.

In February, Sarvam co-founder Vivek Raghavan told Mint that this approach was deliberate: running apps such as Claude and ChatGPT is expensive, while purpose-built, topic-specific apps are easier to monetize by selling them directly to businesses.

The challenge, though, is to build enterprise trust, especially at a time when companies are already cautious about deploying generative AI at work.

How does this approach help financially?

The scales are at completely different levels. According to Anthropic's S-1 filing, the company spent $8.06 billion through 2025, of which $7.3 billion was for computing infrastructure. Sarvam, based on its FY25 financials, incurred total expenditure of $13.8 million, just 0.17% of Anthropic's expenses. Sarvam, however, has raised about $387 million in funding to date, about 0.2% of Anthropic's $140 billion.

Such limited funds, however, have prompted India to develop domain-specific AI use cases, often built on pre-built foundational models and trained on smaller amounts of data, requiring fewer funds.

Such an approach is also squarely targeted at specific fields, such as healthcare and education, which most AI in the country focus on. The risk with this, though, is that enterprises tend to trust larger companies with stronger compliance and support teams, and the idea that a sovereign AI product is 'better' than global ones remains unproven.