Peak XV Forecasts Low-Cost AI Boom as India Targets 100,000 Active GPUs

Domestic tech firms will win India's artificial intelligence market by undercutting Western hyper-scale developers on cost and training targeted models, Peak XV Managing Director Rajan Anandan confirmed.

The compute deficit is evaporating.

India closed 2025 with fewer than 10,000 active graphics processing units across its private and academic sectors—a footprint smaller than that of an individual elite university in the United States. That severe shortage choked local research, prompting domestic venture capital to bypass local startups in favor of hardware hubs across North America, South Korea, and Taiwan.

Yet compute capacity turned sharply over the last nine months. Enterprise GPU allocations expanded nearly ten-fold nationwide.

"Over the next month or two, India will cross 100,000 active GPUs," Anandan stated. "Over the next several years, India will have another 10x growth on that."

Smaller Models Over Trillion-Parameter Ambitions

While United States tech giants burn billions attempting to train 15-to-20-trillion parameter systems for artificial general intelligence, Indian developers are building compact, domain-specific networks centered at 100 billion parameters.

They are doing it at fraction-of-a-cent operational costs.

Affordability remains non-negotiable.

The domestic strategy reflects real-world enterprise adoption: end-users require functional, cheap task completion rather than massive generalist foundation models.

Deep Tech Parallels and Hardware Spin-Outs

Anandan argued that India's deep tech trajectory mirrors past inflection points in aerospace and semiconductor engineering.

In 2015, India housed exactly one private space technology startup. Today, the country lists over 400 space tech ventures, standing as the third nation globally where a private enterprise put a commercial rocket into orbit. Semiconductor ventures followed an identical track, rising from a handful of firms six years ago to more than 100 specialized chip design startups today.

And corporate talent is defecting.

Roughly 20 percent of the world's semiconductor design engineers reside in India, long staffed across multinational subsidiaries of Nvidia, Qualcomm, Intel, and AMD. Senior engineers from those multinational rosters are now resigning to launch domestic design shops.

Backing Eternal and Dismissing Temple Pushback

Turning to consumer hardware, Anandan cited Eternal—the parent entity of Blinkit and India's most valuable venture-backed startup—as proof of domestic operational grit. Eternal currently logs 30 million monthly transacting users across quick commerce, a business model that collapsed in several Western markets.

Anandan also pushed back against medical professionals criticizing Temple, the wearable wellness device developed by Eternal founder Deepinder Goyal.

Physicians had raised questions over data efficacy and alleged the wearable was hyped to inflate startup valuations. Anandan rejected that characterization outright, clarifying that Temple is neither registered nor sold as clinical equipment.

"Nobody actually said it was a medical device," Anandan said, noting the product is restricted to a closed beta testing group of a few hundred users. "It is not a medical device but a wearable that gives you a set of health metrics, and it's up to you to decide what you want to do with it."

Market Context: India's pivot toward lean, 100-billion parameter systems directly mirrors its payments playbook. By letting foreign firms spend hundreds of billions on theoretical AGI training while domestic developers capture enterprise software workflows through low-cost deployment, Indian tech seeks to commercialize artificial intelligence at mass consumer scale.

With hardware deployments scheduled to hit the six-figure GPU threshold over the coming weeks, domestic venture investors are pivoting toward engineering teams converting multi-tenant chip access into production-ready commercial tools.