International Embarrassment

Whenever the new world of AI comes up for discussion, the ones leading the conversation are nations that have made it a mission to advance their domestic capabilities in this field, namely the USA and China. The AI Impact Summit 2026 was India’s attempt to enter the conversation. By inviting more than 20 heads of state and thousands of participants from across the globe — some of whom are prominent figures like Bill Gates and Sam Altman — India wanted to showcase its indigenous innovation and send a signal to the world that it has arrived on the scene. (Bill Gates withdrew from the summit hours before his scheduled keynote on February 19, citing the renewed Epstein files controversy. He arrived in India but did not deliver his address.) However, what was supposed to be India’s debut entry into the world of AI elites soon turned into a global spectacle as a robot dog, claimed to have been developed within the confines of a private university campus, grabbed more attention than it wanted to chew on. As soon as the interview with the professor representing the university went viral, fact-checks across the internet revealed the truth: the dog was actually Chinese. Unitree’s Go2 robodog was claimed to have been “developed” by the Centre of Excellence at Galgotia University, and a day later, the university apologised for the confusion, stating that their representative had been “ill-informed” and was “not authorised to speak to the press.”

The news is embarrassing, but it is also quite revealing. One has to wonder how the university was able to get an opportunity to participate in the summit and who from the organisers’ side oversaw their pitch, display, and even their claims. Galgotia University has not yet established a credible identity for itself, specifically when it comes to research and development. For instance, social media scrutiny after the controversy revealed that though the university filed a record 1,099 patents in the year 2022–23 alone — more than India’s top IITs — of those 1,000-plus patents, only 57 have been granted. A grant rate of roughly 5% makes you question whether filing patents is primarily being pursued to improve ranking metrics and access government incentive schemes, or whether it is a byproduct of genuine, commercially viable innovation. India boasts many bastions when it comes to technology education; it sends millions of engineers around the world who have built the very systems the new world runs on. Yet, when we talk about this summit in the future, Galgotia’s debacle will be the one grabbing all the attention.

In some ways, the Galgotia incident acts as a metaphor for our lived experience. India has become more of a consumer of the latest AI technology shaping the world rather than a leader in innovation. Innovation, after all, is when you bring forth something truly novel; it cannot be outsourced and surely cannot be bought. And why just point fingers at the university when private companies are no less responsible? According to reports, Wipro also showcased the same Unitree Go2 robot dog — branded as “TJ” — at the summit, with the Unitree markings removed from the device. Unlike Galgotia, Wipro’s representative stopped short of explicitly claiming the hardware as their own invention, with company sources clarifying that they are a “software company” and “have never claimed to be a hardware company.” Still, it begs the question of just how many companies have put their name on this robodog — a question that the organisers seemed to have forgotten to ask.

India in the past decade or so has become more focused on educating its youth to become well-trained tech executives and, somewhere along the way, fell off the bandwagon when it came to encouraging its people to innovate and compete. India’s R&D spending hovered at just 0.64% of GDP (approximately $75.7 billion) in 2024, far behind China’s 2.68% (approximately $785.9 billion) and the US’s approximately 3.5% (approximately $784 billion). China surpassed the US to become the world’s largest R&D spender for the first time in 2024. India has tech giants like Infosys and TCS, which thrive on providing tech talent that can build sophisticated software, applications, and systems at an affordable price, but by the very nature of their business model, these companies are not incentivised to innovate solutions that can reduce the need for tech manpower. Infosys’s core business model has historically been labour arbitrage — offering cheaper software development and IT services by leveraging India’s large pool of skilled engineers at lower wages than Western markets. But as AI coding tools mature, this wage gap advantage narrows because a smaller team with AI tools can now do what a large team once did. If competitors adopt AI and Infosys does not, it risks being undercut on both price and speed. Infosys’s revenue depends on headcount-linked contracts. Though Infosys is already investing in AI — it has platforms like Topaz and has been integrating generative AI into its service offerings — while the world was moving away from tech hiring and toward AI, Infosys was still defending its moat of providing tech talent for cheap. Aggressively automating now could cannibalise its own billing model before a new model is ready to replace it. Infosys, like other tech giants that stand alongside it, must learn to float on the changing tides before the waves crash, and they must learn it soon.

DeepSeek’s arrival in January 2025 jolted India into action. MeitY moved quickly, putting out calls for homegrown foundation models and opening up bids for GPU access to private players. Among those selected, Sarvam AI emerged as the most significant bet, tasked with building a large language model tailored to India’s languages and context. The company was no stranger to this space; it had earlier built OpenHathi-Hi-v0.1, a Hindi language model constructed on Meta’s Llama 2 architecture and trained on 40 billion tokens of Hindi and related Indian-language content — at the time, one of the most substantial open-source efforts in Hindi. It had also previously released Sarvam-1, a 2-billion-parameter model spanning 10 Indian languages.

For the government’s larger ambition, Sarvam was granted access to 4,096 Nvidia H100 GPUs. While the original brief envisioned a 70-billion-parameter model, Sarvam arrived at the AI Impact Summit with something more ambitious: two new models — one at 30 billion parameters and another at 105 billion — both built on a mixture-of-experts architecture and trained from scratch on 16 trillion tokens. This sits within the broader IndiaAI Mission, a roughly $1.23 billion national programme launched in March 2024 aimed at strengthening India’s AI infrastructure from the ground up.

But the initiative has not been without friction. A recurring point of criticism was that Sarvam’s government-funded model was being developed as a closed system, raising uncomfortable questions about who ultimately benefits when public money underwrites private technology. At the summit, Sarvam moved to put some of that discomfort to rest by announcing that it would open-source both the 30B and 105B models — though whether that fully resolves the tension between public investment and private ownership remains an open question.

The summit, which was expected to run for five days, has been extended by one more day. Besides the controversy related to the robodog, the summit has also been hit by claims of mismanagement and disorganisation — including reports of a startup founder’s products being stolen from his booth during a security lockdown, UPI failures, overcrowding, and exhibitors being locked out of their own stalls. Owing to an overwhelming turnout, the government added an additional day to the programme. But while we may gloss over the people who attended the summit, its displays, and the number of tickets bought, there is a difference between being adept at AI and holding a conference at an exhibition centre. One involves execution and planning; the other involves buying and selling admission tickets.