According to The Register, India is considering an AI architecture in which the model travels to the data rather than the reverse. Reported on 2 November 2026, the idea remains a proposal: execution terms, access conditions and a timeline have not been specified.
🔑 Key takeaways
- India is studying an architecture in which AI models run close to the data, without centralised transfer to a single server.
- The IT ministry MeitY has circulated a privacy-focused framework designed to ease access to data used for AI training.
- The status remains that of a proposal: no timeline, budget or control mechanism has been published.
- Consumer use of AI in travel is growing fast, but errors on prices and visas remain frequent.
A architecture principle still lacking detail
The Register summarises the principle in its headline: “The model travels to the data, rather than the data travelling to the model.” The approach inverts the conventional logic of centralising datasets to train a model. In principle, it reduces the risk of large-scale data transfers, but it shifts the security question to the environments where the model actually runs.
The available extracts remain sparse. They do not say whether the model would be deployed on site, at data holders’ premises, in public infrastructure or in secure environments. The data categories concerned, priority sectors, participant obligations, control mechanisms and the legal status of the scheme are not described. No specific technical mechanism is mentioned, whether federated learning (training distributed across sites without centralising data), confidential computing or synthetic data generation.
The sources do not document the safeguards either. No withdrawal rights, sanctions or appeal mechanisms are described. Until these elements are published, the analysis must stay cautious: the general direction is clear, but the real scope of the scheme is not.
MeitY’s framework and access to training data
Inc42 reports that MeitY, the Ministry of Electronics and Information Technology, has circulated a new privacy-focused framework intended to ease access to data used to train AI models. The two sources converge on two points: more direct access to training data and privacy protection built into the scheme. However, the material does not confirm whether The Register and Inc42 describe exactly the same document, or whether the “plan” and the “framework” are two parts of one proposal.
“The model travels to the data, rather than the data travelling to the model.”
The Register, article of 2 November 2026
| Element | Reported status | Detail |
|---|---|---|
| Nature of the scheme | Proposal | No plan adopted or implemented |
| Lead actor | MeitY | Privacy-focused framework |
| Data access | Eased | Access conditions not specified |
| Execution location | Undecided | On site, public or secure environment |
| Timeline and budget | Not published | No data available |
A market and skills base in expansion
The proposal sits within a broader strategy. India is presented as one of the pioneers of natural-language AI from the early 2010s, with chatbots such as Haptik, Corover.ai and Niki.ai. Its visibility in generative AI grew in the early 2020s, notably through work by Krutrim, Sarvam, Bintex AI and CoRover. The sector’s development is attributed to a combination of startups, public programmes such as Digital India and public digital infrastructure.
The country also has a record of public computing capacity. The Center for Development of Advanced Computing was created in March 1988, initially focused on supercomputing and computing for Indian languages. NITI Aayog’s National Strategy for AI, launched in 2018, aimed to build applied research projects, tools, frameworks and organised datasets, notably for smart mobility, health and agriculture. On 7 September 2018, NITI Aayog, Intel and TIFRAC announced plans for an International Center for Transformative Artificial Intelligence in Bengaluru, a direct precedent for a policy linking applications, data and public institutions.
The economic scale is equally significant. India’s AI market was projected to reach $8 billion by 2025, with an average annual growth rate of 40% between 2020 and 2025. NASSCOM and Boston Consulting Group estimated that AI services could be valued at $17 billion by 2027. According to UNCTAD’s 2025 report, India ranked tenth globally for private-sector AI investment. Mary Meeker said India accounted for the largest share of ChatGPT mobile users and the third-largest pool of DeepSeek users in 2025.
Obstacles identified
The obstacles cited for responsible deployment are data protection, a shortage of skills and ethical considerations. The same source flags a rise in cyberattacks using AI to target organisations. These elements give a factual backdrop to the privacy orientation, but no figures are provided on data breaches, the skills gap or the number of attacks.
Consumer use: a concrete dependence on data
Indian travellers’ use of AI illustrates the dependence on data. According to a 2025 report from major hotel booking platforms, 47% of Indian travellers had used at least one AI tool to plan a trip in 2025, the highest adoption rate in Asia-Pacific. Skyscanner data indicates that one flight search in three in India now starts with an AI-generated suggestion. HappyFares notes that AI is mainly used for itineraries, packing lists, a first visa search, flight strategies and cultural context. These uses potentially involve passport, payment, booking and location data.
Reported performance remains uneven. The table below summarises the indicators available in the sources.
| Study | Indicator | Result |
|---|---|---|
| OpenAI, 2025 travel benchmark | Itinerary usefulness | 4.3 / 5 |
| OpenAI, 2025 travel benchmark | Price accuracy | 2.8 / 5 |
| Henley Visa Index, 2025 | Visa questions answered correctly | 80% to 85% |
| Phocuswright, 2024 | Recommended hotels closed or rebranded | 22% |
| Phocuswright, 2024 | Gap between training data and reality | 6 to 18 months |
Errors often concern fees and permitted length of stay, two factors that weigh directly on traveller decisions. The time lag explains much of the gap: a model trained on older data cannot see recent changes in pricing or operations.
General assistants versus specialised planners
The sources distinguish general-purpose assistants from specialised planners. HappyFares says a general chatbot cannot see live fares across all aggregators and may rely on price ranges that are six to twelve months old. Bridge, by contrast, describes planners that combine language models, travel application programming interfaces, price histories and behavioural signals. According to Bridge, such a tool can produce a detailed itinerary, cost estimates and logistical suggestions in under 60 seconds, compared with eight to twelve hours for manual planning. These figures come from a sector player and have not been independently verified.
Google and AI-assisted search
The private sector is already adapting. Google launched AI Mode as a Search Labs experiment in March 2025, then extended it to US users in May 2025. According to Google, an average AI Mode query is three times longer than a traditional search, and planning-related queries grew 80% faster than other AI Mode queries over the preceding six months.
For hotels, airlines, online travel agencies and tourism boards, this shift may reduce the visibility gained through keywords, advertising and click tracking, since options can be proposed before a user visits a booking site. On 20 May 2026, Google announced new advertising tools for the AI era, including experiences built with Gemini and new ways to appear in complex search journeys.
Outlook and scenarios
The sources report no market reaction to the Indian proposal: no investor response, no corporate position and no regulator comment. No forecast of funding, data volumes or effects on AI service pricing is made. Three trajectories remain plausible, none confirmed: rapid adoption as an operational framework, a prolonged proposal stage while safeguards are defined, or a delay driven by data protection and skills constraints.
The key item to watch is the publication of operating terms: where models run, which data categories are involved, access conditions and control mechanisms. Until these details appear, the scheme remains a documented political intention rather than an operating infrastructure.
Sources
- The Register, 2 November 2026
- Inc42, MeitY framework
- AltexSoft, Google AI Mode and trip planning
- HappyFares, AI trip planning in India
- Wikipedia, Artificial intelligence in India
- Bridge, how AI travel planning works in India
Cet article est publié à titre informatif et éducatif. Il ne constitue en aucun cas un conseil en investissement. Faites vos propres recherches (DYOR) avant toute décision.
