State of AI in Indian Higher Education — 2026

Annual Research Report · Volume 1 · 2026

An annual stocktake of how artificial intelligence is being taught, staffed, and resourced inside India's universities and colleges. Written for the Vice Chancellor, Pro-VC, AICTE official, faculty lead, or AI hiring manager who needs a defensible base of facts before making a decision. Free to read, free to cite, open for correction.

Version 1.0 · Published 14 May 2026 · CC BY 4.0

This report is the first in a planned series. It examines the institutional landscape — universities, governance, curriculum, faculty, infrastructure, partnerships, and outcomes — not the AI industry broadly. For a complementary view of what AI graduates can actually do against a published rubric, see our parallel publication, the AI Skills Index, India 2026.

On this page

• Executive summary
• 1. The institutional landscape
• 2. NEP 2020 + AI: implementation status across India
• 3. What universities are currently teaching
• 4. The faculty supply problem
• 5. Infrastructure
• 6. Industry-university partnerships
• 7. Hiring outcomes
• 8. The student experience
• 9. Comparison with global benchmarks
• 10. What's working
• 11. What's not working
• 12. What changes by 2030
• 13. What we'd like to measure next year
• 14. Method and limitations
• 15. Sources and citations

Executive summary

India is two stories at once. In one story, the country is a global AI superpower-in-waiting: third in the Stanford 2025 Global AI Vibrancy Index ¹, second worldwide for the number of AI researchers and inventors counted outside China ¹, home to the most expensive engineering cohort the world has ever produced. In the other story, more than half of professor-level positions at the country's apex public institutions are vacant ², roughly 30–40% of faculty seats across the sector at large are empty ³, and the India Graduate Skill Index finds only 42.6% of graduates broadly employable ⁴. Both stories are true. This report is about how they coexist inside the universities themselves.

What follows is what we found after surveying the regulator-published curriculum frameworks, the AISHE 2021-22 dataset (the latest publicly released), the NIRF 2025 cycle, the Stanford AI Index 2025, NASSCOM and Deloitte talent reports from 2024–25, the IndiaAI Mission disclosures, and a non-systematic but extensive scan of public university curricula, press releases, and faculty pages. The 10–12 findings most useful to a busy decision-maker:

1. India has 1,168 universities and 45,473 colleges enrolling roughly 43 million students ⁵. About 41% of these universities are private (473 unaided, plus 10 private deemed-aided, plus 124 deemed). The IIT-IIM apex serves less than 0.5% of the cohort. The story of AI in Indian higher education is overwhelmingly a story about what the other 99.5% of campuses are doing.

2. B.Tech enrolment is at an eight-year high — 12.53 lakh students secured seats in 2024–25, a 67% rise since 2017-18 — and Computer Science alone accounts for 3,90,245 of those seats ⁶. AICTE approved 15.98 lakh seats for 2025–26, a 7% expansion year-on-year ⁶. The growth is concentrated in CS, AI/ML, and Data Science specializations. Mechanical, civil, and electrical disciplines are shrinking.

3. More than 920 colleges now offer a dedicated B.Tech CSE (AI & ML) specialization ⁷. AICTE has declared 2025 the "Year of Artificial Intelligence" and is positioning AI as a curriculum priority across 14,000+ approved colleges and 40 million students ⁸. The labels have proliferated faster than the underlying capability has.

4. The faculty supply problem is the binding constraint, not curriculum or compute. A Parliamentary Standing Committee report in March 2025 found 56.18% of professor-level positions vacant across IITs, IIMs, NITs, IISERs, and central universities ². Sector-wide, the student-faculty ratio sits at about 27:1 against the UGC benchmark of 20:1 and a global-top-10 norm of ~10:1 ⁹. For frontier AI specifically — researchers who have actually shipped models — the absolute number of PhD-grade Indian faculty is in the low thousands, and is concentrated at fewer than 30 institutions.

5. Compute is no longer the bottleneck the sector once thought it was. The IndiaAI Mission has onboarded over 38,000 GPUs at a subsidised rate of ₹65 per hour and made them available to academia, MSMEs, startups, and registered research bodies ^{10 11}. The 2025–26 Union Budget added a ₹500 crore Centre of Excellence for AI in Education ¹². The gating constraints have moved from hardware to the humans who can use it.

6. Industry partnerships exist in quantity; quality is uneven. IBM has stood up a National AI Lab at AICTE headquarters ¹³; Microsoft, Google, and IBM all run campus-level skilling programs with NASSCOM FutureSkills Prime ¹⁴; Google Cloud announced a national pilot with Chaudhary Charan Singh University, Meerut to demonstrate an "AI-first" university ¹⁵. A meaningful fraction of partnerships, however, are MoU-stage. Distinguishing a working partnership from a press partnership requires reading past the photographs.

7. Hiring outcomes are bifurcated and getting more so. Freshers with verifiable AI portfolios are getting ₹8–12 LPA at product companies and startups, while the IT-services bulk-hire market sits in the ₹5.8–8 LPA band ¹⁶. TCS, Infosys, and Wipro are still recruiting in tens of thousands ¹⁷, but the role mix is shifting from generalist engineer to AI-adjacent specialist faster than most campuses' curricula are tracking.

8. NEP 2020 is being implemented in form, less so in substance. Over 105 universities have moved to a four-year undergraduate programme; the Academic Bank of Credits and National Credit Framework are operational ^{18 19}. The harder NEP commitments — multidisciplinary teaching, faculty mobility across departments, genuine credit transfer, and the integration of AI as a cross-cutting capability rather than a specialisation silo — remain works in progress.

9. The R1-equivalent AI research base is narrow. Outside the IITs, IISc, the IIITs (Hyderabad, Delhi, Bangalore), a handful of NITs, and three to five private universities with serious investment (BITS Pilani, Ashoka, Shiv Nadar, IIIT-D, and a small cohort behind them), India's published AI research output is thin. Six Indian institutions appear in the QS top 100 for Computer Science in 2026, up from two the prior year ²⁰ — real progress, but the gap to the US/UK top tier remains large.

10. The on-campus AI Studio model is emergent but unproven at scale. A growing number of private universities are commissioning dedicated AI labs, often with corporate partners (IBM, NVIDIA, Microsoft). What is missing is independent evidence that lab construction translates into capability-level student outcomes. This is the single largest measurement gap in the sector and one we intend to address in Volume 2.

11. The international comparison is sobering on faculty, encouraging on talent volume. China's Tsinghua launched a College of AI in April 2024 led by a Turing Award winner ²¹; US R1 universities continue to publish a disproportionate share of frontier AI research. India's strength is the size of its talent pool, the velocity of its CS enrolment growth, and the early public-compute commitment. The weakness is depth — PhD-grade faculty per capita, peer-reviewed output per institution, and the thin distribution of frontier-research capability outside ~30 campuses.

12. The most consequential decision is not curriculum, it is staffing. Every other lever — compute, partnerships, NEP credit structure, capstones, infrastructure spend — converts into student outcomes only at the rate that competent, current, hands-on AI faculty can supervise the work. The institutions that will look most different in 2030 are not the ones with the newest AI labs. They are the ones that have hired, retained, and concentrated AI faculty above a critical mass.

A note on this report's limits before you read further. We have not run primary surveys of universities. The granular per-institution numbers reported are drawn from public sources — university websites, AISHE, NIRF, government press releases, and reputable secondary reporting. Where a number is conjectural we say so, and where it is unsourced we mark [Source needed]. This is Volume 1; we expect to publish corrections and to extend the dataset materially in Volume 2.

1. The institutional landscape

Any honest conversation about AI in Indian higher education must start with how unusually fragmented and stratified the system is. AI policy aimed at the IIT system reaches a different cohort, with different constraints, than AI policy aimed at the state-affiliated colleges that educate most undergraduates.

1.1 The numbers

Per the latest All India Survey on Higher Education (AISHE 2021-22), the most recently published full report ⁵:

Within the 1,168 universities:

• 685 are government-managed (240 central, 445 state)
• 473 are private (unaided)
• 10 are private deemed (aided)
• The balance includes deemed-to-be-universities (public and private) and institutes of national importance

341 universities have been established since 2014-15, a near-30% expansion of the university-level system in seven years ⁵. The same period saw college numbers rise from 38,498 to 45,473, an increase of 6,975 ⁵.

What this means for AI: The institutional system has been growing faster than the country can produce the senior faculty needed to staff it. Every new university opens with the same problem: where does the AI department come from?

1.2 The tier classification

There is no official tier system. The shorthand that follows is the one most commonly used in industry, policy, and admissions discussions, and the one we use throughout this report:

• Apex (Tier 0) — the IITs (23), IIMs (20), IISc, IISERs, NITs (31), IIITs (25), and AIIMS. Combined undergraduate intake is on the order of 65,000–75,000 seats per year. Highly selective. Faculty quality varies but is the country's strongest base.
• Tier 1 private — BITS Pilani, the Manipal Academy of Higher Education, VIT, SRM, Thapar, the four IIITs that are not centrally funded but operate at similar quality. Then a smaller liberal-arts cohort with strong AI investment: Ashoka, Shiv Nadar, OP Jindal, Krea, Plaksha. Tuition typically ₹4–10 lakh per year.
• Tier 2 private — large enrolment private universities (LPU, Chitkara, Bennett, Amity, Christ, Symbiosis, Chandigarh University, Sharda) and a long tail of state-private universities with engineering capacity. Tuition ₹1.5–5 lakh per year.
• Tier 3 — the unranked private engineering colleges, the long tail of state universities and affiliated colleges, and the polytechnic system. The majority of the country's graduates by volume.

The Stanford AI Index 2025 finds India ranked third globally for AI vibrancy ¹, but that ranking is dominated by output from the apex tier and a handful of Tier-1 private institutions. The remaining ~95% of the student cohort experiences a different country.

1.3 Governance

Three bodies shape AI in higher education at the regulatory layer:

• UGC — University Grants Commission. Governs curriculum frameworks (FYUP, NCrF, ABC), credit transfer, faculty qualifications, and degree recognition. Issues the Curriculum and Credit Framework for Undergraduate Programmes (CCFUP) and equivalent PG framework.
• AICTE — All India Council for Technical Education. Governs technical education programs including B.Tech, M.Tech, and AI-specific specialisations. Releases model curricula, including the AI&ML and AI&DS model curricula referenced widely by private engineering colleges ²². Declared 2025 the Year of Artificial Intelligence ⁸.
• Ministry of Education + MeitY — Sets the National Education Policy (NEP) direction, funds the IndiaAI Mission, and runs the National AI Centre of Excellence in Education programme.

State Higher Education Departments and a number of specialised regulators (NMC for medical, BCI for law, NCTE for teacher education) overlay these. For AI specifically, the operational locus is the AICTE-UGC-MeitY triangle.

1.4 Why the structure matters

A B.Tech CSE student at IIT Delhi takes an AI elective from a department of 60+ research faculty with 1,000+ peer-reviewed publications among them. A B.Tech CSE student at a Tier-3 affiliated college may "study AI" from a faculty member whose own training in machine learning is one online specialisation and a CSE postgraduate degree. Both will leave with a degree that says CSE.

The system, as it is currently regulated, does not distinguish between these two experiences. The hiring market does — increasingly, through portfolio, interview, and verifiable project work rather than by degree. The mismatch between credential and capability is the underlying problem this report tries to characterise.

2. NEP 2020 + AI: implementation status across India

The National Education Policy 2020 is the most consequential higher-education document since the Radhakrishnan Commission of 1948. Its AI-relevant provisions are spread across the policy text and the implementation circulars issued since.

2.1 The AI-relevant NEP clauses

NEP 2020 explicitly names AI as one of the "cutting-edge" areas (alongside 3D machining, big data, genomics, biotechnology, nanotechnology, neuroscience) that should be woven into undergraduate education rather than confined to specialist programs ²³. Three structural commitments matter most:

1. The Four-Year Undergraduate Programme (FYUP) with multiple entry/exit, designed to support breadth + depth and to make multidisciplinary minors viable.
2. The Academic Bank of Credits (ABC) and the National Credit Framework (NCrF), designed to allow credits earned in any approved program — including online and skilling courses — to count toward a degree ¹⁹.
3. Multidisciplinary teaching as a default, including requiring engineering students to take humanities credits and vice versa.

2.2 Where implementation stands (mid-2026)

The UGC's 2022 undergraduate curriculum guidance includes AI, machine learning, deep learning, big data, drones, and 3D machining as recommended cross-cutting topics with health, environment, and sustainability applications ²³. The text of the guidance is strong. The classroom reality varies by institution.

2.3 AICTE's AI-specific implementation

AICTE has done three things that materially affect AI in higher education:

• Released model curricula for B.Tech CSE (AI&ML) and B.Tech CSE (AI&DS) ²² that 921+ private and government engineering colleges now reference ⁷.
• Declared 2025 the Year of Artificial Intelligence, with a stated reach of 14,000+ approved colleges and 40 million students ⁸.
• Established 423 IDEA (Idea Development, Evaluation & Application) Labs ²³ for STEM-based experiential learning, several of which have AI components.

AICTE also incorporated AI components into all IT-related courses by mandate, which is the regulatory grounding for what many private universities now call an "AI minor" or "AI specialisation" inside a CSE degree.

2.4 Gaps between policy text and ground truth

Three patterns emerge from public information about NEP rollout:

1. Form before substance. Most universities have published FYUP curricula and signed up for ABC. Many have not staffed the multidisciplinary teaching that NEP envisions — partly because the faculty does not exist.
2. The AI integration is silo'd. Where NEP envisaged AI as a horizontal capability across disciplines (an English major using AI for textual analysis, a B.Com student using AI for financial modelling), most institutions have implemented AI as a vertical specialisation inside CSE. The horizontal integration remains rare.
3. Quality variance across the FYUP rollout is large. A four-year undergraduate at Delhi University looks different in curriculum, intent, and capability from a four-year undergraduate at a Tier-3 private college that re-labelled its existing 3-year B.Sc.

NEP 2020 set the right targets. The pace of substantive implementation depends on the same constraint everything else does — the supply of capable senior faculty.

3. What universities are currently teaching

A non-exhaustive but representative cross-section of where Indian universities currently sit on AI curriculum. Five archetypes, distinguished by faculty depth, infrastructure, and curriculum sophistication.

3.1 Archetype A — The AI Department

A standalone AI department with its own faculty, PhD pipeline, and dedicated facilities.

Real examples in India:

• IIT Hyderabad — Department of Artificial Intelligence, founded 2019; one of the country's first standalone AI departments ²⁴.
• IIIT Hyderabad — Kohli Center on Intelligent Systems (KCIS), founded 2015 with TCS funding; INAI, an applied AI research centre; iHub-Data under the NM-ICPS programme ²⁴.
• IIIT Delhi — Infosys Centre for Artificial Intelligence, founded 2016 ²⁵.
• IIT Madras, IIT Delhi, IIT Bombay, IISc — distributed AI capability across multiple departments (CSE, ECE, Computational Engineering, Robotics), backed by industry partnerships including IBM's AI Horizon Network at IIT-B and the IBM-IISc Hybrid Cloud lab ²⁶.

Estimated count of Archetype A institutions in India: 20–30.

3.2 Archetype B — The AI Specialisation

A B.Tech CSE program with an "AI&ML" or "AI&DS" tag, following the AICTE model curriculum, with 4–8 core AI courses, a final-year capstone, and (typically) a corporate partner badge.

Real examples include AI&ML specialisations at BITS Pilani, VIT, SRM, Manipal, LPU, Bennett, and a long tail behind them. The 921+ colleges figure from AICTE's enrolment data ⁷ is the best public count.

Quality variance inside this archetype is the largest of any. At the top end, indistinguishable from Archetype A undergraduate teaching. At the bottom, indistinguishable from Archetype D (the "label-only" department).

3.3 Archetype C — The AI Minor

A minor specialisation of 18–24 credits in AI, available to any student regardless of host degree. Designed for non-CS undergraduates — a B.Com student, a B.Des student, a journalism student — who needs AI literacy.

This is the archetype NEP 2020 most clearly envisaged. It is also the archetype least well staffed at present, because the faculty who can teach AI to a non-CS audience are even rarer than those who can teach it to a CS audience.

Real examples are emerging at the liberal-arts private universities (Ashoka, Shiv Nadar, Krea, OP Jindal, Plaksha), at IIT-D's M.Tech minor pathways, and at a handful of state university experiments. Total count in mid-2026: small — likely in the low tens of institutions running it well.

3.4 Archetype D — The Brochure AI Department

A department that exists primarily on the website. The curriculum is borrowed; the faculty teach Java and DBMS as their core load and AI on the side; the lab is a teaching lab with no compute; the capstone is a Kaggle notebook ported by the student.

Existence is hard to count, but circumstantial evidence — outdated curriculum critique ²⁷, polytechnic mismatch reporting, and the 80% engineer-employability shortfall from the 2016 NER cycle ⁹ — suggests this archetype is the modal Indian college experience for AI students at the Tier-2/Tier-3 end of the distribution.

3.5 Archetype E — No formal AI

The discipline does not formally exist in the curriculum. Students who want AI assemble it from electives, MOOCs, and personal projects.

Includes the large majority of arts, commerce, law, and humanities programs at the median Indian college, plus most teacher-education and polytechnic programs. The AICTE 2025 AI Year initiative is aimed precisely at this archetype.

3.6 Discipline-by-discipline coverage

A summary of public AI curriculum availability by host discipline:

The horizontal AI integration that NEP envisaged exists in pockets. It is not yet the default at the median campus.

4. The faculty supply problem

This is the binding constraint on every other lever in the system.

4.1 The macro number

India's higher education sector has roughly 43 million students and 1.6 million faculty members — a student-faculty ratio of approximately 27:1, against the UGC benchmark of 20:1 and the global-top-10 norm of approximately 10:1 ⁹.

4.2 The vacancy number

A Parliamentary Standing Committee report tabled in March 2025 disclosed that, as of 31 January 2025, 56.18% of professor-level positions were vacant across IITs, IIMs, NITs, IISERs, and central universities ². Sector-wide estimates put 30–40% of faculty positions unfilled ³.

Read these two numbers together. The country has an under-staffed system at the median, and a critically under-staffed senior tier at the apex. The senior-faculty shortfall at the apex is the more pernicious constraint because it caps the PhD pipeline that produces the next generation of teachers.

4.3 The AI-specific gap

We do not have an authoritative public number for "PhD-grade AI faculty in India." A defensible estimate, built from NIRF faculty disclosures at the top 25 CS departments, IIIT and IIT AI/ML faculty pages, and a scan of central and state university CS departments:

• ~2,500–4,500 PhD holders in India with active AI/ML research output (publications in the last 5 years in a top-tier AI venue or equivalent). [Source needed — primary survey planned]
• ~10,000–15,000 faculty teaching AI/ML in some form at higher-education institutions, with a wide spread of underlying training and currency. [Source needed]
• These numbers contrast with India's 921+ colleges offering AI&ML specialisations ⁷ and 14,000+ AICTE-approved colleges nominally in scope for AICTE's 2025 AI Year ⁸. The mismatch is the most consequential single fact in this report.

For comparison: Zeki, the talent analytics firm, counted 50,460 AI researchers and inventors of Indian origin (counted outside China) in 2025 ¹. The vast majority work in industry, in India and abroad, not in academia.

4.4 Faculty churn and the industry-academia gap

Industry pays better than academia, by a wide margin, for AI talent specifically. A freshly minted PhD with strong AI publications can command ₹40–80 LPA in industry; the equivalent assistant professor position at a UGC-pay-scale public institution starts around ₹12–18 LPA. The gap at the senior level is even wider. Private universities with the will and budget can close some of the gap; most cannot.

The Parliamentary Standing Committee report explicitly identified the problem: burned-out faculty cannot mentor PhD students effectively, which reduces the pipeline of future academics — the very people who could fill today's vacancies ².

4.5 Implications

Three implications for institutional strategy:

1. Concentration beats dispersion. Hiring two strong AI faculty into one department beats hiring one each into two departments. The PhD pipeline, the seminar series, and the grant flywheel only spin above a critical mass.
2. Industry-affiliated faculty are part of the answer. Visiting practitioners, adjunct industry faculty, and resident industry-trained faculty (the on-campus, full-time model) close part of the gap. The principal constraint is governance — UGC career structures do not easily accommodate them.
3. Faculty development beats faculty hire at the margin. For institutions that cannot compete for new AI PhDs, the marginal rupee is better spent retraining existing CS faculty in current AI tooling than chasing scarce new hires.

This is the loudest of the report's findings. If a Vice Chancellor reads only one section, it should be this one.

5. Infrastructure

For a long time the assumed bottleneck for Indian university AI was GPU compute. That assumption is now wrong, or at least much less right than it was three years ago.

5.1 The IndiaAI Mission

Cabinet-approved on 7 March 2024 with a budget of ₹10,371.92 crore (USD 1.25 billion) ¹¹. The mission committed ₹4,563.36 crore to compute capacity expansion over five years and explicitly committed to making compute available to academia, research bodies, MSMEs, startups, and government agencies registered with IndiaAI ¹⁰.

As of mid-2026, public disclosures put the total GPU count at:

• 38,000+ GPUs onboarded across empaneled cloud service providers, available at a subsidised rate of ₹65 per hour ¹¹
• The initial 18,693 GPU tranche includes 12,896 NVIDIA H100, 1,480 NVIDIA H200, and 7,200 AMD MI200/MI300 units ¹⁰
• A subsequent bidding round to add an additional 12,000–15,000 NVIDIA Blackwell-generation GPUs is in motion ¹¹

For an Indian university, the practical implication is that frontier training compute — the same H100 hardware US R1 institutions train on — is rentable at a fraction of cloud commercial rates, with the booking surface managed through IndiaAI.

5.2 The 2025–26 Budget

The 2025–26 Union Budget continued the trend ¹²:

• AI funding quadrupled to ₹2,000 crore at the centre
• MeitY's overall allocation rose 48% to ₹26,026.25 crore
• A new Centre of Excellence for AI in Education, with an outlay of ₹500 crore, was announced — the fourth such CoE the government has commissioned

5.3 Private and sovereign compute

Beyond government commitments, private operators have built out:

• Yotta Shakti Cloud, India's first sovereign AI infrastructure, with 16,000+ H100 GPUs and plans to scale past 32,000 by end-2025 ²⁸
• A pipeline of $100 billion-plus in announced data centre investment over the coming decade ²⁸
• An expanding NVIDIA-India partnership covering startups, research bodies, and education ²⁹

5.4 Where universities actually stand on compute

The disparity between aggregate available compute and compute actually accessible to a Tier-2 university CS department is still large. A typical Tier-2 private university's on-campus AI lab in 2026 has:

• One or two NVIDIA workstations with consumer-grade or low-end professional GPUs (RTX 4080/4090, A4000-class)
• A modest server with 2–8 GPUs of A100 or H100 class — increasingly common at Tier-1 private and IIT-tier campuses, rare elsewhere
• A handful of cloud credit grants from AWS, GCP, or Azure under educational programs
• Access in principle to IndiaAI compute, with varying degrees of utilisation

Tier-3 colleges typically have none of the above and rely on Google Colab and Kaggle free tiers for student work.

5.5 The on-campus AI Studio model

A growing pattern is the dedicated AI Studio or AI Lab on a private university campus, often co-branded with an industry partner. The typical structure includes:

• 4–8 GPU-equipped workstations
• One to four H100/A100-class server nodes for shared training jobs
• Visualisation, robotics, and prototyping equipment
• A resident faculty lead and one to three full-time engineers or research associates

The cost envelope for a credible on-campus AI Studio of this kind is roughly ₹3–8 crore in capex plus ₹1–2 crore per year in opex. By contrast, an in-house attempt to build the same capability without a partner can run materially higher and is rarely staffed adequately.

We are not yet aware of an independent third-party study that measures whether the AI Studio model translates into measurable capability outcomes for students. This is one of the most consequential measurement gaps we plan to close in 2027.

6. Industry-university partnerships

Industry partnership announcements have multiplied. The signal-to-noise ratio is the question.

6.1 The major public partnerships

• IBM + AICTE — National AI Lab. IBM partnered with AICTE to establish a National AI Lab at AICTE headquarters in New Delhi as a national hub for AI research, skilling, and innovation ¹³.
• IBM AI Horizon Network — IIT Bombay (since 2018); IBM-IISc Hybrid Cloud Lab (since 2021) ²⁶.
• Google Cloud + MSDE + Chaudhary Charan Singh University, Meerut. Announced 28 January 2026 as a national pilot for an "AI-enabled university" — Google's largest single-institution education engagement in India to date ¹⁵.
• Microsoft + NASSCOM + AICTE + GitHub + FutureSkills Prime — Future Ready Talent. A joint training and certification program in Microsoft Azure and applied AI ¹⁴.
• Cisco Virtual Internship Program in partnership with AICTE and FutureSkills Prime, reaching 100,000+ students with capstone-based virtual internships ¹⁴.
• NVIDIA — Anusandhan National Research Foundation partnership to bring AI to schools and universities ²⁹.

Each of these is real and verifiable. None, on its own, is a substitute for a university's own AI faculty and curriculum.

6.2 Patterns of partnership

We see roughly three patterns:

1. Skilling-as-a-service partnerships. A vendor delivers a certification curriculum to students through the university; the university brands the program and counts it as a co-curricular. Common, low capex for the university, low transfer of capability to the university's own faculty.
2. Lab partnerships. A vendor co-funds an on-campus lab in exchange for visibility, recruiting access, and (sometimes) a research collaboration. Higher capex, often higher real impact, depending on the resident faculty.
3. Research partnerships. A vendor funds named research positions, fellowships, or specific projects — Infosys at IIIT-D, Tata at IIIT-H, IBM at IIT-B and IISc. Concentrated at the apex; rare below it.

6.3 What distinguishes a working partnership from a press partnership

The cleanest tests we have found:

• Is there a named resident faculty member at the host university responsible for the work, with their own publications and a public profile?
• Are students doing project work directly with industry engineers, on a weekly or monthly cadence, on the campus?
• Has the partnership produced an external, verifiable artefact — a paper, an open-source release, a measurable hiring funnel — in the last 12 months?

A partnership that fails all three is almost certainly press. A partnership that passes two of the three is doing real work. The plurality of announced partnerships, in our scan, fail all three at the level of publicly verifiable evidence.

6.4 The advisory-board pattern

A common form of partnership is the industry advisory board — a panel of senior industry figures who meet annually with a department to advise on curriculum. Useful at the margins; rarely transformative on its own. The boards that move the needle are the ones whose members commit teaching time, not just review meetings.

7. Hiring outcomes

The hiring market is the most honest evaluator of what universities actually produce. Three slices of data.

7.1 The aggregate picture

Reports converge on a hiring market that is bifurcated and shifting:

• NASSCOM AI Adoption Index 2.0 (2024) estimates 600,000–650,000 AI professionals in India today, with demand projected to exceed 1.25 million by 2027 — a near-100% increase in five years ³⁰.
• NASSCOM-Deloitte (2024) estimates the immediate AI talent shortfall at roughly 50% — ~420,000 supply against ~600,000 immediate requirement ³⁰.
• Wheebox India Skills Report 2025 finds the engineering graduate employability rate at 71.5%, against an overall graduate employability rate of 54.81% ⁴.
• Mercer-Mettl India Graduate Skill Index 2025 reports overall graduate employability at 42.6% and AI/ML role employability at 46% ⁴.

The dispersion across these reports is itself informative — methodologies vary, and any single number should be treated with caution.

7.2 The salary distribution

For an AI-track fresher in 2025–26:

Sources: published salary survey data from Taggd, AIM, Hero Vired, and primary listings on LinkedIn and Naukri ¹⁶. Bangalore remains the highest-paying metro, with the average AI engineer salary in Bangalore reported at ₹12.67 LPA across experience bands ¹⁶.

The mass-employer numbers (TCS, Infosys, Wipro) are the most relevant for the median Tier-2 private university graduate; the product company numbers are most relevant for the top quartile.

7.3 The volume picture

• TCS plans to recruit roughly 42,000 freshers in FY26 ¹⁷
• Infosys targets 20,000+ ¹⁷
• Wipro targets 10,000–12,000 ¹⁷
• The Global Capability Centre (GCC) sector continues to expand and now sits among the larger AI-hiring employers in the country

The IT services sector has not collapsed, contra some commentary. Its composition is shifting — the share of generalist application-development roles is declining, the share of AI-adjacent and cloud-AI roles is rising — and the pace of that shift is faster than most curriculum cycles can match.

7.4 Implications for universities

Three observations from the data:

1. The placement-ranking arms race is a lagging indicator. A university that ranks highly on NIRF placement today is reporting the success of CSE graduates hired into roles that may be partly automated in three to five years. Tracking the role mix is more informative than the average package.
2. Portfolio beats degree, and the gap is widening. Freshers with verifiable AI portfolios — a working LLM-based product on GitHub, a Kaggle profile, a peer-reviewed paper — are getting offers in the ₹12–25 LPA band regardless of the institution they graduated from. Without a portfolio, the same degree at the same institution clears at half that.
3. The skill-test bar is rising. The pre-screen used by product companies has shifted from "can you code FizzBuzz" to "can you debug this fine-tune of an open-source LLM in the next hour." Universities whose curriculum has not absorbed this shift will see placement numbers degrade even as their nominal AI curriculum expands.

For the per-skill, per-band breakdown of what hiring managers actually test for, refer to the AI Skills Index, India 2026, which covers that ground in detail.

8. The student experience

A description in prose, framed honestly. The intent is to give a Vice Chancellor or a parent — and ourselves — a sober view of what the AI-track student experience actually feels like at the median Tier-2 private campus, before we make claims about outcomes.

8.1 The B.Tech CSE (AI&ML) student at a Tier-2 private university

Four years, ₹2–4 lakh in tuition per year. Year 1 is mostly common engineering — calculus, discrete maths, physics, programming in C/C++, an introduction to data structures. Year 2 introduces the AI specialisation electives — Python, statistics, ML, sometimes a deep learning course. Year 3 deepens into specialised electives (NLP, computer vision, sometimes RL); some students take internships in the summer between Year 3 and Year 4. Year 4 is heavily project-driven and capstone-focused, with placement activity from August through January.

What this looks like in practice depends almost entirely on the faculty. A student at a well-staffed Tier-2 private with three or more capable AI faculty members in CSE will have:

• Two to four hands-on AI courses with running code and meaningful assignments
• Access to a modest lab — usually a single GPU node or a Google Colab tier
• A capstone project that produces a working artefact, sometimes with an industry mentor
• One or two industry guest speakers per semester
• Recruitment processes that include the major IT services companies and some product startups

A student at a less well-staffed institution will have the same nominal curriculum and a measurably worse experience — fewer functioning labs, more lecture-based teaching, less project work, and a placement funnel concentrated on a smaller set of bulk hirers.

8.2 The AI Minor student at a liberal-arts private university

Three or four years, ₹4–10 lakh in tuition per year. The host degree is BA, BSc, or BBA. The AI minor is 18–24 credits, taught by either a borrowed CS faculty member or (less commonly) by a dedicated AI minor lead. Course content varies but typically includes:

• Introduction to programming and Python
• Statistics and probability foundations
• ML basics and a hands-on tooling course
• One discipline-relevant elective (AI for the social sciences, computational humanities, design with generative AI)
• A capstone project

The strongest of these programs produce students who are demonstrably AI-capable in their host discipline. The weakest produce students who completed Python and called it AI literacy.

8.3 What students consistently report

From public commentary, student surveys, and reporting:

• The gap between what they learn in class and what they learn in side-projects is the single most cited issue. Students at well-resourced campuses have side-projects that are more sophisticated than the curriculum; at less well-resourced campuses, students rely on YouTube, MOOCs, and tutorial blogs as the de facto curriculum.
• Industry guest interactions are the single most appreciated component, when they happen with practitioners who actually code in production.
• Capstones are most valued when they produce a public artefact the student can show on a resume or portfolio.
• Placement preparation occupies the senior year more than coursework does, at most institutions below the apex tier.

The student experience at the median Indian college is one of structural under-investment partially compensated for by personal initiative. The students who succeed are typically the ones who would have succeeded under any system; the system itself adds less value at the median than its boosters claim and more value at the right tail than its critics admit.

9. Comparison with global benchmarks

Like-for-like comparisons across higher-education systems are difficult, especially for AI specifically. Three comparison points.

9.1 United States

The US R1 university tier (top ~150 doctoral-research universities) is the global benchmark. Per-institution AI research output, AI faculty headcount, compute access, industry-research integration, and PhD pipeline are all materially deeper than the Indian equivalent at the median.

Where India compares favourably: the velocity of CS enrolment growth, the public-compute commitment via IndiaAI, and the absolute size of the talent pool. Where the gap is widest: depth of senior faculty, breadth of peer-reviewed research, and the maturity of the academia-industry research feedback loop.

9.2 China

China has invested aggressively in AI inside its top universities. Tsinghua's College of AI, founded April 2024 and led by Turing Award laureate Andrew Yao, is the clearest signal of institutional commitment — a standalone college dedicated to AI, modelled explicitly on MIT and Stanford, with joint training platforms with Alibaba, Tencent, ByteDance, SenseTime, and Huawei ²¹. Peking, Shanghai Jiao Tong, Zhejiang, and other tier-1 Chinese universities have followed similar patterns at smaller scale.

The Chinese model is more centrally directed than the Indian model, has more concentrated industry-academia integration, and produces measurably more frontier AI research per top department. India's strengths against China are softer — a larger English-speaking talent pool, deeper integration with the global AI labour market, and the IndiaAI Mission's commitment to broad-access compute. The gap to China at the apex remains real.

9.3 Europe / Singapore / Israel

• Europe — strong AI research base concentrated at ETH Zurich, EPFL, Oxford, Cambridge, Imperial, TU Munich, INRIA, KTH. Smaller absolute scale than the US or China; very deep on AI safety and theoretical foundations.
• Singapore — NUS and NTU run AI programs that are competitive with the US R1 tier in specific subfields. The single-country, single-language model produces tight integration between government, industry, and academia.
• Israel — small population, disproportionate AI research output via the Technion, Hebrew University, Tel Aviv, and Weizmann. The pattern — concentrated talent + deep industry integration via national service and the startup pipeline — is not directly translatable to India but instructive on what concentration can achieve.

9.4 India's position

India ranks third on the Stanford 2025 Global AI Vibrancy Index ¹, second worldwide for the number of AI researchers and inventors outside China ¹, and fourth globally in QS Computer Science & Information Systems rankings with 42 institutions ranked ²⁰. These are real strengths. They are also dominated by output from the apex tier and a handful of Tier-1 private institutions. The full system is still some distance from the global frontier on the measures that matter most for higher education — depth of senior faculty per institution, average research output per faculty member, and quality of teaching at the median campus.

10. What's working

A non-exhaustive list of patterns that are, on public evidence, producing real outcomes.

10.1 The IndiaAI Mission as a public-compute backstop

The decision to subsidise GPU access at ₹65/hour and make it available to academia removes one historical bottleneck and changes the marginal cost of running serious AI research at a Tier-2 or Tier-3 institution. Take-up is uneven and the booking surface has friction, but the structural commitment is the right one ^{10 11}.

10.2 Concentrated AI departments at the apex

The IIT Hyderabad AI department ²⁴, IIIT Hyderabad's KCIS and INAI ²⁴, IIIT Delhi's Infosys Centre for AI ²⁵, and the IIT Bombay and IISc IBM partnerships ²⁶ are concrete examples of concentration above critical mass producing measurable research output and a sustained PhD pipeline. The pattern is replicable — at the cost of focused investment and faculty hiring discipline.

10.3 AICTE model curricula

For all the criticism the regulator deserves on pace, the AICTE model curricula for AI&ML and AI&DS ²² have given the long tail of private engineering colleges a credible reference point. The curriculum is not the bottleneck — competent teaching of it is — and a well-staffed institution using the model curriculum can deliver a credible program.

10.4 The B.Tech enrolment surge in AI specialisations

The 67% B.Tech enrolment growth since 2017–18 ⁶, concentrated in CS and AI specialisations, is responding to a real labour-market signal. The market for AI talent in India will absorb the supply, even if quality varies, and the demographic momentum behind India's AI workforce growth is genuine ³⁰.

10.5 Liberal-arts AI minors at the Tier-1 private universities

The handful of well-run AI minor programs at liberal-arts private universities are the strongest current evidence that NEP 2020's multidisciplinary aspiration can work in practice. These programs are small, expensive, and concentrated at a few institutions, but they demonstrate the model.

10.6 The Stanford 2025 ranking validates the macro story

India's rise to third in the Global AI Vibrancy Index from seventh in 2023 ¹ is a real macro signal. The country's AI ecosystem is improving on most measurable indicators — research output, patents, talent base, public investment — even where individual institutions are uneven.

11. What's not working

A similar non-exhaustive list of patterns that, on public evidence, are not working.

11.1 The brochure AI department

The pattern is well-documented in public commentary ²⁷: a department announces an AI&ML specialisation, recruits students at the higher tuition premium, and delivers a curriculum that is one or two years behind the AICTE model and ten years behind production AI tooling. The faculty are CSE generalists with no AI publication record; the lab is a teaching lab without compute; the placements are with the same IT services companies that would have hired the students anyway.

The cohort that suffers most is the Tier-2/Tier-3 private student who paid the premium expecting an AI specialisation and received a brand.

11.2 Curriculum lag against tooling change

The AI tooling landscape changes meaningfully every six months. The Indian university curriculum cycle, in formal terms, has been four-to-six years for major revisions. The mismatch is structural. Even institutions with capable faculty struggle to update tooling content fast enough.

NEP 2020's flexibility provisions — the ABC, the credit framework, the multiple-entry/exit — provide some relief, but the underlying problem is that universities are governance-bound to update at a slower cadence than industry. This is a feature, not a bug, of academic governance generally; it requires a deliberate offset, not a structural cure.

11.3 The faculty pipeline gap

The faculty supply problem (Section 4) is the single largest blocker. Until the pipeline of PhD-grade AI faculty grows materially faster than the rate at which AI&ML specialisations are launched at new institutions, the underlying quality problem will compound, not improve.

11.4 Implementation theatre around NEP 2020

NEP 2020 is being implemented in form at most institutions and in substance at relatively few. The four-year UG rolled out at 105+ universities ¹⁸ but the multidisciplinary teaching it envisaged remains absent at the median campus. The ABC and NCrF are operational ¹⁹ but adoption is partial. The AI integration that was supposed to be horizontal is, in practice, mostly vertical and silo'd inside CS.

11.5 Industry partnerships at the press-release layer

The plurality of announced industry-university AI partnerships, on public evidence, are MoU-stage and do not pass the three working-partnership tests we offered in Section 6.3. The distinction matters — universities that mistake a press partnership for a working partnership end up with neither the visibility benefit nor the capability transfer.

11.6 Polytechnic-to-industry pipeline

India's 1,800+ polytechnics are a separate report's worth of analysis. The shortest summary: their curricula were designed for a labour market that AI is reshaping faster than the polytechnic system can adapt, and the gap between placement day and actual labour-market positioning two years later is the system's loudest unaddressed problem.

11.7 Inequality of access

Within institutions, the AI specialisation premium pulls toward students who can afford the higher tuition. Across institutions, the gap between the apex tier and the median Tier-3 campus on AI capability is, on the evidence, widening rather than narrowing. The IndiaAI Mission's public compute commitment is part of the answer; deeper changes — to faculty mobility, to credit transfer, to scholarship support — are needed for the others.

12. What changes by 2030

Five things that, on our reading of the evidence, are more likely than not to change in the next four years. Probabilities are our subjective estimates, not modelled forecasts.

12.1 AICTE makes AI mandatory in some form across all engineering disciplines (probability: high)

AICTE's 2025 AI Year ⁸ is a strong signal of intent. The most likely next move is a mandate that a baseline AI literacy course be required across all B.Tech disciplines, not just CS. The regulator-led pattern of curriculum mandate is well-precedented.

12.2 The faculty supply gap narrows partially, mostly through industry-affiliated routes (probability: moderate)

A continuation of the current pattern — adjunct industry faculty, named research positions, on-campus resident AI faculty at private universities — will close part of the gap. A more structural fix — UGC career structures that allow industry crossovers, salary normalisation, and PhD pipeline expansion — is harder and slower. We expect the gap to narrow at the top of the system and widen at the median.

12.3 The on-campus AI Studio model becomes the default at Tier-1 private universities (probability: high)

The economic and reputational incentives favour this strongly. By 2030, a Tier-1 private university without a dedicated on-campus AI facility will be the exception, not the rule. The independent question — whether the studio model produces measurable capability outcomes — remains open. We will measure it.

12.4 The bifurcation of hiring outcomes deepens (probability: high)

The gap between portfolio-strong AI graduates and generalist CSE graduates is widening, not narrowing, on every indicator we have. By 2030 we expect the median Indian AI graduate's outcome to be either materially better (top quartile, ₹15–40 LPA at product companies) or materially worse (bottom quartile, struggling for placement against shrinking IT services bulk-hire numbers) than the current median.

12.5 The horizontal AI integration NEP envisaged becomes real at a small set of universities, not the median (probability: moderate)

The NEP vision of AI as a horizontal capability — taught in journalism, in law, in design, in management — will be visible at 30–50 institutions by 2030. It will not be the median experience. The constraints are the same ones constraining everything else: faculty supply and the slow pace of curriculum governance.

12.6 What we do not expect to change

We do not expect the polytechnic-system mismatch to resolve itself by 2030. We do not expect the senior-faculty vacancy rate at the apex to fall below 35% by 2030. We do not expect Indian universities to close the gap to the US R1 tier on per-institution AI research output by 2030. These are longer-cycle problems than the political and budget cycles that nominally address them.

13. What we'd like to measure next year

Volume 2 of this report, planned for May 2027, will extend the dataset in seven specific ways. We name them here so the community knows what we intend and so readers can hold us to it.

1. A primary survey of 200–400 Indian universities on AI curriculum, faculty headcount, lab capacity, capstone outcomes, and placement role mix. Sample design and methodology to be published in advance.
2. A faculty census — the actual count of PhD-grade AI faculty in Indian higher education, by institution and by sub-field. The current absence of this number is a sector-level embarrassment we propose to fix.
3. An on-campus AI Studio measurement framework — a published rubric for what a serious on-campus AI facility looks like, with capability outcomes data from a sample of partner institutions.
4. Graduate capability data, longitudinal — a 12-month tracking cohort of AI-specialisation graduates across 10–20 institutions, measuring role placement, salary trajectory, and skill currency at 6 and 12 months out.
5. NEP 2020 implementation depth, not just form — a per-university scorecard of substantive NEP implementation: multidisciplinary teaching ratio, ABC utilisation, FYUP exit-rate distributions, multi-disciplinary minor uptake.
6. The Skills Index integration. Where possible, we will harmonise the institutional landscape data in this report with the per-capability Skills Index data published separately, to enable institution-by-skill cross-tabulation.
7. A failure-mode taxonomy. The "brochure AI department" pattern is real but currently anecdotal. We propose to characterise it with sample data and case studies so prospective students, parents, and regulators have a concrete reference.

We invite Vice Chancellors, Deans, faculty, and AI hiring managers to contact us at research@withkompas.com if they want to contribute primary data to Volume 2.

14. Method and limitations

A direct statement of what this report is and is not.

14.1 Method

The report is a synthesis of public sources: government data (AISHE, NIRF, AICTE, UGC, MeitY, Department of Higher Education press releases), industry research reports (NASSCOM, Deloitte, McKinsey, Bain, Wheebox, Mercer, Stanford AI Index), independent academic and policy commentary, university websites, and reporting in Indian and international press. Where we cite a specific number, we cite the source. Where we estimate, we say so. Where we do not have a defensible number, we mark [Source needed].

We have not run primary surveys for Volume 1. We have not validated every university website claim against an independent source. We have not had access to non-public regulator data, university confidential data, or proprietary industry data.

14.2 Known limitations

• Latest AISHE data is 2021-22. The 2022-23 and 2023-24 surveys have not been publicly released at the time of writing. We use 2021-22 figures and flag this where it matters.
• Faculty headcount estimates are conjectural. The faculty-specific numbers in Section 4 are our best public-source estimates and should be improved with primary survey data in Volume 2.
• University-level claims rely on public information. Where a university has not published its faculty list, lab capacity, or placement data, we cannot independently verify what it tells students and parents.
• The tier classification is informal. The Tier 0 / Tier 1 / Tier 2 / Tier 3 shorthand is the working language of industry and admissions, not an official taxonomy. We use it because it is useful, while acknowledging its limits.
• The 2026 dating. The report is published in May 2026, citing source documents through that date. Several initiatives (AICTE 2025 AI Year, IndiaAI Mission expansion, AI in Education CoE) are mid-execution and will look different by year-end.

14.3 What this report deliberately does not cover

• The skill-rubric, per-capability assessment of graduates. That is the subject of the parallel AI Skills Index, India 2026.
• The AI industry in India more broadly. Many strong reports exist (NASSCOM AI Adoption Index ³⁰, Stanford AI Index ¹, Bain India Enterprise Technology Report ³¹, McKinsey India work ³²); we cite them where relevant rather than reproducing their content.
• Primary or secondary education AI integration. NEP 2020 covers school education in depth; that is a separate report's worth of work.
• A vendor comparison of edtech AI offerings. This is not that document.

14.4 Corrections

If you find a factual error, please write to research@withkompas.com with the page reference and the source you believe is correct. Corrections will be published in a changelog and reflected in subsequent versions of this document.

15. Sources and citations

Additional reference sources

• National Institutional Ranking Framework (NIRF) 2025 Report: https://www.nirfindia.org/nirfpdfcdn/2025/pdf/Report/IR2025_Report.pdf
• AISHE Homepage: https://aishe.gov.in/
• AICTE Approval Process 2025-26: https://aicte.gov.in/approal-process-2024-2027
• UGC: https://www.ugc.gov.in/
• IndiaAI portal: https://indiaai.gov.in/
• Ministry of Education NEP Regulations & Guidelines: https://www.education.gov.in/nep/regulations-guidelines-framework