The honest headline on AI adoption in India and the Gulf is that both markets have bought in and neither has broken through. In India, the NASSCOM-EY AI Adoption Index 2.0 — a study of 500 companies representing roughly 75% of GDP — scores aggregate enterprise AI maturity at 2.47 out of 4, with 87% of companies stuck in the middle stages and only around 2% operating as genuine leaders. In the GCC, McKinsey's survey of the region reports adoption jumping from 62% to 84% between 2023 and 2025 — close to the global 88% — yet only 31% of organisations have scaled AI beyond pilots, and only 11% report earnings impact of 5% or more.
Two different economies, two very different funding environments, one identical shape: wide adoption, narrow production. This piece assembles the verified numbers on both markets, with their sources and their caveats, and then draws out what the pattern means if you are the one deciding where an AI budget goes.
A note on the data before using it. The NASSCOM-EY figures come from the 2024 edition of the index — the latest available, but roughly two years old now, and newer surveys suggest talent and data-readiness have risen as barriers alongside the original blockers. The GCC figures rest on a small self-reported sample of roughly 130–140 organisations, so they are directional rather than precise. And the major consulting houses that produce this research have an obvious interest in framing "value gaps" that consulting closes. None of that invalidates the numbers; it just sets how hard to lean on any single one.
How mature is AI adoption in India?
The NASSCOM-EY index describes a market that has left the starting line and bunched up in the middle of the track. A 2.47-out-of-4 average with 87% mid-stage means the typical large Indian enterprise has run pilots, proven feasibility somewhere, and not yet turned any of it into a production capability that moves the P&L. The roughly 2% who have are the outliers, not the pattern.
The blockers the index identifies are worth reading carefully, because none of them is a technology problem: inconsistent leadership commitment, ad-hoc PoC-driven budgeting, difficulty selecting use cases, and weak partner-led innovation. Budgets released pilot by pilot, with no committed path to production, produce exactly the mid-stage bunching the maturity scores show — a mechanism dissected in why enterprise AI projects fail. Notably, NASSCOM's own prescription is structural: it explicitly recommends that enterprises partner with tech SMEs for swift PoC-to-production, an acknowledgement that the missing ingredient is execution capacity rather than ambition.
Sector by sector, the index finds manufacturing and telecom/media the most AI-mature industries, healthcare lagging, and BFSI — despite its data wealth — still more than 50% PoC-heavy. That last figure is a useful corrective to intuition: the sectors with the most data are not automatically the ones that ship, because shipping depends on workflow ownership and organisational commitment, not data volume.
The money, meanwhile, has started moving in a more serious way. Gartner reports that Indian CIOs began allocating GenAI budgets beyond proof-of-concept in 2025, and that Indian IT-services spending grew 11.4% to $33.5 billion, with consulting driving the growth. Read together with the maturity data, the picture is a market where the funding constraint is easing faster than the execution constraint — budgets are now ahead of the capacity to convert them.
Where should a mid-stage Indian firm actually start? The research offers an unusually specific answer. PwC's work on Indian manufacturers at lower AI maturity names a single entry use case: using GenAI to respond to RFPs and draft proposals. It is a telling choice — tender and proposal work is high-volume, document-heavy, deadline-bound, and measurable, which makes it one of the few workflows where a first production system can prove itself inside a quarter rather than a fiscal year. The general principle transfers beyond manufacturing: the fastest route out of the 87% is a workflow whose output the business already counts.
Two India-specific operating realities sit underneath these numbers. First, the compliance layer: any AI system touching personal data now operates under the DPDP Act, and working through a DPDP compliance checklist for AI systems belongs in project scope, not as an afterthought. Second, the procurement layer: an enormous share of Indian B2B and B2G workflow runs through public tendering, which makes GeM portal tender automation one of the most concrete, high-volume places the adoption gap shows up as unread documents and missed deadlines.
What do the GCC adoption numbers show?
McKinsey's GCC survey — small sample, self-reported, and to be read accordingly — shows a region that has closed the adoption gap with the world and not yet closed the value gap. Adoption at 84% sits near the global 88%. But only 31% have scaled, and only 11% clear the bar of AI contributing at least 5% of earnings. The same survey reports 89% of organisations planning to increase AI budgets and roughly three-quarters reporting genuine executive commitment — materially stronger top-level conviction than India's blocker list implies. The GCC's version of the gap is not funded reluctantly; it is funded enthusiastically and still not crossed.
One barrier stands out in the GCC data: output inaccuracy, cited by 53% of respondents as a top concern. That is a signal worth taking literally. It says the region's problem has moved past "should we use AI" to "we cannot yet trust what it produces" — a production-engineering and evaluation problem, which is further along the adoption curve than a strategy problem, and a large part of why governance and verification requirements now shape Gulf procurement.
Saudi Arabia makes the pattern unusually crisp because the spending and the shortfall are both measured. Per PwC's Middle East analysis, Saudi firms spend around 8% of revenue on AI against a 6% global average — yet report ROI of roughly 30% against 37% globally. Outspending the world while under-collecting on returns is the value-capture gap in a single comparison, and it echoes the region's structural bet: money is the abundant input, execution the scarce one.
The Saudi numbers also show governance operating as a first-class buying criterion rather than a checkbox: 62% of Saudi organisations have documented Responsible AI frameworks against 47% globally, backed by SDAIA's mandatory AI framework and the PDPL. The state has built the infrastructure to match — SITE Cloud, the PIF-owned sovereign cloud, and DEEM Cloud running IBM watsonx.ai alongside SDAIA's ALLaM Arabic model for government workloads. For any organisation selling into or operating near the Saudi public sector, the practical entry points are the SDAIA and NCA cloud requirements and the procurement mechanics of an Etimad tender workflow; the wider policy context sits under Vision 2030's approach to AI procurement.
The UAE runs the same state-backed playbook with its own institutions, and its regulatory layer is maturing in parallel — the practical picture is covered in sovereign AI in the UAE and the compliance mechanics in UAE PDPL compliance for AI. Across the Gulf, "sovereign" has become a load-bearing word in AI procurement, and it pays to be precise about what sovereign AI actually means before a vendor uses it on you.
India vs GCC: the numbers side by side
| Dimension | India | GCC |
|---|---|---|
| Headline measure | Maturity 2.47 / 4; 87% mid-stage; ~2% leaders (NASSCOM-EY, 500 firms, 2024 edition) | Adoption 62% → 84%, 2023–2025 (McKinsey, small self-reported sample) |
| Scaled / leading cohort | ~2% "Leaders" | 31% scaled; 11% with ≥5% earnings impact |
| Budget direction | GenAI budgets moved beyond PoC in 2025; IT-services spend +11.4% to $33.5B (Gartner) | 89% plan to increase AI budgets |
| Executive commitment | A named blocker — inconsistent | ~75% report committed executives |
| Distinctive barrier | Ad-hoc PoC budgeting; use-case selection | Output inaccuracy (53%) |
| Governance posture | DPDP Act; maturing | Saudi: 62% documented RAI frameworks vs 47% global; SDAIA framework + PDPL |
| Sector signal | Manufacturing, telecom most mature; BFSI >50% PoC-heavy | Saudi spends 8% of revenue on AI vs 6% global; ~30% vs 37% ROI |
The asymmetries are as informative as the similarities. India's constraint profile is leadership commitment and budgeting discipline with cost pressure everywhere; the GCC's is trust in outputs and value capture with money largely solved. A vendor or a buyer moving between the two markets is not carrying the same problem across a border.
What the shared pilot-to-production gap means for buyers
Strip the regional detail and one structural fact remains: in both markets, the distance between adopting AI and profiting from it is the pilot-to-production crossing, and most organisations are standing at it. Three practical conclusions follow for anyone allocating a budget.
Budget for the crossing, not the pilot. The verified Indian blocker — ad-hoc PoC-driven budgeting — is a spending pattern any company can simply decline to repeat. Fund use cases with a committed path: pilot, integration, acceptance criteria, and run costs approved as one decision. A pilot funded without its production budget is, on the evidence of both markets, most likely a donation.
Treat accuracy and governance as scope, not overhead. The GCC's leading barrier is output inaccuracy, and Saudi buyers have responded by making documented governance a purchasing criterion. That is the mature response. Evaluation against your own documents, monitoring in production, and a written compliance mapping — DPDP in India, PDPL and SDAIA requirements in the Gulf — belong in the project's definition of done from day one.
Start where the documents pile up. In both regions, the workflows where the gap is most visible — and most cheaply closed — are procurement- and compliance-shaped: tenders read by hand, proposals assembled under deadline, invoices and contracts keyed into systems. These are high-frequency, measurable, and already digital at one end, which is why they keep surfacing as the recommended entry point in the adoption research. A first production win in a document workflow builds the internal evidence and the organisational muscle that the second, harder use case will need.
Judge partners on production evidence. In a market whose bottleneck is implementation capacity, the discriminating question for any vendor or consultancy is not their strategy credentials but their production record: systems running, referenceable, integrated with clients' real ERPs and document flows. India's own industry body pointing enterprises toward tech-SME partnerships for PoC-to-production is a strong hint about where that evidence tends to live.
The adoption race that the headline numbers describe — percentages of firms "using AI" — is effectively over in both regions, and it was never the race that mattered. The numbers that will separate the next cohort of leaders are the production numbers: the 31% scaled becoming 50%, the 2% of Indian leaders becoming 10%. Those figures move one shipped workflow at a time, which is to say they are not really market statistics at all. They are the sum of individual companies deciding to fund the crossing instead of another pilot — a decision available to any buyer in either market, this budget cycle.