Why 96% of Govt AI Projects Stall—And It’s Not About Money

India's sovereign AI push remains heavily centralised. While states are competing to attract AI data centres, GPUs, and AI startups, the bulk of AI investments still come from the Centre through the ₹10,371 crore IndiaAI Mission. States are largely acting as facilitators by providing land, power, data centre policies, startup incentives, and AI labs rather than committing large standalone budgets of their own.

Recently,a new IDC study commissioned by Dell Technologies mentions that India is emerging as one of the most distinctive sovereign AI markets in the Asia-Pacific region, with 96% of government leaders actively pursuing sovereign AI strategies. 

While 46% are evaluating sovereign AI technologies and an equal number are running proof-of-concept projects in their departments, only a small fraction have progressed to large-scale deployments. As per the report, only 4% of Indian government organisations have moved beyond experimentation to make significant investments in sovereign AI.

The numbers point to a broader challenge facing India's public sector AI ambitions. While agencies are increasingly experimenting with AI, converting pilots into production-grade systems remains difficult.

Industry experts argue that the problem is not a lack of interest or funding, but the complexities of operationalising AI across fragmented government systems, legacy databases, procurement frameworks, and organisational processes.

AI PoCs are Treated as Experiments

There are some notable examples of government departments implementing AI capabilities. The Ministry of Electronics and IT is spearheading BHASHINI for real-time multilingual translation, while the Indian Railways has deployed Machine Vision Inspection System to detect any hanging, loose, or missing components of moving trains.

However, most government departments tend to treat AI as trials.

"Most AI PoCs fail to reach production because organisations treat them as technology experiments rather than business transformation initiatives," Biswajeet Mahapatra, Principal Analyst at research firm Forrester, tells AIM.

He states that projects often struggle due to unclear business outcomes, poor data quality, weak governance frameworks, integration challenges, and the absence of an AI-ready operating model.

"Organisations that succeed typically align AI initiatives to measurable business value, establish accountable governance, and build the capabilities needed to scale, operate, and continuously improve AI solutions beyond the pilot stage," he notes.

Mahapatra adds that India's public sector has already made significant progress in digital transformation and AI experimentation. However, the next phase requires agencies to move beyond pilots and focus on operationalisation.

"Many agencies have made progress in digital transformation, data modernisation, and AI experimentation, but scaling AI requires stronger connections between data, governance, procurement, business ownership, and workforce readiness," he explains.

He argues that government agencies need repeatable processes, shared platforms, and stronger accountability structures that can make AI part of everyday governance rather than isolated innovation projects.

Mahapatra also calls for a more agile procurement framework tailored specifically for AI deployments.

"Traditional public sector procurement models are often designed for large, well-defined technology projects, while AI solutions require rapid experimentation, iterative development, and continuous refinement," he states.

He advises that a fast-tracked AI procurement mechanism could help agencies evaluate solutions faster, reduce pilot fatigue, and shorten the gap between proof-of-concept and deployment, while maintaining governance and transparency safeguards.

Real Challenge Lies Beyond Funding

NSN Murty, Partner and Government Consulting Leader at Deloitte India, argues that AI pilots should not be viewed as failures simply because they do not immediately reach production. "To put it in perspective, PoCs don't fail—they provide direction," Murty states.

He argues proofs-of-concept, minimum viable products, and pilot deployments are all part of a broader "test bed" process, particularly when government departments are still defining the outcomes they expect from AI-enabled transformation.

Murty believes that many government systems remain fragmented, federated, and partially digitised, limiting the effectiveness of AI deployments. He points to citizen grievance management as an example. "A citizen grievances management system enabled through a multilingual/native language LLM and AI can effectively capture grievances. However, resolution depends on the backend department's technological readiness," he says.

While many observers argue that government agencies are hesitant to invest in AI, Murty disagrees. "Governments are not holding back investments in AI PoCs," he claims.

He points to the IndiaAI Mission, under which states have received support for sector-specific AI use cases, Centres of Excellence, capacity-building programmes, and startup engagement initiatives.

However, he stresses that financial support alone cannot guarantee successful deployment.

"Successful deployment requires backend modernisation and the technical capacity to design, build, and operate AI solutions at scale," he adds.

The Rise of PoC Fatigue

Alouk Kumar, CEO of Inductus Group, who is working closely with governments on GCC initiatives, identifies that the sectors that have moved fastest on AI—banking, financial services, retail, healthcare, and logistics—share one common characteristic: they already have large volumes of structured digital data.

"Because they already have structured data sitting ready, running an AI experiment is a logical next step rather than a massive leap into the unknown," Kumar mentions.

He explains that AI systems often perform well in controlled environments but struggle when connected to fragmented real-world datasets spread across multiple departments and systems.

Murty agrees that data quality is an issue, which can reduce the accuracy of AI outputs. "Another challenge is integrating AI with backend databases that are incomplete or contain discrepancies, as is the case with several government databases that are not regularly updated," he states.

Outside these sectors, however, enthusiasm for AI has been tempered by what Kumar describes as growing "PoC fatigue."

"Many business leaders spent the past couple of years funding flashy AI pilots that ultimately delivered little to no actual business value."

As a result, organisations are increasingly demanding measurable returns before approving large-scale AI budgets.

"There is also the reality of legacy tech: plugging modern AI tools into decades-old core systems is messy, expensive, and slow," Kumar believes.

Government is Still Building Trust

Government officials familiar with ongoing AI initiatives tell AIM that pilots are serving another purpose beyond technology validation: building trust between government agencies and private-sector partners.

"This is the first time we are experimenting with something like this, tying up with corporations and giving them access to government data," officials from the Karnataka government state on the condition of anonymity.

They say private companies are increasingly interested in working with datasets ranging from transport and metro systems to land records and revenue department information.

"All this data is available. But how do we make it more meaningful? How do we reduce the turnaround time? We have to do a pilot," they explain.

According to the officials, pilots are important as they help validate use cases and establish confidence before any wider deployment takes place.

"We have to build trust in each other. Then you scale it up further," they affirm.