The AI adoption crisis has evolved into a cost management crisis, with companies now facing unprecedented AI bill increases. Pylon, a support platform for B2B companies, experienced a 3.5x overnight cost jump from Anthropic, with monthly bills rising from $400,000 to $1.4 million per year. Engineering teams at Pylon spent $3,107 per person per month on Claude alone, while the support team's AI costs reached more than $11,000 per month across a 10-person team. The marketing team spent $739 per person yet hasn't seen results to justify that level of investment, with AI-written copy described as "terrible" and far from perfect. Despite these challenges, companies continue to invest heavily in AI infrastructure, with 42% prioritizing upgrading IT infrastructure and 39% focusing on cybersecurity over AI-specific initiatives.
According to Kyndryl's Readiness Report, which surveyed 3,700 business and technology leaders across 21 countries in 2025, a quarter of mission-critical servers, storage and networking systems are already at or nearing end-of-service. More than half of those surveyed, 57 per cent, say innovation is being delayed by foundational technology issues. Among organisations that have not seen positive returns on their AI investments, 35 per cent point to integration difficulties as the reason. The report treats end-of-service infrastructure as an urgent opportunity for modernisation, with 22 per cent of surveyed companies saying technical debt is actively holding their organisation back. As companies face mounting AI costs, the urgency to modernise infrastructure has become critical for cost optimization.
A separate Kyndryl report surveyed 1,100 business and technology leaders across eight markets, including India, revealing a parallel human readiness crisis. Even as 57 per cent of leaders say AI is now deployed broadly or embedded in core business processes, the share who believe their workforce is actually ready to work with it fell to just 23 per cent, down six points from the year before. 79 per cent of leaders agree that the speed of AI adoption will outpace their organisation's ability to adapt its workforce, governance and operating model. The gap is particularly stark between leaders and employees, with non-C-Suite respondents more likely than their C-Suite leaders to describe skills gaps and role redesign as major implementation challenges.
The scale of infrastructure problems varies sharply by industry, with Kyndryl's data showing retail highest at 32% of systems at or near end-of-service, followed by manufacturing and telecommunications at 29% each. Banking and financial services sit close to the overall average at 26%. As reported by Kyndryl, nearly three in four Indian organisations see their AI efforts stall once they move past proof-of-concept, with the same reasons affecting global markets: fragmented data, legacy systems, and business processes not built for AI. The sectors under the most pressure to adopt AI quickly are also carrying the heaviest legacy load.
Despite extending significant autonomy to AI systems, only 25% of leaders say they completely trust AI systems that make decisions or act without human oversight. Yet 66% have already given AI permission to read from and write to core systems of record fully autonomously. Governance has not kept pace either, with just 23% of leaders saying governance and compliance are currently ready to support AI adoption. Only 33% have fully implemented clear policies on which decisions AI is prohibited from making on its own, and just 41% say they are confident in the guardrails their organisation has put in place. As companies face mounting AI costs, the need for proper governance frameworks becomes increasingly critical for managing both risk and expenditure.
As asked about their top priorities for reducing risk, 42% named upgrading IT infrastructure and 39% named strengthening cybersecurity, both ahead of any AI-specific initiative. Three in four organisations are investing in AI for cybersecurity purposes, making that the single most common use of AI. The report identifies a small group of organisations, 13% of surveyed companies, that are 32 percentage points less likely than laggards to have innovation delayed by basic technology problems. These pacesetters are more likely to be upgrading infrastructure and implementing comprehensive workforce readiness programs, suggesting that companies with their basics in order are also getting more out of AI investments. Companies are now implementing token budgets at the team and individual level, with examples ranging from $200 per week at Tesla to $1,500 per employee per month at Uber, while maintaining manager approval workarounds to prevent AI access removal.
