AI tools, models, agents, and unification became much harder to control in 2026. It is the uncontrolled AI expansion in an organisation, and eventually, the process leads to lower revenue. The spending on AI-native applications rose 108% by 2025. Companies and investors are facing a big reality check from Artificial Intelligence (AI). However, businesses are finding out that making money from AI is much harder than they thought. Here are a few breakdowns of the reasons for such big losses a company bears every year;
Adoption of AI in enterprises - AI corporate projects
Many businesses utilise AI tools and launch AI programs to attract investors, as it tends to have higher potential for growth in the long term. However, they repeatedly end up facing diseconomies of scale due to poor workflow and excess usage of AI tools even for the smallest task. Companies are losing thousands a month to "shadow AI"—individual employees signing up for point solutions on corporate cards without going through official procurement or security checks, and forgetting them.
Furthermore, teams are paying premium standalone prices for basic features that are now built natively into tools they already own, like Slack, Notion, or Google Workspace. Up to 70% of purchased corporate software licenses go unused because tools are deployed without a clear workflow or team onboarding. The average organisation designate $1.2M to tools like Claude.ai, Perplexity, and the OpenAI API. The variance in company usage of AI is still widening. Finance and procurement teams are hunting down and cancelling unapproved AI tools to save cash.
Less than 1% of executives report ROI of 20% or greater from AI, and 53% report only 1–5% ROI. Around 60% of organisations see minimal or no value, and roughly 30% of generative AI projects are being abandoned after proof-of-concept. About 30% of AI projects are stopped after testing because they don’t work well enough and indicate high risk of failure.
B2B Tech Procurement and Corporate Operations levelled their gun at business owners, team leads, or IT managers who are watching their software budgets bleed out. AI tends to save hours of work, while many gimmicky AI feature-The 5 second traps barely do the work. The high-visibility features that deliver near-zero operational ROI. License management- tracking and controlling the software license is important for enterprises as it prevents unnecessary expenses.
Workers are forced to bounce between 10 different AI tools that break their focus and drain energy, leading to low productivity. In addition, when a team exploit separate tools, company knowledge gets trapped in various places, which usually leads to data leaks and other privacy issues. Training employees to use AI properly takes a lot of time. On the other hand, businesses have to spend money on training staff to audit the AI's work for mistakes. 180k annual waste in unused software seats in medium size corporate company.
Data centres are being built. Building AI needs big data centres that use a lot of electricity, and these centres cost a lot to run. But the money companies make from AI services is not sufficient to pay the bills yet. Usage-based pricing is starting to emerge because flat-fee models are losing money — if a developer uses too many tokens, the company loses money. This shows the current business model for AI is not perfect and needs changes. Monetising AI applications at scale remains challenging; many AI-powered services are offered for free or at low cost to build user bases.
Average organisations now spend $55M on SaaS, an 8% increase annually. 78% of IT managers reported unexpected charges, while 61% were forced to cut projects due to the surprise cost. Running AI requires massive computer power, which creates huge electricity and technology bills.
Physical bottlenecks
The most significant barriers to AI acceleration are thermodynamic — relating to power generation, grid capacity, and the physics of getting electricity where it needs to go. This is limiting how fast AI infrastructure can actually scale. Expanding data centres requires upgraded power grids, reliable energy supply, and managing thermal transfer. AI servers are not easy because the power system cannot keep up quickly.
High interest rate means investor low profitability and preventing them to put the high amount of real money into stocks, causing AI stock prices to drop. Investors no longer care about big promises and want clear evidence of their earnings. A recent poll shows that more than half of Americans worry about AI bringing out more harm than good by taking over jobs, privacy issues, and ethical problems, which could influence regulatory and adoption trends. Plus, it could slow down the adoption of AI technologies. The initial excitement is slowing down because the actual financial benefits are small compared to the high costs.
The shadow IT problem could be overcome by establishing a clear policy for AI tool usage, educating employees, and monitoring the shadow IT application. They use shadow IT because it is convenient but may not consider the implications of using these systems.
References: