Enterprise AI adoption is shifting from experimental spending to a strict focus on return on investment (ROI). With inference costs rising, companies are prioritizing smaller, cost-efficient models over raw power, directly impacting how IT services firms like TCS, Infosys, and HCLTech structure their client contracts.
The initial wave of corporate excitement surrounding artificial intelligence is undergoing a significant reality check. As of late 2026, enterprise leaders are moving away from the "experimentation at any cost" phase, pushing instead for strategies that prioritize financial accountability and measurable business outcomes.
The primary driver for this shift is the ballooning cost of AI operations. While early phases focused on training models, the current challenge lies in the rising expense of inference—the process of running AI models to perform tasks. As companies deploy agentic workflows that require constant data processing, token consumption and infrastructure costs have surged. Reports indicate that these expenses are frequently outpacing initial budget projections, forcing CFOs to demand a clearer link between AI spending and corporate profits.
Evolving AI Architectures
To balance performance with budget, IT services majors are adopting tiered strategies. Instead of relying solely on the most powerful, expensive Large Language Models (LLMs), firms are increasingly deploying heterogeneous model ecosystems. This approach involves mixing high-power LLMs with smaller, more specialized Small Language Models (SLMs). By delegating routine tasks to efficient models and reserving expensive processing power for high-stakes decisions, enterprises can maintain quality while keeping costs in check.
This shift is also changing the competitive landscape among IT service providers. Companies like Tata Consultancy Services (TCS) and HCLTech have pursued aggressive investments in physical infrastructure, such as proprietary data centers, to provide secure, stable environments for their clients. In contrast, Infosys has maintained a more asset-light stance, focusing on platform-based automation—such as its Topaz Fabric—to manage model selection and cost-efficiency without heavy capital spending on hardware.
The ROI Reality Gap
For investors, the critical question is whether these AI investments are actually translating into bottom-line growth. Recent industry data from 2026 highlights a persistent gap in this area: while nearly 90% of organizations have adopted AI, only about 37% have reported a positive impact on their earnings before interest and taxes (EBIT). This disconnect puts pressure on IT services firms to prove that their AI solutions deliver productivity gains that directly increase profits.
As a result, contracts are changing. IT firms are moving away from traditional models based on hours worked and are increasingly tying their fees to specific performance outcomes and productivity metrics. While this can provide long-term stability for service providers, it also introduces execution risk. If an AI project fails to deliver the promised efficiency, it may lead to contract disputes or revenue pressure.
Risks and Investor Monitorables
Investors should keep an eye on how these shifting strategies impact company margins. The risk of uncontrolled infrastructure costs—combined with energy and hardware constraints—could erode profitability if not managed through effective governance. Furthermore, the lack of precise tracking for AI-related costs across many large organizations creates a risk of budget overruns, which could lead to project cancellations or delays.
The next important monitorable will be the translation of these productivity gains into quarterly financial results. Investors may look for management commentary on how outcome-based pricing models affect long-term profit margins and whether the divergence in capital strategy—between heavy infrastructure investment and asset-light approaches—provides a sustainable business advantage for the respective IT firms.