The tech sector is in focus. The fears of AI-led disruption have taken a new dimension. Anthropic proposed to slow the pace of AI capability development, highlighting that AI’s improving capabilities could outpace the development of safety measures needed to control increasingly capable AI systems. However, Kotak Institutional Equities said that even though developments are new, they follow same old patterns and create “a double-edged sword”.
“It can reduce the viability of new cases, offsetting the tailwind of lower deflation. It can also have an opposite effect, i.e., the increase in deflation if the capability slowdown is more pronounced for non-software-related tasks compared with software engineering,” Kotak noted.
The rapid adoption of AI could have a net deflationary impact of around 3-3.5% annually on IT services revenue growth between FY27-FY29, according to Kotak.
Kotak believes that faster improvements in AI capabilities, particularly in software engineering, could put greater pressure on the traditional IT services business. However, stronger AI capabilities could create new opportunities as enterprises adopt AI across different business functions.
AI adoption creates both risks and opportunities
The fresh debate of AI impacting IT services arises due to the increase in cheaper but capable AI models, particularly those developed by Chinese AI labs growing at a faster speed.
“Better affordability of AI through the use of cheaper yet capable models such as from Chinese labs will serve to both increase deflation in the base business and increase new demand opportunities,” Kotak noted.
Safety Standards & Recursive AI: How Anthropic’s Proposal Impacts Services
Anthropic’s proposal does not call for a complete pause in AI development. Instead, Kotak highlighted, it proposes that the pace of capability improvements should be limited by the ability to ensure safe AI usage.
Anthropic has said AI development has accelerated in recent months because AI systems are increasingly capable of helping develop the next generation of AI models, called ‘recursive self-improvement (RSI)’.
Anthropic also calls for third-party safety evaluations of AI models and government-supported coordination among frontier AI labs. Leaders from OpenAI, Grok and Google have supported the proposal, while the US and Chinese governments, Nvidia and Meta have seemingly not agreed to it.
Kotak sees low probability of major AI slowdown
Kotak said it assigns a low probability to a significant slowdown in AI capability development.
According to the brokerage, such a slowdown would require coordination across AI labs globally. It also said Chinese AI labs have limited incentives to slow down because they are still behind US labs on the AI capability curve.
US frontier AI labs, meanwhile, have incentives to continue developing their models as long as they can manage safety concerns and secure funding.
A slowdown could also happen if AI labs struggle to secure enough funding to train new-generation models. Otherwise, Kotak expects frontier AI labs to remain focused on improving AI capabilities.
New Growth Vectors: AI Safety Audits and Enterprise Monitoring Work
Greater focus on AI safety could also create new opportunities for IT services companies.
Kotak said enterprises will need systems for monitoring AI agents, maintaining audit trails, protecting data privacy, ensuring AI systems follow instructions and verifying AI-generated outputs.
These requirements could increase demand for IT services firms that can implement enterprise-grade AI systems.
The brokerage said enterprises are also becoming more focused on the return on investment from AI applications, rather than simply increasing AI usage.
“IT firms need to be nimble enough to adapt to such shifts and deliver enterprise-grade AI systems,” Kotak said.
Model Disruption: How DeepSeek V4.1 & Cheaper Chinese LLMs Squeeze Margins
Chinese AI labs are continuing to narrow the capability gap with leading AI models while offering them at significantly lower prices, according to the report.
Kotak highlighted DeepSeek’s V4.1-Flash model, which has 552 billion parameters and uses a mixture-of-experts architecture. The model uses 8 billion active parameters for input and 16 billion for output.
The brokerage said DeepSeek V4.1-Flash delivers performance similar to some frontier models on software-related benchmarks such as Terminal Bench 2.1, although it still trails on some newer benchmarks, including Terminal Bench 4.0.
The combination of relatively strong capabilities and lower costs could encourage greater experimentation with agentic AI, Kotak said.
This could have a dual impact on IT services companies by increasing pressure on their existing business while also creating demand for new AI-led services.
OpenAI continues to build anticipation around new model
Kotak also pointed to OpenAI’s ongoing development of a new AI model.
According to the report, OpenAI has indicated that the model helped solve an aspect of the Millennium Prize problem related to the Navier-Stokes equations.
OpenAI has also indicated that the new model has significantly higher capabilities in mathematical research than its predecessor and could deliver a “step function” improvement across several benchmarks.
Kotak noted that OpenAI had made similar claims about AI’s ability to assist with new mathematical research before the release of its latest Astra model.
AI slowdown could reduce some pressure on IT firms
A significant slowdown in AI capability development could reduce the deflationary pressure on traditional IT services businesses in the medium term, Kotak said.
For instance, IT companies could lower their expectations for future productivity improvements in multi-year fixed-price contracts. This could reduce the extent of price deflation when these contracts are renewed.
However, the brokerage said such a slowdown would be a “double-edged sword” because it could also reduce the viability of new AI use cases.
If AI capability improves more slowly in non-software areas than in software engineering, the impact could even be the opposite and increase deflationary pressure on IT services, Kotak said.