Artificial intelligence may not remain a cloud first business for long. Enterprises are increasingly expected to move AI workloads from public cloud platforms to their own infrastructure as concerns over data privacy, security and costs intensify, according to a recent report by Jefferies.
The investment bank said the launch of China’s GLM 5.2 model marks another turning point in the AI race. The report noted that the model delivers performance close to leading western models while costing only about one quarter as much per token. As AI inference becomes cheaper, businesses are expected to deploy AI applications more widely.
However, rather than strengthening the dominance of public cloud providers, Jefferies believes falling AI costs could encourage enterprises to run AI models inside their own data centres.
“GLM 5.2 proves enterprises no longer have to sacrifice intelligence for privacy. We are seeing a massive acceleration in companies pulling their AI workloads out of the public cloud and back onto local corporate servers,” the report said.
Do follow us on X @VainetraAi
According to Jefferies, lower token costs make it economically viable for companies to operate smaller language models on premise, allowing them to keep sensitive corporate information within their own infrastructure. This shift could significantly increase demand for enterprise servers, AI accelerators, storage systems and private data centre capacity.
The report also highlighted growing cybersecurity risks associated with AI adoption. It referred to a recent warning from the Five Eyes Alliance, which cautioned that advances in frontier AI models could dramatically accelerate cyberattacks within months rather than years. Such risks are expected to strengthen the case for organisations to retain greater control over their AI infrastructure.
Jefferies further argued that declining AI costs should not be viewed as negative for infrastructure spending. Instead, applying the economic principle known as the Jevons Paradox, the report said cheaper AI services are likely to increase overall AI usage, resulting in higher demand for computing power, memory chips and data centre infrastructure.
If this trend gathers pace, enterprise AI infrastructure could emerge as one of the biggest beneficiaries of the next phase of the global AI boom, complementing continued investments by hyperscale cloud providers while opening new growth opportunities for private data centres and hardware vendors.
Source: Jefferies, GREED & Fear: The Mother of All Cycles (25 June 2026)
