Fears that the artificial intelligence industry is headed for a brutal correction are premature, according to new research from Barclays.
The bank argues that far from peaking, capital expenditure across the sector is poised for another leg higher, fueled by stronger-than-expected financial performance at OpenAI and the prospect of major technological breakthroughs late in the decade.
The findings counter rising concerns that investors, cloud-computing providers and Big Tech companies may soon pull back after several years of frenzied spending on models, data centers and semiconductors. Barclays analysts say the evidence instead points to sustained—and in some areas accelerating—demand for computing power.
A key driver of the bank’s optimism is OpenAI’s latest internal revenue outlook. Chief Executive Officer Sam Altman recently told investors the company’s 2025 revenue is now projected to come in about 15% above its mid-year forecast.
Expectations for 2027 revenue have also been lifted sharply, from US$60 billion to roughly US$90 billion. OpenAI now aims to reach US$100 billion in annual recurring revenue by 2027—one year earlier than previously planned.
The upgraded revenue projections have pushed up other core metrics, including inference costs, weekly active users and annual average revenue per paying customer. Together, they signal accelerating commercial adoption of AI tools and reduce the likelihood that the industry is nearing a speculative bubble, Barclays said.
Behind the numbers lies OpenAI’s surging appetite for computing resources. The company’s budget for compute from 2024 through 2030 is expected to exceed US$450 billion, with a peak of roughly US$110 billion in 2028.
To secure long-term capacity, OpenAI has signed 10-year infrastructure agreements worth about US$650 billion in total with major partners including Microsoft and Oracle. These deals support OpenAI’s expansion while forcing its partners to undertake massive capital expenditure programs of their own.
OpenAI’s next-generation products are driving the trend further. The company is developing its GPT-6 large language model and Sora 3 video-generation model, but the biggest catalyst could come from an anticipated breakthrough in “recursive self-improvement” around 2027–2028. The technology, which would allow AI systems to autonomously upgrade their capabilities, could dramatically increase compute requirements. OpenAI has set aside roughly US$43 billion in monetizable compute to prepare for the surge.
The company’s rapid expansion has also intensified pressure on competitors. Google and Meta are accelerating model iteration and boosting compute budgets to narrow the gap. Google has increased the computational resources allocated to its Gemini models by 30%, while Meta plans to double its AI data-center footprint by 2026. Their spending is cascading across the technology supply chain, lifting demand for advanced semiconductors and related infrastructure.
Barclays estimates that global AI data-center capacity will double by 2030, with OpenAI alone requiring more than US$600 billion in partner-driven expenditures for cluster construction and upgrades.
With the industry preparing for breakthroughs later in the decade, the current capital-spending boom is likely to persist—and potentially accelerate—as the race to dominate the next era of AI intensifies.
