This is not Internet Bubble 2.0! Citigroup sharply raises its AI capital expenditure forecast, saying that AI infrastructure deployment is "accelerating rapidly."
Citi Group’s latest research suggests that investment and deployment in artificial intelligence infrastructure is “accelerating rapidly” at a pace far exceeding expectations. This is not a repeat of the internet bubble of 2000—the key difference lies in the real, enterprise-level demand providing an “outlet” for this wave of investment, forming a closed loop of value validation.
According to news from Chase the Wind Trading Desk, based on this judgment, Citi analysts have significantly raised their forecasts for AI capital expenditure by tech giants: The AI capital expenditure forecast for 2026 has been raised from $420 billion to $490 billion. Meanwhile, the accumulated capital expenditure forecast up to 2029 has also been lifted from $2.3 trillion to $2.8 trillion.
Citi points out that the revised growth expectation for capital expenditure in 2026 is 24%, significantly higher than the current 20% market consensus. The report expects major tech giants will reflect this incremental spending in their Q3 earnings guidance. The entire AI infrastructure industry chain—including semiconductors, hardware, and other infrastructure suppliers—will benefit from this wave of accelerated investment.
Meanwhile, Citi emphasizes the essential difference between this wave of AI investment and the internet bubble. The rapid growth in current enterprise demand for AI services provides “clear external validation” for investment, creating a crucial “outlet” for the investment cycle. Citi estimates that by 2030, global AI computing power demand will add 55GW of electricity capacity worldwide, bringing $2.8 trillion in incremental AI computing spending.
Frequent High-profile Partnerships, Investment Frenzy Accelerates
Citi’s report notes that in recent weeks, moves by industry giants underscore that the scale and speed of AI infrastructure construction are rising. Since Oracle disclosed a $300 billion deal with OpenAI in early September, the market has witnessed a series of major collaborations, including:
NVIDIA and OpenAI announcing a $100 billion partnership to deploy 10 gigawatts (GW) of NVIDIA systems;CoreWeave’s agreement with OpenAI grew from an initial $12 billion to $22.4 billion;Microsoft’s “Stargate” project in partnership with OpenAI continues to expand;Alibaba expects its data center capacity to grow tenfold and is fully integrating its PAI software stack with NVIDIA.
Citi believes these developments clearly show that the pace of infrastructure investment is dramatically accelerating to support enterprise AI service demand and AI labs’ training capacity.
AI Infrastructure Demand Exhibits Exponential Growth
Citi’s report shows that AI infrastructure construction is entering an acceleration phase. According to Epoch AI data, since 2010, frontier model training computing power has grown 4.6 times annually, more than double the speed of Moore’s Law. Although computing performance and density of GPU clusters continue to improve, hardware and algorithmic efficiency improvements haven’t kept pace, resulting in the hardware costs of leading AI supercomputers increasing exponentially (1.9 times annually), while training power consumption is also surging rapidly (2.1 times annually).
Citi believes the acceleration of infrastructure investment is essentially due to terminal customer demand. This trend is already reflected in the latest earnings reports of tech giants, whose backlogged orders show strong demand. Citi expects that, as companies like AthenaHealth, Hitachi, Eli Lilly, and Wolters Kluwer move their AI projects from concept validation to actual production deployment, this trend will be further confirmed in Q3 earnings and guidance.
In addition, based on estimates in the report, global demand for AI computing power will add 55 GW of electricity capacity by 2030, translating into up to $2.8 trillion of incremental AI compute spending, with the US market accounting for $1.4 trillion.
Not a Repeat of the Internet Bubble
Faced with such frenzied investment, the market is widely concerned whether this will repeat the internet bubble of the early 21st century, especially the phenomenon of funds “circulating” among giants, such as NVIDIA investing in OpenAI, while OpenAI is a major customer of NVIDIA and Oracle.
Citi’s report directly responds to these doubts, arguing that there are “key differences” between the current AI investment frenzy and those times. The report acknowledges that the vendor financing model is similar to deals circa 2000 between internet startups and network equipment firms like Nortel and Cisco.
However, the report emphasizes, this round of AI investment has a clear “off-ramp”—the constantly growing external demand driven by enterprise adoption of AI services. The report points out that many internet companies of that era spent vast sums on marketing to fight for a “customer that ultimately did not exist,” while today’s AI companies enjoy “clear and reliable levels of demand.” Actual application gains in areas such as knowledge retrieval, customer service, and healthcare provide “clear external value validation” for this investment wave.
Even so, the report cautions that while companies see AI gains, they are also attentive to the risks brought by large-scale adoption.
~~~~~~~~~~~~~~~~~~~~~~~~
The above wonderful content is from Chase the Wind Trading Desk.
For more in-depth analysis, including live commentary and frontline research, please join [Chase the Wind Trading Desk · Annual Membership]
Risk Disclaimer and Exemption ClauseThe market carries risks, and investment should be approached with caution. This article does not constitute personal investment advice, nor does it take into account the individual investment objectives, financial conditions, or needs of any user. Users should consider whether any opinions, views, or conclusions in this article are suitable for their unique situation. Investing based on this is at your own risk.