A new report from Citi describes the AI industry's demand as a "vertical wall," with revenue at major labs exploding far faster than the supply of computing power can keep up.
A new report from Citi describes the AI industry's demand as a "vertical wall," with revenue at major labs exploding far faster than the supply of computing power can keep up.

Artificial intelligence labs are facing a "vertical wall of demand" that is sending revenues soaring, with Anthropic expecting its second-quarter revenue to jump 130% to $10.9 billion, according to a new Citi report. The analysis shows enterprise adoption is creating an unprecedented demand surge that is colliding with supply-side bottlenecks for computing power and talent.
"This vertical wall of demand, a phrase first used by OpenAI CFO Sarah Friar, is now borne out by the data," the Citi report, titled "Inference Outlook," said on May 25. "The demand is hitting a 'slanted' supply curve that is growing slowly, creating an absolute seller's market for infrastructure."
Anthropic's revenue is projected to grow from $4.8 billion in the first quarter to $10.9 billion in the second, putting it on track for an annual recurring revenue of approximately $50 billion. For comparison, OpenAI reported $5.7 billion in revenue for the first quarter of 2026 alone, while Google's token processing has increased sevenfold year-over-year to 3.2 quintillion per month.
The imbalance is forcing a strategic shift, as rising compute costs and a scarcity of top talent become the primary constraints on growth. In response, AI leaders like OpenAI are beginning to lock customers into one- to three-year contracts for guaranteed capacity, fundamentally changing the economics of AI and signaling a period of intense capital investment and pricing power for infrastructure providers.
The core issue facing the industry is that while demand for AI processing is growing exponentially, the supply of the necessary components is expanding on a much slower, more linear track. The primary constraints are the availability of advanced semiconductors, particularly from Nvidia; the physical data center capacity to house them; and the massive amounts of electricity required to power them. This infrastructure is no longer a supporting input but the core cost structure of the AI business.
This scarcity extends to human capital. The report notes that the most severe bottleneck in delivering frontier AI capabilities remains the small pool of top-tier research talent, highlighted by recent high-profile hires like Andrej Karpathy's move to Anthropic.
AI companies are actively leveraging this supply-demand imbalance to create new pricing structures that maximize revenue. OpenAI has introduced a "Guaranteed Capacity" model, allowing customers to sign one- to three-year contracts to reserve computing power at a discount, a model more akin to cloud infrastructure providers than traditional software.
At the same time, providers are implementing aggressive pricing stratification. Google recently introduced a $100 per month "prosumer" tier for its top-tier Ultra AI, while new flagship models are being priced at a significant premium. According to Citi's analysis, OpenAI's new GPT-5.5 model costs between 49 percent and 92 percent more per workload than its predecessor. This trend of rising costs for top-tier models is already beginning to impact enterprise adoption decisions at the margin.
This closed-source, high-price strategy is not without challengers. Chinese AI lab DeepSeek, which has focused on a research-first, open-source approach, recently released its V4 Flash model at a price point reportedly up to 100 times cheaper than comparable Western models. This puts long-term pressure on the pricing power of the industry leaders.
While OpenAI, valued at approximately $852 billion, and Anthropic, which is targeting a $900 billion IPO, are capturing massive revenues, their costs are scaling directly with growth. The key question for investors is whether the immense productivity gains from AI will be captured by the AI platforms themselves, their enterprise customers, or the foundational infrastructure providers like Nvidia that supply the picks and shovels for the gold rush.
This article is for informational purposes only and does not constitute investment advice.