AI has its own Hormuz choke point now.
Introduction
Artificial intelligence is no longer limited by imagination alone. In 2026, the bigger constraint is becoming supply. The world wants more AI answers, more coding assistants, more video generation, more enterprise automation, and more AI agents. But all of this depends on a scarce resource: compute capacity.
Between January and March 2026, weekly token usage on OpenRouter reportedly quadrupled, showing how quickly demand for AI model usage is rising. At the same time, AI companies are facing shortages of chips, data-centre space, power equipment, memory, CPUs, and electricity infrastructure.
This supply crunch is changing the economics of AI. The winners may not only be the companies with the smartest models. They may be the companies that control the chips, data centres, power access, and cloud capacity needed to run AI at scale.
.png)
Let's dive deeper.
1. AI demand growing faster than infrastructure
The most visible sign of the crunch is token demand. Tokens are the small units of text processed by large language models. Every chatbot response, coding suggestion, AI search result, document summary, and agentic workflow consumes tokens.
In early 2026, demand rose sharply as more users adopted AI coding tools, research assistants, business automation tools, and autonomous agents. The OpenRouter data showing a fourfold rise in weekly token usage between January and March 2026 is a strong signal that AI consumption is accelerating faster than the industry expected.
This means AI companies are no longer only competing on model quality. They are competing on who has enough compute to serve users reliably.
2. Compute the New Oil of AI
For AI companies, compute is now a strategic resource. Without enough GPUs, CPUs, memory, networking hardware, and power, even the best AI model cannot serve millions of users.
OpenAI CFO Sarah Friar reportedly said in April 2026 that the company is skipping some opportunities because demand is outpacing available compute. Reports also suggest OpenAI has had to make difficult trade-offs, including pulling back resources from projects such as video-generation work to prioritize core AI products.
This shows an important shift. In earlier software businesses, scale was mainly about servers and distribution. In AI, scale is about massive physical infrastructure.
3. Usage limits becoming a business tool
The supply crunch is already changing how AI products are priced and rationed. Anthropic adjusted Claude usage limits in 2026 to manage growing demand. A company spokesperson said about 7% of users would hit session limits they would not previously have hit.
Anthropic also introduced a March 2026 promotion that doubled Claude usage during off-peak hours, outside peak weekday windows. This is important because it shows AI platforms are beginning to manage demand like airlines, electricity grids, or telecom networks.
In simple terms, when compute is scarce, time of use matters.
4. Hyperscalers spending at historic levels
The companies best positioned in this crunch are the large cloud and data-centre players. Amazon, Google, Meta, Microsoft, and Oracle are expected to spend around $750 billion in capital expenditure in 2026, according to reports cited in late April 2026.
Amazon alone has reportedly guided toward around $200 billion in 2026 spending, while Meta has indicated a range of roughly $115 billion to $135 billion.
This scale of spending makes one thing clear: the AI race is no longer just a software race. It is also a capital expenditure race.
5. Nvidia and TSMC control AI's Hormuz Strait
The biggest profits in the AI boom are flowing to the companies that control the hardest-to-replace parts of the supply chain.
Nvidia dominates advanced AI accelerator chips, while TSMC manufactures many of the world’s most advanced chips. These companies benefit because demand is high and alternatives are limited.
Nvidia’s gross margin has been reported at around 75%, while TSMC’s gross margin is above 60%, far higher than many traditional contract manufacturers. This pricing power shows how valuable choke points become when supply is scarce.
In 2026, the lesson is clear: the deeper you are in the AI infrastructure stack, the stronger your pricing power may be.
6. Custom Chips becoming a survival strategy
High AI hardware costs are pushing major software and cloud companies to design their own chips. Amazon has Trainium and Graviton, Google has TPUs, Microsoft has Maia, and Meta is investing in custom AI infrastructure.
Amazon is reportedly considering selling Trainium chips directly to customers within the next two years. Its AI chip strategy is tied to major cloud commitments from companies such as OpenAI, Anthropic, and Uber.
Custom chips can reduce dependence on Nvidia, but they are difficult to design, manufacture, and scale. Google’s TPU effort took more than a decade to mature. For most companies, replacing Nvidia is not easy. Replacing TSMC is even harder.
7. Cheap AI pricing may not last forever
So far, users have become used to falling AI prices. Inference costs have dropped sharply over the past year, and companies have used low subscription prices to attract users in markets such as India.
But this cheap pricing hides a difficult reality. AI firms are spending heavily to serve users, and many are expected to lose billions before becoming profitable. As OpenAI, Anthropic, and others move toward public-market expectations, they will need to show that their businesses can eventually make money.
That means future AI pricing may become more disciplined. Users may face higher prices, tighter usage caps, tiered access, or charges based on compute intensity.
8. Businesses need to use AI more efficiently
The final impact of the supply crunch will be behavioral. Today, many companies ask, “Are we using AI?” In the coming years, the better question will be, “Are we using AI efficiently?”
As AI becomes embedded in coding, customer support, finance, law, healthcare, education, marketing, and operations, demand for tokens may grow by orders of magnitude. But if compute remains costly, companies will need to optimize how they use AI.
They will need to choose the right model for the right task, avoid unnecessary long prompts, use smaller models where possible, cache repeated outputs, design efficient AI workflows, and monitor token consumption like a business cost.
The future of AI adoption will not only depend on access. It will depend on AI cost management.
Conclusion
The AI supply crunch of 2026 is a turning point. The industry is discovering that intelligence at scale is not only a matter of algorithms. It is also a matter of chips, power, memory, data centres, capital expenditure, and supply-chain control.
This will reshape the economics of AI. Model-makers will face pressure to become more efficient. Cloud companies will gain more influence. Chipmakers will enjoy enormous pricing power. Enterprises will need to use AI more carefully. Consumers may see more usage limits and premium pricing.
The age of unlimited cheap AI may not disappear, but it will become more complex. The next phase of AI will belong not only to those who build the smartest models, but to those who can secure, optimize, and afford the infrastructure required to run them.
[The Billion Hopes Research Team shares the latest AI updates for learning and awareness. Various sources are used. All copyrights acknowledged. This is not a professional, financial, personal or medical advice. Please consult domain experts before making decisions. Feedback welcome!]
