Summary of Key Points
Over the past three years, the AI industry has been dominated by the belief that only GPUs were essential for high-performance computing. However, with the rise of agents capable of independently completing complex tasks, the role of CPUs has evolved from a mere "auxiliary tool" to a "key coordinator." CPUs are now responsible for splitting tasks, invoking other tools, and coordinating sub-agents, becoming a new bottleneck in AI computing power. The three major chip manufacturers—NVIDIA, AMD, and Intel—are all increasing their investment in the CPU market, leading to supply shortages: longer delivery times for server CPUs, rising prices, which could eventually be passed on to consumers, potentially making the next computer more expensive.
Detailed Analysis
#### 1. The Era of Agents: CPUs Moving from Supporting Roles to Central Players
Previously, AI was primarily used for model training (such as with ChatGPT), where GPUs excelled at parallel computing (processing large amounts of data simultaneously). CPUs were relegated to performing miscellaneous tasks like system initialization, data transfer, and task management. However, agents are different. They act like super assistants, breaking down complex tasks into multiple steps (for example, booking a flight involves checking flights, comparing prices, filling out forms, and confirming payments), as well as invoking external tools (searching, running code, and accessing databases) and coordinating the work of dozens of sub-agents in parallel. These coordination and scheduling tasks cannot be handled by GPUs; they rely entirely on CPUs. Studies show that CPU processing accounts for 50%-90% of the total waiting time in agent-based systems. The ratio of CPUs to GPUs in AI servers has shifted from 1:8 (one CPU per eight GPUs) to 1:4, and it may soon become 1:1, indicating that CPUs are becoming more central in AI computing.
#### 2. The Three Giants Competing for Market Share
- NVIDIA: Launched its own CPU, "Vera," with the aim of not engaging in price wars with AMD and Intel but rather ensuring that external CPUs do not hinder GPU performance. Vera is optimized for coordinating agent activities, improving GPU efficiency and potentially opening up a new market worth billions of dollars. Vera has already been supplied to companies like OpenAI and Anthropic.
- AMD: Focusing on the "Venice" processor, the industry's first 2-nanometer mass-produced high-performance chip, with plans for the "Verano" processor to address the large memory demands of agents. AMD's data center business revenue increased by 39% in the fourth quarter of 2025, reaching a record $5.4 billion.
- Intel: Thanks to the success of its "Xeon 6" and "Core 3" processors, Intel saw a 156% increase in net profit in the first quarter of 2026. However, production capacity still falls short of demand, with the CEO stating that growth will continue into next year.
#### 3. Why is There a Sudden Shortage of CPUs?
- Surging Demand: The new demands driven by agents have pushed the annual growth rate of the CPU market from single digits to over 35%.
- Chip Manufacturing Capacity Competition: Chip manufacturers (such as TSMC) prioritize higher-profitting GPU orders, reducing CPU production quotas.
- Purchasing Havoc: Global memory shortages have led customers to stockpile CPUs in anticipation of price increases, further exacerbating the supply shortage. As a result, Intel's server CPU delivery times have reached up to six months, and some AMD products are waiting 8-10 weeks to be delivered. Intel's server CPU prices in the Chinese market have increased by 10% (China accounts for 20% of its revenue).
#### 4. The Impact on Consumers
The majority of CPU production capacity is being allocated to servers, leaving less for consumer products (personal computers and small businesses). Suppliers prioritize supplying higher-bidding customers (such as cloud providers), forcing consumers to wait longer and facing higher prices. In short, the more popular AI becomes, the higher the cost of CPUs in new computers, as demand outpaces supply.
#### 5. The "Barrel Effect" of AI Computing Power
AI computing power is no longer solely dependent on GPUs; other factors such as CPUs, memory, high-bandwidth networks, and efficient cooling systems are also critical bottlenecks. Just like a barrel that can hold only as much water as its shortest plank determines, the efficiency of an entire system depends on all components. The return of CPUs to a central role in AI computing is a testament to this principle—achieving true AGI (Artificial General Intelligence) requires a robust infrastructure across all aspects, not just powerful individual chips.
In essence, this news highlights that the AI industry has come full circle and returned to the realization that foundational hardware is essential. The resurgence of CPUs is a clear indication that advanced AI technologies cannot be sustained without a solid computational foundation.