Agent opens up a new battlefield for computing power. The CPU track now has three pillars: red, green and blue
The three semiconductor manufacturers Intel, AMD, and NVIDIA are each gathering heavy forces in the new computing battlefield of CPU, and the three-legged "red, green and blue" (nicknames given by consumers based on the colors of their trademarks) emerging. On July 23, local time, Intel announced its second-quarter financial report as scheduled, achieving revenue of US$16.1 billion. The Data Center and AI Division (DCAI) has become a well-deserved growth engine - achieving revenue of US$6.3 billion, a year-on-year increase of 59%, far exceeding the company's overall growth rate, and exceeding the previous analyst expectations of 5.54 billion yuan.
The three semiconductor manufacturers Intel, AMD, and NVIDIA are each gathering heavy forces in the new computing battlefield of CPU, and the three-legged "red, green and blue" (nicknames given by consumers based on the colors of their trademarks) emerging. On July 23, local time, Intel announced its second-quarter financial report as scheduled, achieving revenue of US$16.1 billion. The Data Center and AI Division (DCAI) has become a well-deserved growth engine - achieving revenue of US$6.3 billion, a year-on-year increase of 59%, far exceeding the company's overall growth rate, and exceeding the previous analyst expectations of 5.54 billion yuan. The core driving force behind the soaring performance of this department is undoubtedly the CPU. Talking about the reasons for performance growth, Intel said that the outbreak of agent AI (Agentic AI) has led to the recovery of server CPU demand. The company has signed a number of long-term server CPU procurement contracts, with a term of 3 to 5 years, and some include volume and price commitments. CEO Chen Liwu said: "In the data center field, CPU demand is taking off, and demand is exceeding our growing supply capacity." Huge demand has given Intel stronger pricing power, with Chief Financial Officer David Zinsner noting: "From a pricing perspective, things are better than we expected." At the beginning of this month, Intel confirmed that it would raise the prices of some consumer-grade and server CPUs. Consumer-grade processors will increase by between 30 and 50 US dollars, and data center-level products will increase by hundreds or even thousands of dollars. In response to the Science and Technology Innovation Board Daily, it stated that the price adjustment was based on regular monitoring of the supply chain and related costs. Investment bank Wedbush believes that Intel still has the ability to implement price increases without affecting market demand. "Given the continued shortage of server CPU supplies, we believe Intel still has room to raise prices without impacting demand," analyst Matt Bryson wrote in a note to clients. In the past, GPU was almost synonymous with computing power. With the gradual rise of agent-based AI, the cost per token has become extremely critical. CPUs, which are good at planning tasks, have taken over from GPUs, becoming a new battlefield in the computing power arms race. Intel is not the only player in this wave. Recently, Nvidia disclosed a number of measured data on the Vera Rubin platform, including the core component Vera CPU. NVIDIA emphasizes that Vera CPU is designed and built for the era of intelligent agents, and says that its customized Olympus core achieves a 2x improvement in single-thread performance, a 3x increase in inter-core bandwidth, and a 40% reduction in memory latency compared to chiplet designs from competitors such as AMD. However, just two days later, AMD took a strong counterattack at the annual flagship AI conference "Advancing AI 2026": It compared the latest sixth-generation server CPU "Venice" with the Vera CPU, claiming that the throughput performance of "Venice" was 2.2 times that of Vera, and the per-core performance was 20% ahead in the out-of-box state without performance tuning. AMD predicts that its traditional data center CPU market will reach US$220 billion by 2030, and the total market size it competes in will reach US$2 trillion. Regarding the CPU dispute between the "red and green factories", Bank of America analyst Vivek Arya pointed out that the issue that investors are most concerned about is whether artificial intelligence requires a higher single-core frequency or a larger number of cores. NVIDIA and AMD represent these two different technical solutions: NVIDIA advocates reducing the delay of tool calls and task switching; AMD pursues concurrency and full rack throughput first, and plans to stuff more cores and threads. Different from the technology dispute between NVIDIA and AMD, "blue factory" Intel announced that it will increase its capital expenditure plan for 2026 from US$18 billion to US$20 billion. Specifically, it will spend approximately US$5.7 billion to expand the manufacturing capacity of Xeon 6 and next-generation Xeon processors (server CPUs) based on the Intel 3 process. Under many variables, the CPU market may move from the current "one super, many strong" situation to the future "three-point world" situation.
The above-mentioned organizations predict that by 2030, commercial ARM CPUs launched by NVIDIA and other companies will account for 36% of server CPU revenue. At the same time, Intel's market share is expected to decline from 41% to 24%, while AMD's market share will remain basically stable, between 25% and 27%. From an industry perspective, all parties are optimistic about future CPU demand. Joint research between Georgia Tech and Intel shows that in five Agent task tests, the CPU running time accounted for more than half of the task; TrendForce predicts that the ratio of CPU to GPU in the Agent era will change from 1:4 to 1:8 to 1:1 to 1:2. CITIC Securities stated that the 1:1 to 1:2 ratio of CPU to GPU is still just the starting point, not the end point: under optimistic circumstances, the CPU to GPU ratio will reach 8:1, and the forward market size is expected to reach 2 trillion US dollars, which is higher than the current GPU market size. Against this background, 2026 will also be the year of accelerated penetration of domestic CPUs. It is expected that the penetration rate of domestic CPUs in the domestic commercial market will rapidly increase from single digits to double digits.
The above-mentioned organizations predict that by 2030, commercial ARM CPUs launched by NVIDIA and other companies will account for 36% of server CPU revenue. At the same time, Intel's market share is expected to decline from 41% to 24%, while AMD's market share will remain basically stable, between 25% and 27%. From an industry perspective, all parties are optimistic about future CPU demand. Joint research between Georgia Tech and Intel shows that in five Agent task tests, the CPU running time accounted for more than half of the task; TrendForce predicts that the ratio of CPU to GPU in the Agent era will change from 1:4 to 1:8 to 1:1 to 1:2. CITIC Securities stated that the 1:1 to 1:2 ratio of CPU to GPU is still just the starting point, not the end point: under optimistic circumstances, the CPU to GPU ratio will reach 8:1, and the forward market size is expected to reach 2 trillion US dollars, which is higher than the current GPU market size. Against this background, 2026 will also be the year of accelerated penetration of domestic CPUs. It is expected that the penetration rate of domestic CPUs in the domestic commercial market will rapidly increase from single digits to double digits.