Intel to mass-produce 1.4-nanometer chips in 2028, posts biggest sales growth in 15 years

Intel posted an earnings surprise as surging AI data center investment by big tech companies drove its highest quarterly sales growth in 15 years. The company has won wide industry praise after lifting yields on its 18A (1.8-nanometer) process, and it is now pushing into advanced packaging and developing a memory solution to challenge South Korea's dominance in the HBM market.

Intel said Thursday that second-quarter sales reached $16.13 billion, up 25.4 percent from a year earlier, topping the market consensus of $14.42 billion. Operating profit came in at $1.79 billion, swinging back to the black.

Adjusted earnings per share were $0.42, double the market forecast of $0.21. Adjusted gross margin improved sharply to 41.8 percent from 2.5 percent in the same period last year. The quarterly sales growth rate was the highest since the third quarter of 2011.

The recovery was led by CPUs for AI data centers. Intel's data center and AI segment posted sales of $6.26 billion, exceeding the market estimate of $5.37 billion. The shift from AI training to agentic AI inference drove a sharp increase in CPU demand.

Foundry sales rose 31 percent year-on-year to $5.8 billion. Intel began mass-producing the "Panther Lake" CPU and "Xeon 6+ (Clearwater Forest)" on its 18A process last October, becoming the first in the world to do so — ahead of both Samsung Electronics and TSMC.

"During the second quarter, we exceeded our internal output targets across all factories using Intel 7, Intel 3 and Intel 18A, driven by yield improvements, cycle time reductions and increased wafer start volumes," CEO Lip-Bu Tan said. "18A output in particular showed meaningful growth during the quarter, and yields continue to exceed expectations." He added that the company is ramping production of new 18A-based products.

Tan had previously suggested Intel might halt 14A development if it failed to secure major customers, but external customers have since come forward, prompting the company to accelerate the node's rollout. "The 14A (1.4-nanometer) process is outpacing 18A's development trajectory on both defect density and transistor performance," he said. "PDK 0.5 is already complete, and PDK 0.9 is on track for an October release."

Tan said Intel plans to begin 14A pilot production in the second half of 2027 and move into full mass production in 2028. "As customers begin to see yields on the 0.9 PDK, they are starting to show interest in which products they want to apply this technology to and how much production capacity we can provide," he said.

Intel's roadmap closely mirrors TSMC's. TSMC plans to begin mass production on its A16 (1.6-nanometer) process in the fourth quarter, followed by A14 (1.4-nanometer) in 2028 and A13 and A12 processes sequentially in 2029. Samsung Electronics, focused on stabilizing its 2-nanometer process, has said it will skip 1.8-nanometer and target 1.4-nanometer mass production around 2029 — roughly a year behind Intel.

Intel also highlighted plans to expand its packaging business during the earnings call. The company had previously unveiled EMIB-T, its proprietary 2.5D packaging technology positioned as an alternative to TSMC's packaging bottleneck. The technology is expected to become a new growth engine for Intel's foundry operations.

"Customer interest in packaging is very high, our order backlog continues to grow, and yields and reliability are meeting targets," Tan said. "We are focused on scaling this technology into high-quality, high-volume production to support customer ramp in 2027."

Intel is also developing a memory solution to rival HBM. The company has been researching Z-Angle Memory (ZAM) technology with SiMemory, a subsidiary of Japan's SoftBank, in a bid to crack South Korea's grip on the HBM market.

"Memory is a key bottleneck in building AI infrastructure and plays a critical role in delivering exactly the capabilities customers need," Tan said. "We recently brought on Executive Vice President Lee Suk-hee, the former CEO of SK hynix. We are researching how to integrate computing and memory, and how to further improve memory utilization through memory stacking."