Intel's Data Center and AI group posted $6.26 billion in Q2 2026. Revenue rose 59 percent year on year. Sequential growth reached 24 percent.

The division recorded its highest server CPU growth in fifteen years, the strongest expansion Intel has reported since AMD's Epyc reshaped the commercial server market in 2017.

What is driving the demand?

Agentic AI sandboxes execute code rather than interpret tokens. That workload favors server CPUs with strong cores and high memory bandwidth.

Intel benefits despite competing against AMD's Epyc and homegrown Arm server chips used inside the largest hyperscalers and cloud builders worldwide.

Operating profit at the Data Center and AI group climbed 3.9 times to $2.47 billion. Margin reached 39.5 percent of revenue.

Intel had not posted that profitability level in five years. The segment remains below the 50 percent margins Intel booked before 2016.

How is the foundry side performing?

The Intel Foundry group generated $5.77 billion in Q2 2026. Revenue rose 30.5 percent year on year.

External foundry customers contributed just $293 million. The unit recorded a $2.09 billion operating loss, a 34 percent improvement from a year earlier.

Most foundry revenue still came from inside Intel. Desktop and server parts ramped in volume during the quarter.

Intel logged yield gains on Intel 4, Intel 3, and 18A processes as 18A wafer output ran 25 percent above target.

The 18A process ramp ran 25 percent above the company's internal target in Q2. Wafer volume rose more than 50 percent sequentially.

Intel is also advancing an 18A-P high performance variant on the same node family.

CEO Lip-Bu Tan told analysts on the Q2 conference call that the 14A node is gaining traction with outside customers. "I am pleased to see the increasing momentum on customer engagements for Intel 14A, and I am increasingly confident that the 14A will be highly competitive process offering across key vectors of performance, power, density, cost, and schedule."

Intel plans 14A risk production for internal products in the second half of 2027 and a high volume ramp in 2028.

Can EMIB-T packaging generate real revenue?

Intel's EMIB-T technology has drawn interest from Taiwan Semiconductor Manufacturing Co. TSMC cannot allocate enough CoWoS-L capacity to every customer.

Intel previously used an earlier EMIB version on the Ponte Vecchio GPU, which combined 47 chiplets across five process nodes from both Intel and TSMC in one package.

The packaging shortfall creates a real opening for Intel. EMIB-T could earn revenue from chipmakers unable to secure CoWoS-L slots.

Nvidia, AMD, and Broadcom continue to absorb most advanced packaging output on the market today.

How large is the custom silicon opportunity?

Custom processor work now runs at a $2 billion annualized rate at Intel. Chief Financial Officer David Zinsner said the business targets a $4 billion run rate "in the not too distant future."

Intel places the total addressable market at $100 billion.

The custom work spans CPUs and infrastructure processing units. Hyperscalers account for most demand in the category.

The business puts Intel against Marvell, Broadcom, and Alchip in custom silicon work for cloud customers.

What does Intel still lack?

Intel has no AI inference accelerator that matches Nvidia's GPUs or Groq's LPUs in commercial deployment.

The SambaNova partnership matters, but Intel does not control that roadmap. Tan doubles as SambaNova's chairman of the board.

SambaNova raised $1 billion in a Series F round three weeks before the Q2 report. The round valued the startup at $11 billion. Nvidia paid $20 billion earlier to acquihire Groq's LPU team.

That Groq price tag likely ended any Intel move to absorb SambaNova outright. Intel cannot fund such an outlay while supporting its foundry ambitions through US government aid and Nvidia equity.

Each new SambaNova funding round makes the math harder for Intel.

How far must Intel still climb?

The datacenter business peaked at $9.06 billion in Q2 2020. Operating income that quarter reached $3.43 billion.

Margins ran 45 to 50 percent between 2009 and 2015. Today's results mark recovery rather than a return to prior scale.

Intel's combination of AI host demand, foundry traction, and packaging capacity points toward steady improvement through the 14A high volume ramp in 2028, though inference accelerator competition will continue to weigh on the broader product portfolio.