Intel CEO admits it's "too late" to catch up in AI training

Intel's new CEO just told employees what everyone else was already thinking: when it comes to the AI boom, the chip giant is completely out of the race.

In a leaked global all-hands Q&A reported by Tom's Hardware and OregonTech, Intel CEO Lip-Bu Tan delivered a brutally candid reality check to staff. The former king of silicon hasn't just lost its edge—it has officially fallen behind.

The fall from grace

"Twenty, 30 years ago, we were really the leader," Tan told employees. "Now I think the world has changed. We are not in the top 10 semiconductor companies."

It is hard to overstate the contrast. Decades ago, Intel held an iron grip over computing, famously toying with the idea of buying Nvidia for $20 billion. Today, Nvidia is valued at $4 trillion, while Intel bled a staggering $16 billion in Q3 last year. Deep layoffs are cutting thousands of jobs worldwide as ballooning R&D costs for next-gen manufacturing nodes bite hard.

Intel's legendary vertical integration has largely turned into a liability. Stumbling fabrication nodes forced the company to outsource roughly 30% of its chip production to rival TSMC in 2025, including the entire compute tile for Lunar Lake. Meanwhile, AMD's EPYC processors have steadily eroded Intel's stronghold in data center CPUs.

Giving up on training

Tan made it clear that Intel is done chasing lost causes in the cloud. "On training, I think it is too late for us," he admitted regarding the giant machine learning models powering today's AI rush.

Instead of going head-to-head with Nvidia's runaway data center GPUs, Intel is pivoting toward edge AI—running AI workloads locally on client devices like PCs. Tan also singled out agentic AI as a core growth pillar and teased that several high-profile hires are being brought in to execute this narrower playbook. Rebuilding the company, Tan stressed, will have to be a "marathon" built on humility.

Why it matters

We think Tan's harsh assessment is the most sensible thing to come out of Intel's executive suite in years. Chasing Nvidia in hyperscale model training was burning billions with zero realistic path to victory. Admitting defeat lets Intel stop bleeding resources and focus on where it still has skin in the game: the massive PC client footprint.

For developers and PC buyers, this pivot means Intel will double down on device-side AI inference and NPUs. But edge AI won't be a cakewalk either. AMD, Apple, and Qualcomm are moving at lightning speed on local silicon, and Intel's historical software missteps won't magically vanish overnight.

For the wider industry, this cements Nvidia's near-monopoly on frontier AI training for the foreseeable future. More importantly, it brings Intel one step closer to what many see as inevitable: splitting its troubled foundry division into an independent entity and becoming a leaner, fabless designer. If 18A doesn't deliver a miracle within the next year, Tan's marathon might turn into a breakup.