Analysis Intel plans to layoff more than 16,000 staff, or at least 15 percent of its workforce, with most cuts coming by the end of the year as the x86 giant scrambles to get its finances under control.
“By implementing our spending reductions, we are taking proactive steps to improve our profits and strengthen our balance sheet,” Intel CFO David Zinsner wrote in the mega-corp’s second-quarter earnings report Thursday.
We expect these actions to meaningfully improve liquidity and reduce our debt balance
“We expect these actions to meaningfully improve liquidity and reduce our debt balance while enabling us to make the right investments to drive long-term value for shareholders.”
While Intel was widely expected to announce sweeping layoffs, the scope of the cuts is larger than anticipated. “Intel expects to reduce headcount by greater than 15 percent with the majority completed by the end of 2024,” the biz stated. That’s up from the eight percent rumored this week.
The Xeon maker employs about 110,000 workers across its various divisions, putting the total cuts at more than 16,000 heads.
These losses will impact multiple divisions including research and development, marketing, and general and administrative roles. Intel will also cut capital expenditures by more than 20 percent to between $25 billion and $27 billion in 2024 now that it’s achieved its goal of five process nodes in four years. The x86 titan plans to cut cap-ex spend back further in 2025 to $20 billion to $23 billion.
And all non-essential work is being stopped. “Intel is now shifting its focus toward capital efficiency and investment levels aligned to market requirements,” the company’s earnings report reads. Whether these reductions will disrupt the construction of Intel’s foundries, we’ll have to wait and see.
And in a blow to shareholders, Intel is also waving goodbye to its quarterly dividend beginning in Q4, as it recognizes “the importance of prioritizing liquidity to support the investments needed to execute its strategy.”
Combined, Intel expects these cuts to save it more than $10 billion in 2025, which it says will help it “achieve clear line of sight toward a sustainable business model.”
Intel caps ‘disappointing’ Q2 with $1.6B in losses
The slashing comes on top of a dismal three months for Intel, which racked up $1.6 billion in losses on revenues that fell one percent to $12.8 billion in its second quarter. This time last year, the corp recorded a $1.5 billion profit for Q2.
The one-two-three punch of layoffs, losses, and no dividends saw its share price plummet nearly 20 percent in after-hours trading to $23.50 apiece Thursday night.
“Profitability was disappointing despite continued progress on product and process roadmaps with our new operating model firmly in place,” CEO Pat Gelsinger confessed on Thursday’s earnings call.
Making matters worse, Intel’s Q3 outlook suggests the chipmaker’s road to recovery could be a long one. “Q3 will be impacted by a modest inventory digestion in CCG, with DCAI and our more cyclical businesses of NEX, Altra, and Mobileye, trending below our original forecasts,” Gelsinger warned.
CCG being its client or personal computing group, DCAI its datacenter AI wing, and NEX the networking and network edge unit.
In other words, Intel is in for a tough three months ahead, at least, with its third-quarter revenue forecast calling for between $12.5 billion and $13.5 billion — that’s roughly a 5 to 12 percent decline from the same time last year.
AI PCs now, AI accelerators later
Gelsinger blamed some of Intel’s performance during the second quarter on US export restrictions on sales to China and the decision to ramp production of its Core Ultra mobile processors faster to capitalize on the emerging AI PC segment.
In spite of this, Intel’s personal computing wing managed to be one of the few bright spots on its Q2 results, at least in terms of revenues, which grew nine percent year-on-year to $7.4 billion.
“We have now shipped more than 15 million Windows AI PCs since our December launch — multiples more than all of our competitors combined,” Gelsinger beamed. “We remain on track to ship more than 40 million AI PCs by year end with over 100 million accumulative by the end of 2025.”
We’ll note that while Intel is calling them AI PCs, the Meteor Lake processor parts Gelsinger is referring to fall well short of Microsoft’s minimum spec of 40 NPU TOPS for Copilot Plus PCs. It won’t have hardware available that meet this criteria until later next month.
While CCG revenues were up, the same can’t be said of Intel’s datacenter group, sales of which fell three percent year-on-year to $3 billion.
Intel’s datacenter revenues come in stark contrast to AMD, which, earlier this week, reported datacenter sales of $2.8 billion — $1 billion of which was directly attributed to sales of its MI300X GPUs for AI deployments.
But why?
Intel hasn’t managed to replicate this success, in part, because its Gaudi3 accelerators, announced this spring, won’t ship in volume until Q3. It’s a similar story with many of Intel’s Xeon 6 processors, including its 128-core Granite Rapids Xeons. Intel did push a 144 e-core Xeon back in June, but we expect it’ll be at least a few more quarters before we see any positive impact from that on Intel’s balance sheet.
This puts Intel in a somewhat awkward spot of being unable to capitalize on the bulk of the AI hype. Previous disclosures put Intel far behind rivals AMD or Nvidia’s successes in the space, with Intel’s neural network accelerators expected to drive just $500 million in revenues during the second half of 2024. By comparison, Nvidia’s datacenter division netted $22.6 billion in revenues in Q1 alone.
Despite this, Gelsinger argues Intel’s Gaudi3 accelerators will be competitive: “We expect Gaudi3 to deliver roughly 2x performance-per-dollar in both inference and training versus [Nvidia’s] H100.”
Don’t forget, though, that Intel will also have to contend with Nvidia’s H200, and by the end of the year, the first of Nvidia’s Blackwell and AMD’s 288GB MI325X accelerators.
Intel’s NEX, meanwhile, brought in $1.3 billion in revenues, down one percent from the year-ago quarter.
Gelsinger remains hopeful growing demand for high-speed networking for AI clusters will help to drive growth in this sector before long, touting Intel’s work on Infrastructure Processing Units — sometimes referred to as SmartNICs, SuperNICs, or DPUs — and its work on the Ultra Accelerator Link standard, which our sibling site The Next Platform has broken down in detail here.
Outsourcing to TSMC
Intel Foundry – the division that manufactures chips not just for Intel but for customers from their own blueprints on a contract basis – showed some signs of recovery with revenues up a modest four percent year-on-year in Q2 to $4.3 billion.
Intel Foundry was a topic of concern on Thursday’s earnings call, as financial analysts attempted to wrap their heads around what Intel’s cap-ex cuts might mean for the unit. Intel has committed to investing tens of billions of dollars on new fab sites across the US, Europe, and Middle East, and will use subsidies from America’s CHIPS Fund at least to help pay for it all.
“Fundamentally we believe our strategy will be maintained even as we get these more efficient steps in place,” Gelsinger said. “On the CHIPS [Act] side, these are milestone-based investments; we still believe that we’re comfortably able to execute against those milestones and across the projects that we’ve announced.”
However, at least as far as Intel’s own products, it seems that more chips traditionally built on its internal process tech will be outsourced to Taiwanese rival TSMC. This will include its Lunar Lake mobile chips due out on Sept. 3.
And this outsourcing of fabrication to the rival factory is expected to continue through much of 2025 before product returns home in 2026, Gelsinger explained on the call.
While Intel has used outside fabs for certain components, including GPUs, SoC tiles, and its Gaudi accelerators, it has, with a few exceptions, manufactured its CPUs in house.
The implication here is that Intel plans to use TSMC as a stop gap while it gets its internal 20A and 18A process nodes ready for volume production. ®