Intel Sinks $19B Into The Red, Kills Falcon Shores GPUs, Delays Clearwater Forest Xeons

Intel capped off a tumultuous year with a reality check for its product roadmaps.

Plans to commercialize the American giant's next-gen GPU architecture, codenamed Falcon Shores, have been scrapped, while its next generation of many-cored CPUs has been delayed until 2026, interim co-CEO Michelle Johnston Holthaus revealed Thursday along with the biz's financial figures for 2024.

Falcon Shores has undergone several transformations over the years. It was first envisioned as an XPU that would meld CPU and GPU tiles, and later pared back to a GPU that would combine Intel's Xe graphics lineage with Habana's AI chemistry. Now it seems the chip will never leave Intel's labs.

We plan to leverage Falcon shores as an internal test chip only

"Many of you heard me temper expectations on Falcon Shores last month," Holthaus told financial analysts on a conference call a few hours ago. "Based on industry feedback, we plan to leverage Falcon shores as an internal test chip only, without bringing it to market. This will support our efforts to develop a system-level solution at rack scale with Jaguar Shores to address the AI data center more broadly."

The revelations mark the latest blow to Intel's AI ambitions for the datacenter world. Last quarter, Intel admitted its Gaudi3 accelerators had failed to deliver on the $500 million revenues forecast for 2024.

"This is an attractive market for us over time, but I am not happy with where we are today," Holthaus said. "On the one hand, we have a leading position as the host CPU for AI servers, and we continue to see a significant opportunity for CPU-based inference on-prem and at the edge as AI-infused applications proliferate.

"On the other hand, we're not yet participating in the cloud based AI data center market in a meaningful way."

Intel's Xeon roadmap gets lost in Clearwater Forest

While Falcon Shores has effectively been canned, Intel's Clearwater Forest Xeons have only been delayed. The parts, slated to be among Intel's first datacenter products on its all-new 18A (1.8nm-class) process nodes, are now expected to arrive in the first half of 2026.

According to Holthaus, the rationale boils down to demand for Xeons bristling with E-cores: "What we've seen is that's more of a niche market, and we haven't seen volume materialize there as fast as we expected."

The decision to delay Clearwater Forest is less surprising as its original release schedule had it awkwardly close to the release of Intel's 288-core Sierra Forest Xeon 6 processors due out earlier this year.

While Clearwater Forest won't make its debut this year, Holthaus insists the 18A process tech is still on track and will ramp alongside the launch of Intel's Panther Lake client CPUs in the second half of the year.

"As the first volume customer of Intel 18A, I see the progress that Intel Foundry is making on performance and yield, and I look forward to being in production in the second half, as we demonstrate the benefits of our world class design," Holthaus emphasized.

She also revealed that Panther Lake would be followed by a next-gen product called Nova Lake, which would utilize a mix of internal and external fab processes, meaning a reliance on non-Intel factories.

Pat's Foundry vision lives on while search for CEO continues

Speaking of 18A, it's been nearly two months since chief exec Pat Gelsinger's abrupt "retirement" and the chipmaker has not yet named a worthy successor.

"The board remains intensely focused on the search for a permanent CEO," interim co-CEO and CFO David Zinsner said on the conference call. "The search is progressing, but we have nothing new to report," he added, preemptively shutting down questions ahead of a Q&A with Wall Street.

Even without a permanent CEO at the helm, Holthaus and Zinsner are pushing ahead with Gelsinger's grand plan to reinvent Intel as a manufacturing heavyweight. The CFO emphasized that building a successful foundry business requires more than competitive process technology; it needs trust from the semiconductor industry, which can take time to build.

We're going to systematically attack our costs

In other words, don't expect the foundry's financial situation to change overnight, though improvements are on the horizon with the impending ramp of 18A, allegedly. "We're going to systematically attack our costs and remain highly focused on our goal of delivering break even operating income for Intel foundry by the end of 2027 and we expect to demonstrate improvements this year," Zinsner said.

Work in support of this goal is still ongoing. According to Intel, tool installation at its Fab 52 site in Arizona is underway to support the ramp of 18A later this year. To support this project, Intel says it collected $1.1 billion of the $7.86 billion in subsidies awarded by the US Department of Commerce late last year, and another $1.1 billion this month.

In terms of foundry customers and who on Earth they are exactly, Intel remains fairly tight lipped. It claimed in December it completed a tape out of a 16nm-class design for a customer expected to begin volume production at its Ireland facility later this year.

Intel is expected to share more details regarding its manufacturing ramp in late April during its second annual foundry event in San Jose, California.

• Want Intel in your Surface? That'll be $400 extra, says Microsoft
• Intel Capital next into the chip giant's trebuchet, to be shot as far over the wall as possible
• Intel’s datacenter architecture boss and Xeon lead jumps to Qualcomm
• Intel sued again over struggling foundry business
• Intel pitches modular PC designs to make repairs less painful

Seasonality hits as Intel staunches the bleeding

Compared to Intel's disastrous performance over the past few quarters, which were punctuated by multi-billion-dollar losses, missteps, and unmet promises, its final three months of 2024 were comparatively uneventful.

Intel is still bleeding cash, recording a $100 million net loss in the fourth quarter, down 105 percent from the year-before's $2.7 billion profit. However, it managed to book revenues on the upper end of its guidance at $14.3 billion, albeit down seven percent year-on-year. So-called inter-segment eliminations also cost the chipmaker $4.3 billion in sales, it noted.

Digging a little deeper, Intel's Client Computing Group (CCG) once again did the heavy lifting, accounting for $8 billion of the corporation's revenues for the quarter, a decline of nine percent from the year prior.

The Datacenter and AI group fared a little better falling just three percent to $3.4 billion, while Intel's Networking and Edge (NEX) group was the lone bright spot in Intel's product mix with revenues growing 10 percent to $1.6 billion. However, NEX's rebound in edge demand wasn't nearly enough to overcome a six percent year-on-year decline in revenue, down to $13 billion, from the Intel products division.

And while Zinsner remains confident in Intel's foundry fortunes, that division continues to struggle with revenues down 13 percent year-on-year to $4.5 billion and operating losses for the quarter totaling $2.26 billion, nearly twice that of Q4 2023's losses.

Despite a challenging 2024, Intel's full-year revenues were almost flat at $53.1 billion, declining two percent compared to 2023. The same can't, however, be said of profits. The past 12 months saw Intel record $18.8 billion in losses versus a tidy $1.7 billion profit the year before. Gross margins dipped to 32.7 percent from the year-ago's 40 percent.

Looking ahead to Q1 2025, Intel is targeting revenues of $11.7 to $12.7 billion, with Zinsner citing regular seasonality – Q1 is usually slower than Q4 for sales – for the weak outlook. ®