The most important figure Intel reported during its 2Q25 earnings call was –$1.1 billion. That’s its adjusted free cash flow, a self-reported (and, honestly, hard-to-replicate) figure that’s a better measure of business health than profitability. When a business burns a billion or more dollars per quarter, surgery is required. Think tree surgeon, not brain surgeon. Consequently, the company has been slashing staff, curtailing capex, and spinning off noncritical operations.
The good news is that free cash flow was much better in Q2 than in prior quarters. The outlook for the remainder of 2025, however, isn’t great. The company expects 2H sales to be weak compared to 1H on the assumption that customers bought in advance of tariffs. Purchase commitments limit Intel’s ability to cut capex further this year. Meanwhile, it’s likely to incur additional restructuring costs. Next year, however, the company will reduce capex again and, we expect, restructuring costs will decrease. Moreover, Intel sees gross margins increasing to 40–60%, bad by its historical standards but a big improvement from recent levels.
Whither 18A?
Intel implicitly reiterated its support for the 18A process and the 14A follow-on. CEO Lip-Bu Tan (LBT) stated that the company will fabricate three product generations in 18A, which we interpret to be the Panther Lake, Nova Lake, and rumored Razer Lake PC processors. CFO David Zinsner went so far as to say that peak 18A volumes would occur next decade, which is only possible if additional products employ the process. Zinsner went on to state that while Intel hadn’t landed many external customers, the company expected to do so after performance and yield improved.
New Capex Strategy
Performance and yield, along with volume commitments, will also gate capex, LBT stated. By contrast, Pat Gelsinger, Intel’s previous CEO, closed joint-venture deals (semiconductor coinvestment programs—SCIPs) to build fabs. These reduced the company’s capital outlay compared with going it alone but put it on a path to spending and building more than it needed. During the 2Q25 earnings call, LBT was unsparing in his assessment of this strategy, saying, “The capacity investments we made over the last several years were well ahead of demand and were unwise and excessive.”
14A Future
Regarding 14A, Intel is addressing the various elements required to be a foundry business as it develops the process, suggesting that with 18A it was too focused on just the technology. Moreover, external customers are essential for the new node to be economical. The company expects 14A production to begin in 2028 or 2029, in line with TSMC.
PC Processors
Intel disclosed nothing new about PC processors, repeating past plans that Panther Lake will launch this year and Nova Lake will come out at the end of 2026. (The former will employ the 18A process, and the latter will be split between 18A and a TSMC node.) Panther is on track; its first wafers have started at the 18A fab.
Server Processors
The server-processor discussion differed. Last quarter, Intel described the current-generation P-core Xeon, Granite Rapids, as a good first step toward closing the gap with AMD and underscored the importance of delivering its successor, Diamond Rapids. This quarter, the prepared remarks didn’t explicitly mention Diamond. However, LBT referred to “past mistakes regarding multithreading capabilities of our P-cores” and stated, “Clearly, we have some mistakes we made on the high-end performing server area. And one thing is simultaneous multithreading. … It used to be Intel’s strength, but somehow we overlooked it.”
Simultaneous multithreading (SMT, which Intel brands HyperThreading) had been a P-core feature, but Intel’s most recent PC processors don’t implement it. On the basis of LBT’s comments, we can infer that Diamond doesn’t have SMT, either. Diamond, therefore, won’t be a good second step toward closing the competitive gap. Intel, however, referenced a subsequent chip, Coral Rapids, saying it would arrive in 2028 or 2029, aligning it with the 14A process.
AI
Intel disclosed nothing meaningful about its AI products, deferring discussion to the coming months.
Bottom Line
Intel’s 2Q 2025 earnings call made clear that the company’s turnaround is at a midpoint, not near the end. Staffing cuts and conservative capital expenditures are buying it time to complete the process. On the manufacturing side, that includes improving 18A performance and yield while simultaneously readying 18A and 14A for external use, keeping a finger on the detonator in case things go wrong.
On the product side, a detailed AI plan is still TBD. Setting aside manufacturing issues, PC processors are on track. (We have concerns about Intel’s post-Razer plans, however.) We had thought server processors were on the right track, catching up with AMD Epyc’s core counts. Learning Diamond doesn’t implement SMT, a feature cloud providers depend on for their vCPU offerings, we conclude AMD will keep taking share, at least until Coral launches.
In most ways, LBT’s strategy is like Gelsigner’s: retain manufacturing, offer foundry services, and pursue AI. Their execution differs, however. The new CEO is outspoken about his dissatisfaction with previous decisions on issues as diverse as capex and P-cores. Facing financial challenges, he is resolutely reducing costs.
Nonetheless, LBT’s tenure hinges on process technology as much as his predecessor’s did. Gelsinger had his 5N4Y plan, which culminated in the 18A process. Had the company proven unable to deliver 18A, he would’ve been replaced. While LBT hasn’t staked as much on 14A, it’s hard to see him keeping his job without the node succeeding. However, it was Intel’s finances, not its process development, that undid Gelsinger. By contrast, LBT is managing free cash flow and other business metrics. Gating capex on 18A and 14A success, he has indicated a willingness to walk away from manufacturing to avoid financial catastrophe.