Lunar Lake for the Win: Is This the Comeback Intel Needs? (Premium)

Intel’s multiple missteps this past year would be comical if they weren’t so tragic. But the microprocessor maker may have just pulled off an engineering coup with its “Lunar Lake” processors, which appear to live up to the company’s promises at a time when it desperately needs a win.

The past year has been a difficult one for once-mighty Intel, the hardware half of the powerful Wintel duopoly that’s dominated the PC industry since there’s been a PC industry. Intel’s market power can’t be overstated, but that’s even truer when you learn how this chip making colossus has leveraged that power and its size to undermine rivals like AMD and now Qualcomm through partnerships that often amount to little more than cash payments and other subsidies. Intel has been both great and terrible, the one hardware company that both Microsoft and PC makers can’t ignore or avoid, even when their strategies diverged.

But Intel has been knocked down a peg in the past 12 months, and it suddenly seems vulnerable. Long accustomed to ignoring the market forces that favored thin, light, and efficient mobile hardware, Intel played to its strengths with increasingly more powerful but less relevant chips. Until it couldn’t: In late 2023, Intel for the first time in the modern era shifted from its “tick-tock-tock” (what it called Process-Architecture-Optimization) release schedule and pushed out a new “Meteor Lake” family of mobile chips at a time when it traditionally led with desktop chips. It did that after sticking with its previous architecture for just two years, and then immediately ghosted Meteor Lake for the far more advanced–and expensive to develop–“Lunar Lake” and “Arrow Lake” architectures it’s now launching.

We might fairly refer to this dramatic series of quick shifts as “tick-tick-tick.” But whatever you call it, Intel has belatedly gotten the thin, light, efficient religion Microsoft, especially, has been proselytizing to it for years. That delay it tragic. But it’s impressive seeing the speed at which this behemoth has moved to correct a decade-plus of mistakes in such a short period of time.

Less impressive is the impact this shift has had on the company’s finances. Its rapid push to move quickly from architecture to architecture, a process that would typically take at least three or four years for each processor generation, has led to massive reliability issues that dwarf infamous past episodes like the Pentium 5 FPU (floating-point unit) bug or Skylake. And we’ve only heard the tip of the iceberg about these issues: While Intel took the rare step of publicly acknowledging a serious flaw in its 13th and 14th generation Core processors, and then the even rarer step of resolving the problems and extending warranties for free, other issues remain. And its PC maker partners, responsible for shipping fixes for these issues, have remained silent, as always. Intel is always listening, don’t poke the bear.

Intel’s quarterly financial reports over the past year aren’t that dire if you look just at its PC microprocessor business, which weathered an industry-wide post-pandemic sales slowdown as well as is possible. But the firm’s in trouble because of other issues, issues which, granted, are self-inflected. By ignoring the mobile boom and sticking with its inefficient x86 architecture, Intel ceded mobile to Arm, and it’s now suffering similar defeats in the data center. The next milestone is Intel’s last hurrah, the PC market. Apple dropped the company like a bad habit in 2020, switching the Mac from Intel x86 to its Arm-based Apple Silicon designs. And Qualcomm, a company that had, until this year, fielded nothing but pointless PC processors, engineered a turnaround of historic proportions with its Snapdragon X architecture. Suddenly, even the PC market isn’t safe.

Intel CEO Pat Gelsinger came to the company with a plan: He would reimagine the microprocessor maker as a U.S.-based alternative to TSMC, by far the world’s biggest manufacturer of microchips. This idea has some merit–the U.S. government sees this initiative as a national security issue, thanks to fears that China might invade Taiwan and seize control of TSMC. But it’s also expensive. Really expensive: Intel is investing north of $100 billion of its own money financing new and improved chip fabrication facilities in the U.S. and elsewhere. And that’s before you add in the tens of billions of dollars in loans and subsidies the U.S. government is contributing. To date, there’s not much to show for this effort, which lost $7 billion in just the most recent quarter.

But Intel’s financial woes extend beyond its chip fabrication aspirations. By jumping from chip architecture to chip architecture so quickly, as it has in the past few years, Intel is tossing aside its historic ability to refine an existing architecture over two to four years, a process that is much less expensive than drumming up new designs. The “Lunar Lake” chips that Intel just shipped were hastily thrown together using technologies it wouldn’t have otherwise shipped for at least two years normally. And the industry’s sudden to AI has piled on the hurt: Last year, Microsoft informed Intel that its NPU components were too underpowered to drive the innovations it was planning for Windows and a new generation of PCs, forcing the firm to push an improved component to market in a shorter period of time than originally planned. The costs of getting Lunar Lake to market in its shipping form so quickly were so great that the company was forced to call them out in its most recent earnings report.

If there’s a silver lining in this tale of woe, and I think there is, it’s that Intel apparently pulled it off. We need more time and experience with real-world Lunar Lake-based PCs before we’ll know for sure. And I am notably suspicious, given my experiences with Meteor Lake-based PCs this past year. But all the initial reports, and all the industry insiders I’ve spoken to, point to a positive reception for Lunar Lake. Intel, which has been face-raking on repeat for the past year, may have turned the corner, at least in this crucial area. (I can’t speak to its chip fabrication hopes.)

So what does that mean?

It means that Intel–like AMD, but more aggressively–has pushed to adopt the benefits of Arm with relinquishing everything that makes x86 not just a safe bet for customers but also truly great. This positioning was on full display at Intel’s pre-IFA Lunar Lake launch event, which was so full of bravado and chest pumping that you could almost forget its troubles. I found Intel’s hyper-focus on Qualcomm and Arm amusing given its years of publicly ignoring this company and its products. But the story it told was solid, and there was only one caveat in its claims, and it wasn’t particularly damaging.

“We have the fastest silicon in the industry, with industry-defining x86 performance per watt, incredible system designs built from the ground up to take full advantage of Lunar Lake, and an ever-growing set of optimized applications, games, and AI models,” Intel executive vice president and general manager Michelle Johnston Holthaus said at the event. “Our momentum is way above expectations.”

Those words aren’t that incendiary. But as she spoke, the slide that appeared behind her onstage spelled out the depth of Intel’s success with Lunar Lake. Compared to its modern competitors–Qualcomm, with the Snapdragon X Elite, and AMD, with its Zen 5-based Ryzen AI 300 series–both of which just launched, Intel is, by all accounts, suddenly in the lead. It has the fastest microprocessor, with an 18 percent advantage over its near competitor. It has the best integrated GPU, with a 30 percent performance advantage over its predecessor, a 35 percent advantage over AMD and a 2X gain over Qualcomm. It has the most efficient x86 processor–“x86” being the caveat noted above–with a 50 percent efficiency advantage over its predecessor. It offers “historic” performance per watt, up an incredible 200 percent over its predecessor. And it has the fastest NPU, with over 3x the performance of the admittedly underpowered Meteor Lake.

That’s broad strokes, yes. And ignoring the fact that Qualcomm is still more efficient and AMD and Qualcomm both offer what anyone would logically state is equivalent NPU performance, this is quite impressive. Intel’s core promise as the creator of the x86 has always been about compatibility and performance. And it wielded these strengths like cudgels during the launch event, repeatedly raking Qualcomm over the proverbial coals for its Snapdragon X compatibility problems, particularly in gaming. Compatibility with software and hardware is still the asterisk in the Qualcomm story, though I feel it’s a bit overblown in the real world, especially with mainstream users. But if Intel can match the efficiency and AI performance of Qualcomm–or at least be in the ballpark, as I like to say–then, what’s the point of Qualcomm and Arm?

That was Intel’s message last week in the form of that specific rhetorical question. Sure, it took Intel what seems like forever to get there. But it also seems that they did it. Speaking at IFA with several long-time friends who shall remain nameless for obvious reasons–they work for silicon makers, PC makers, and/or are hardware experts–a consistent narrative emerged. We’ve long known that Intel has the engineering chops to do this work. But we’ve also watched it not do so for years. With Lunar Lake, though, it appears to have responded to an emergency in the best way possible.

The one Intel talking point that everyone disputed was battery life. Intel claims that Lunar Lake topped an office productivity benchmark with 20 hours of battery life, compared to 18.4 hours for Snapdragon X Elite and 14.2 hours for its predecessor (Meteor Lake 155H). This “busted” the x86 battery life “myth,” Intel senior vice president Jim Johnson claimed. But that’s not representative of real-world battery life. Everyone I spoke with agreed that Snapdragon still has a comfortable lead in that category. And then I got mixed signals on where AMD and Intel landed. The source I trust the most said Intel was in second place, but it wasn’t clear where each landed.

It doesn’t matter. As with the Snapdragon X (Surface Laptop) and Apple Silicon M3 (MacBook Air) comparison I obsessed over for much of this year, it’s not reasonable to expect a newcomer to run the rack like Larry Bird at a three-point contest. Instead, what we’re looking for is for that newcomer to at least be in the ballpark. Qualcomm did it with Snapdragon X. And it appears that Intel has done it with Lunar Lake, at least from an efficient perspective. When you look at the other comparison metrics, Intel almost certainly comes out on top. That’s not just a win, it’s a rout.

What’s interesting about this is that Intel has only announced its most efficient next-generation chips: We won’t see what the company is doing in its more traditional space until it launch Arrow Lake, a related architecture that focuses less on efficiency and more on performance. That is, Lunar Lake is being marketed as Intel Core Ultra Series 2 chips in a new “V” variant. Arrow Lake will encompass U, S, and H variants. Right. Intel led with efficiency. I don’t even recognize this company anymore.

That may be the point.

As I noted earlier, Intel ignored obvious market needs for far too long, and it has had to scramble to catch up. And it is freely taking ideas from its competitors to get there. Lunar Lake chips include 16 or 32 GB of RAM that’s integrated on-die, just as we see with Apple Silicon and Snapdragon X. That means the RAM is non-upgradeable. But it’s also faster and more efficient. Each Lunar Lake chip has the same 4 performance cores, 4 efficiency cores, 8 Xe graphics cores, and 6 NPU cores. When you look at the Lunar Lake Core Ultra 9, 7, and 5 designs, the only real differences are clock speed and cache. (And the Core Ultra 5 has one fewer Xe and NPU core, each.)

There are still questions, of course. The most important will be answered by reviews of real, shipping Lunar Lake-based hardware. There are longer-term questions about reliability, given the issues we’ve seen in recent years. And then there’s Apple, which should launch M4-based Macs as soon as October. Intel, like Qualcomm and Microsoft, has made many comparisons with the M3, and especially with the MacBook Air. But that machine only ships with Apple’s base CPU, and we’ll almost certainly see higher-end M4 chips in various Macs. Perhaps the answer there is some combination of Lunar Lake and Arrow Lake solutions, each aligned against a specific M4. We’ll see.

Will Lunar Lake “save” Intel? That too, is unknowable in the short term. Here, I have bad news. One source told me that Lunar Lake is so expensive that Intel essentially loses money on each chip it sells. This wasn’t an issue when an earlier version of this architecture was seen as a vertical solution for higher-end products that would sell at higher prices. But the accelerated schedule and architecture hopping means that Lunar Lake is heading out to a much bigger market of less expensive PCs. And Intel can’t charge a premium for that. So Lunar Lake might be a negative on the front-end–during its rushed development–and on the backend as the resulting products head to market.

This reminds me vaguely of the Xbox and the video game console market, where the hardware is a loss-leader that’s hopefully balanced, over time, by game and peripheral sales and cost-cutting. But Intel is a microprocessor maker, so it can’t benefit from most of that. Perhaps in the long term, it can save money by manufacturing its own chips, should its 18A manufacturing process plans ever come to fruition. That still feels uncertain.

I don’t believe that “the x86 vs. Arm discussion is done,” as Intel now claims, but Intel has a good story here. It’s leading in key areas and are in the ballpark otherwise. The burden of responding is once again back in Qualcomm’s lap, and I assume we’ll learn about that in October during its annual Snapdragon Summit. For now, it’s a bit melancholy to realize that Snapdragon X enjoyed just a few months in the sun atop the PC industry in some ways. But that window has closed. And now consumers have more good choices than ever before in the history of this 40-year-old market. What a time to be alive.

Intel may be back. I wish I could be more definitive. But it looks good.