Intel has put a number on its robotics ambitions. The company’s senior vice president, Cho Min-sung, told attendees at the Semiconductors Powering Humanoids conference in South Korea on October 16, 2026, that Intel’s planned “Razor Lake-AX” platform is targeting roughly 700 TOPS of AI compute when it arrives in 2028. The disclosure, first reported by TechPowerUp, reframes a chip that chip-watchers had long filed under “future laptop silicon” as something aimed squarely at humanoid robots and other physical AI systems.
Razor Lake-AX is not shipping next quarter, and Intel has not published a product brief. What exists right now is a roadmap slide and a quote. But the quote is specific enough, and the venue unusual enough, that it is worth unpacking: what 700 TOPS means, how it stacks up against Intel’s current Panther Lake platform, and why Intel chose a humanoid-robotics conference, not a PC trade show, to say it out loud.
What Intel Actually Confirmed About Razor Lake-AX
According to Cho Min-sung’s remarks, reported first by Korean outlets and picked up internationally by TechPowerUp, Wccftech, and GameGPU, Intel plans to introduce the next-generation Nova Lake platform in early 2027, followed by Razor Lake-AX in 2028, targeting performance of around 700 TOPS. That is the extent of what Intel has put on record. The company did not disclose process node, core configuration, memory architecture, power envelope, or price.
It also did not specify whether the 700-TOPS figure is measured using dense or sparse compute, a distinction that can swing reported AI performance numbers by a wide margin depending on which vendor is doing the measuring. Digital Trends flagged the same gap, noting that the reported figure is a roadmap target rather than a confirmed shipping-performance specification. Readers should treat 700 TOPS the way they would treat any pre-silicon marketing number: a directional claim, not a benchmark result.
What is confirmed is the comparison point Intel itself offered. The company’s current robotics development platform, built on Panther Lake, is rated at up to 180 TOPS combined across CPU, GPU, and NPU. Razor Lake-AX’s 700-TOPS target works out to roughly 3.9 times that figure. That is a large jump for a chip still two years from launch, and it is the headline number every outlet covering the story has repeated.
Why Intel Picked a Robotics Conference, Not a PC Launch
The venue matters as much as the number. Intel has spent the past two years pitching Panther Lake, Core Ultra Series 3, and the Nova Lake chips still in development primarily as AI PC silicon, chips meant to run Copilot-style assistants and local inference inside laptops. Razor Lake-AX was announced somewhere else entirely: a conference dedicated to semiconductors for humanoid systems.
That is a deliberate signal. Intel is not positioning Razor Lake-AX as the next Core Ultra for laptops. It is positioning the chip as infrastructure for what the industry has started calling physical AI: systems that pair cameras, lidar, force sensors, and proprioceptive feedback with local inference and real-time control, running inside a body that has to balance, walk, and manipulate objects without a round trip to the cloud. Peak TOPS is one input into that kind of system, but it is far from the only one. Latency has to be deterministic, power draw has to fit inside a battery-powered chassis, and the chip has to run perception, planning, and motor control workloads at the same time rather than in sequence.
Intel’s framing suggests the company expects that market to exist at meaningful scale by the end of the decade. Whether that bet pays off depends on factors well outside chip design, including how fast humanoid robot makers can solve the much harder problems of locomotion, manipulation, and safety certification.
Intel’s AI Compute Roadmap: Panther Lake to Nova Lake to Razor Lake-AX
The three platforms now form a visible progression in Intel’s public roadmap, even if two of the three remain unreleased. Panther Lake is current and shipping in robotics development kits today. Nova Lake is the near-term successor, due in early 2027, positioned as the next step up for PCs and development platforms generally. Razor Lake-AX sits a year beyond that, carrying the 700-TOPS target and the robotics-specific branding.
| Platform | Planned Timing | Reported AI Compute | Primary Target Market |
|---|---|---|---|
| Panther Lake / Core Ultra Series 3 | Current | Up to 180 TOPS (combined CPU, GPU, NPU) | AI PCs; robotics development platforms |
| Nova Lake | Early 2027 | Not disclosed | Next-generation PC platform |
| Razor Lake-AX | 2028 | Approximately 700 TOPS (roadmap target) | Humanoid robots, autonomous mobile robots, physical AI |
Read the roadmap this way: Intel is no longer treating Nova Lake as the end of the line. Razor Lake-AX extends the sequence by a full platform generation, dedicated specifically to a market Intel barely discussed in public eighteen months ago. That is the real story behind the 700-TOPS headline. Intel did not just raise a performance number; it created an entirely new branch of its roadmap for robotics silicon.
700 TOPS in Context: What the Number Does and Doesn’t Tell Us
TOPS, trillions of operations per second, is a peak theoretical figure. It describes the maximum number of low-precision arithmetic operations a chip can perform under ideal conditions, not how a robot running a vision-language-action model will actually perform in a warehouse or a living room. Vendors have leaned on TOPS as a marketing shorthand for years, in part because it is easy to put on a slide and hard for a reader to independently verify.
The available reporting does not establish several things that matter for real-world robotics performance: Razor Lake-AX’s actual memory bandwidth, its power envelope, its software stack, or whether the 700-TOPS figure is measured with dense or sparse computing. Each of those variables can change what “700 TOPS” means in practice by a large margin. A chip that hits a theoretical peak using sparse matrix tricks under laboratory conditions can perform very differently from one hitting the same number on dense, real-world workloads.
That is also why the Panther Lake comparison needs a caveat. Panther Lake’s 180-TOPS figure is itself a combined CPU, GPU, and NPU number tied to INT8 dense precision, not an NPU-only figure. If Razor Lake-AX’s 700-TOPS target is calculated on a different basis, the “3.9 times” multiplier that outlets have reported, while mathematically accurate based on the two published numbers, may not represent an apples-to-apples jump in usable AI performance.
Razor Lake-AX: What’s Confirmed Versus What Intel Hasn’t Said
Given how much speculation tends to fill the gap around early chip roadmap disclosures, it is worth separating what Intel has actually confirmed from what remains open.
| Detail | Status | What’s Known |
|---|---|---|
| Product name | Confirmed | “Razor Lake-AX” |
| Planned timing | Confirmed | 2028 |
| AI performance target | Confirmed (roadmap target) | Approximately 700 TOPS |
| Intended application | Confirmed | Robotic systems, including humanoid robots and physical AI |
| Comparison baseline | Confirmed | Roughly 3.9x Panther Lake’s reported 180 TOPS |
| Processor architecture | Unconfirmed | Not disclosed |
| Power consumption | Unconfirmed | Not disclosed |
| Memory capacity and bandwidth | Unconfirmed | Not disclosed |
| Measurement precision (dense vs. sparse) | Unconfirmed | Not disclosed |
| CPU/GPU/NPU performance split | Unconfirmed | Not disclosed |
| Price | Unconfirmed | Not disclosed |
That is a long unconfirmed column for a two-years-out product, but it is normal for this stage of a chip’s public life. Intel disclosed Nova Lake’s existence and rough timing long before it released core counts, cache sizes, or power ratings, and those details only firmed up gradually through leaks and partial confirmations over the following year. Razor Lake-AX appears to be following the same script, just with a more specific headline number attached from day one.
Market Impact: A New Line Item for Intel’s AI Story
Intel has been fighting to keep its AI narrative relevant against Nvidia’s dominance in training and inference silicon and against AMD’s gains in CPU market share. A dedicated robotics roadmap gives Intel a story that does not require it to out-muscle Nvidia on data center GPUs. Instead, it stakes out a category, edge silicon for embodied AI, where the competitive field is less settled and where Intel’s existing manufacturing and platform relationships could matter more than raw accelerator performance.
For Intel’s investors and partners, the practical takeaway is modest in the near term. Razor Lake-AX will not affect Intel’s revenue, product mix, or client relationships before 2028 at the earliest. What it does is give Intel a public answer to a question that has been circulating in robotics circles for the past year: which major chipmakers are building silicon specifically for humanoid and mobile robotic systems, rather than adapting PC or data center chips after the fact. Intel has now put itself on that list with a specific number attached, which is more than most of its rivals have done publicly.
Historical Context: How We Got From AI PCs to Robot Brains
The AI PC category that Intel, AMD, and Qualcomm have all chased since 2024 was built around a different premise: run a chatbot-style assistant and some local image or audio processing on a laptop NPU, largely to sell a refresh cycle. Microsoft’s Copilot+ PC branding pushed NPU performance into mainstream marketing, and TOPS figures in the tens became a standard spec line on laptop boxes.
Razor Lake-AX represents a jump to a different order of magnitude and a different use case entirely. Rather than incrementally raising laptop NPU performance by a few TOPS per generation, as Intel did moving from Lunar Lake to Panther Lake, the company is proposing a platform aimed at robots that need to process multiple camera feeds, run perception models, and execute control loops simultaneously, all without a cloud connection they can depend on for safety-critical decisions. That is a genuinely different design target than a laptop chip optimized for battery life and occasional AI assistant queries, which is part of why Intel chose to announce it at a robotics conference rather than folding it into its usual PC platform messaging.
The Competitive Landscape: Where AMD, Qualcomm, and Nvidia Stand
Intel is not the only chipmaker eyeing physical AI as a growth category, though the available reporting on Razor Lake-AX does not include directly comparable TOPS figures from AMD or Qualcomm for robotics-specific silicon, and attributing specific numbers to their roadmaps here would risk mixing NPU-only claims with total platform TOPS, a mistake Intel’s own Panther Lake figure already illustrates the danger of.
What is publicly known is directional. AMD has said its AI chip demand is expected to outpace supply well into 2028, a timeline that overlaps almost exactly with Razor Lake-AX’s planned launch window. Qualcomm’s leadership has separately suggested that phone-class chips could eventually run AI models with tens of billions of parameters by around the same period, pointing to its own ambitions in on-device AI compute, even if that specific roadmap is aimed at phones rather than robots. Nvidia, meanwhile, remains the dominant force in data-center AI silicon and has been expanding into robotics-adjacent platforms of its own, giving it a head start on the software and developer ecosystem side that Intel will need to match if Razor Lake-AX is going to attract robotics customers rather than just headlines.
The upshot is that Intel has made the most specific public robotics-chip performance claim among the major CPU and GPU vendors so far, but specificity on a roadmap slide is not the same as a demonstrated lead. Two years is a long runway for AMD, Qualcomm, or Nvidia to publish competing numbers of their own.
What Robotics Developers Should Actually Take Away From This
For engineers building humanoid or mobile robot platforms today, Razor Lake-AX is not an actionable data point yet. There is no development kit, no SDK, no reference design, and no confirmed instruction set to target. The practical advice for anyone building robotics software in 2026 is unchanged: continue developing against today’s available platforms, including Panther Lake-based robotics kits, and treat Intel’s 2028 target as a planning signal rather than a procurement option.
That said, the announcement is a useful signal for anyone mapping out which silicon vendors are likely to compete for robotics design wins over the next several product cycles. Intel joining that race formally, with a named platform and a headline performance figure, changes the calculus for startups and established robotics companies deciding which chip ecosystems to build long-term software investments around.
Predictions: Where Intel’s Robotics Push Goes From Here
- Expect Intel to release partial Razor Lake-AX specifications, likely architecture and process node details, well before 2028, following the same gradual-disclosure pattern it used with Nova Lake.
- Expect AMD and Qualcomm to respond with their own robotics-specific roadmap announcements within the next 12 to 18 months, if only to avoid ceding the “physical AI silicon” narrative to Intel by default.
- Expect continued scrutiny of how TOPS figures are measured industry-wide, as more vendors make robotics-specific claims that are difficult to compare without a standardized benchmark.
- Expect Nova Lake’s early 2027 launch to serve as the more immediate test of Intel’s execution, since any slippage there would cast doubt on the 2028 Razor Lake-AX timeline as well.
- Expect humanoid robotics hardware makers to treat chip TOPS targets as one input among many, with battery life, actuator control latency, and software maturity likely to matter more to early commercial deployments than peak compute alone.
The Bigger Picture: Intel’s Bet on Physical AI
Razor Lake-AX is a bet that the humanoid robotics market will be large enough, and mature enough, by 2028 to justify a dedicated chip platform. That is not a guaranteed outcome. Humanoid robots remain, for the most part, in pilot programs and demonstrations rather than widespread commercial deployment, and the hardest problems in the field, reliable locomotion, safe physical manipulation, and cost-effective manufacturing, are not primarily compute problems. A 700-TOPS chip does not solve a robot’s balance control or its ability to safely hand a human a cup of coffee.
What it does do is position Intel to supply the brains if and when those other problems get solved at scale. That is a reasonable hedge for a chipmaker with Intel’s manufacturing base and existing relationships with PC and embedded-systems customers who may pivot toward robotics. Whether it becomes a meaningful revenue line, or another unreleased roadmap slide that quietly slips, will depend on execution over the next two years, a window that has not been kind to several of Intel’s recent platform promises as reported by Windows Forum.
What Still Needs to Happen Before Razor Lake-AX Ships
Between now and 2028, Intel needs to clear several milestones that have not yet been publicly addressed. It needs to finalize and ship Nova Lake on schedule in early 2027, since any delay there would push the entire roadmap back. It needs to disclose Razor Lake-AX’s actual architecture, fabrication process, and power envelope, none of which have been confirmed. And it needs to build a software and developer ecosystem around the platform that gives robotics companies a reason to design around Intel silicon specifically, rather than treating it as a generic accelerator. Eastern Herald noted that the information disclosed so far does not establish how the chip would perform in a working robot, a gap that will need to close well before 2028 if Razor Lake-AX is going to attract real design wins rather than just press coverage.
Frequently Asked Questions
What is Intel Razor Lake-AX?
Razor Lake-AX is the name Intel has given to a planned 2028 chip platform aimed at robotic systems, including humanoid robots and other physical AI applications. It was disclosed by Intel senior vice president Cho Min-sung at a humanoid robotics conference in South Korea.
How much AI performance will Razor Lake-AX deliver?
Intel is targeting approximately 700 TOPS of AI compute. This is a roadmap target, not a confirmed shipping specification, and the measurement basis, including whether it uses dense or sparse computing, has not been disclosed.
When will Razor Lake-AX be released?
Intel has said Razor Lake-AX is planned for 2028, following the Nova Lake platform, which is expected in early 2027. No exact launch date has been confirmed.
How does Razor Lake-AX compare to Panther Lake?
Intel’s current Panther Lake platform is reported at up to 180 TOPS combined across CPU, GPU, and NPU. Razor Lake-AX’s 700-TOPS target is roughly 3.9 times higher by that peak-value comparison, though the two figures may not use identical measurement methods.
Is Razor Lake-AX a laptop chip?
No. Unlike Panther Lake and Nova Lake, which target PCs and AI PCs, Razor Lake-AX is specifically aimed at robotic systems, including humanoid robots and autonomous mobile robots, rather than conventional laptops.
Who announced Razor Lake-AX for Intel?
The disclosure is attributed to Cho Min-sung, identified in reporting as Intel’s senior vice president, speaking at the Semiconductors Powering Humanoids conference.
What details about Razor Lake-AX remain unknown?
Intel has not disclosed the chip’s architecture, release date beyond the year 2028, price, power consumption, memory capacity, real-world performance, or the precision basis used for the 700-TOPS figure.
Does this mean Intel is exiting the AI PC market for robotics?
No. Razor Lake-AX appears to be an addition to Intel’s roadmap rather than a replacement for its PC-focused platforms. Nova Lake, arriving in early 2027, remains targeted at conventional AI PCs.