Qualcomm Will Pay Huawei for Patents for the First Time in 25 Years: the Deal Covers 5G, AI and the LogicFolding 3D Chip Architecture
On October 5, 2026 Qualcomm and Huawei signed a multi-year patent deal covering 5G, AI and networking, including the patents behind LogicFolding, Huawei's stacked-chip technique. For the first time in 25 years, the money flows the other way.
What Was Signed on October 5, 2026
The announcement was a joint one: Huawei and Qualcomm published it the same day, with the same core text. It is not a leak or a press interpretation.
Cross Licenses Across 5G, Compute, AI and Networking: What "Cross" Actually Means
A cross license is mutual permission: each company lets the other use its patent portfolio and, in the end, whoever holds the heavier portfolio collects the difference in cash. That is the real shift here: the balance moved Huawei's way.
The fields covered, per the joint statement, are 5G, compute, artificial intelligence and networking. That this is the first licensing agreement between the two that includes 5G is the detail industry coverage singled out.
Read also
Qualcomm Will Buy Certain Huawei U.S. Patents (Pending Regulatory Approval)
Beyond the cross licenses, Qualcomm will purchase certain Huawei U.S. patents in the areas of compute, AI, networking and other technologies. Both companies said the transaction will close once regulatory approvals are received: timings depend on third parties.
The $6.9 Billion Figure Huawei Gave: Cumulative Value, Not This Contract
Huawei said the deal will push the cumulative value of its patent licensing agreements above $6.9 billion once completed. Mind the fine print: that is the total across its licensing portfolio, not the amount of this contract, which has not been disclosed.
Why Reporters Call Qualcomm the "Net Payer" After 25 Years of History
Huawei paid its first license to Qualcomm in 2001, and its first revenue from licensing its own technology came a decade later, through Motorola. That flow kept the same direction for 25 years; several outlets describe the new deal as the moment Qualcomm becomes the net payer, a framing the companies themselves do not use.
What LogicFolding Is, Without the Jargon
The Problem: Moore's Law No Longer Gives Free Gains, and Huawei Can't Buy the Best Machines
For decades, improvement came from building smaller transistors. That path has gotten expensive, and export restrictions cut Huawei off from the most advanced tools. When you cannot shrink the part, the way out is to change the shape of the chip.
The Idea: Folding Layers of Logic to Shorten the Paths Data Travels
LogicFolding splits a functional block at the logic-gate level and spreads it across two active silicon layers bonded face to face with hybrid bonding. When circuits that talk to each other sit physically closer, data travels less distance: less time and less energy.
The Tau Scaling Law: the Name Huawei Gives Its Own Roadmap
Huawei frames the strategy as the Tau Scaling Law, its alternative framework to Moore's Law. The company states a goal of transistor density equivalent to a 1.4 nanometer process by 2031 without relying on EUV lithography. That is a stated target, not a chip that exists: there is no Huawei silicon manufactured at 1.4 nm today.
What You Gain and What You Pay
You gain performance per watt: more work on the same energy budget. You pay in manufacturing: more layers require more wafers for the same number of chips and more area where a defect can land. That is the price of not being able to shrink the transistor.
Not the Same as TSMC's or AMD's Stacking, but the Same Family
TSMC stacks with SoIC and CoWoS, and AMD popularized the idea with its 3D V-Cache, but those cases join blocks designed separately. LogicFolding splits one functional block across two layers from the design phase. Same family of solution: when node shrinking stalls, performance per watt comes from stacking.
Where LogicFolding Is Already Shipping
The Mate 90 Series and the Kirin 9050 Pro, the First Production Chip With This Architecture
Huawei unveiled the Mate 90 series on October 1, 2026. The Mate 90 Pro Max carries the Kirin 9050 Pro, which the company describes as the first mass-produced flagship SoC to fully integrate LogicFolding. This is not a lab announcement: it is a chip already on sale.
Which Models Have LogicFolding and Which Only Have the Tau Law
The architecture did not reach the whole lineup: the base models and the Mate 90 Pro use Kirin variants based on the Tau Law, such as the Kirin 9030 and Kirin 9035, without stacking.
What Huawei's Performance Numbers Say, and What They Don't
Percentages are circulating: 42 percent overall performance, 38 percent on GPU, 140 percent on NPU. Those are Huawei's own figures or figures from outlets quoting the company, not independent measurements, and they vary by source and by model compared.
Why a Patent License Is Front-Page News
Sanctions Cut the Chips, Not the Patent Business
U.S. export controls apply to the shipment of products, not to intellectual property, which gets licensed and bought through other channels. That is why this deal coexists with the restrictions in place, while Qualcomm's silicon sales to Huawei all but disappeared. The pattern showed up the same month in the Nvidia-chip server smuggling case, with one difference: here no hardware moves, knowledge does.
The Recent Precedent: Huawei's Deal With HP
On August 26, 2026 Huawei and HP signed a multi-year global patent cross-licensing agreement that includes a license to HP for certain Huawei Wi-Fi patents, ending a Wi-Fi 6 lawsuit. The patent portfolio works as a business line, not just a legal shield. For context on Huawei's AI chip strategy, the blog covered the Ascend 960DT being pulled forward to early 2027.
Patents as a Revenue Line
Huawei has spent years turning its R&D into royalties, with that line positive since 2021. For a company with its hardware markets closed off, charging for what it already invented does not depend on buying any machine.
What This Means for You
If You Buy Hardware
Shopping by nanometers falls short: two chips on the same node perform differently depending on how they are organized inside. The useful question becomes how much work per watt.
If You Build Software
Silicon efficiency decides battery, heat, and how much local inference a phone can handle without reaching for the cloud. Every step moves the floor of what a small device is expected to do.
If You Follow Geopolitics
Restricting hardware sales does not block the movement of knowledge: a company sanctioned in the product market can end up collecting royalties from the one sanctioning it.
What Comes Next
Three questions stay open: regulatory approval of the patent purchase, whether other manufacturers license LogicFolding — turning it into a de facto stacking standard — and whether the architecture reaches Huawei's AI chips, where performance per watt matters even more than in a phone.
Conclusion
What happened on October 5 is neither a surrender nor a military victory: it is a broad licensing deal with interests on both sides. Qualcomm gets access to technology and to U.S. patents; Huawei turns its R&D into revenue and recognition. Underneath sits what will last longest: when shrinking transistors gets hard, the way out is to stack, and that road no longer belongs to a single manufacturer. If you follow the chip industry, stay with the blog: we will track the approvals and the 2031 targets.
