Ten years. That is how far ahead Huawei asked its audience to look on Friday, at an international media roundtable in Shenzhen, and the length of the horizon was itself the message. Companies worried about next quarter do not talk in decades. Richard Yu, the company's executive director and consumer business chairman, laid out an argument that runs well past any single handset: the smartphone in a customer's pocket, he said, is only the first of five places Huawei intends to run that customer's digital life.
The handset in question is the Mate 90 series, just put on sale in China. Its chip is the Kirin 9050 Pro. Huawei says overall device performance is 31 percent higher than the previous generation.
Read that figure the way launch figures deserve to be read. It comes from the company's own labs. It has not yet been tested independently. What is not in dispute — and what matters more — is how Huawei says it got there: with a list of components it designed itself. The Balong modem. The image signal processor. A dedicated mobile security processor. The Lingxi CPU. The Maleoon GPU. And a neural processing unit built on the company's own Da Vinci architecture.
The quiet components matter most
Two items on that component list carry more weight than their billing suggests, and neither of them is the CPU.
Start with the security processor. Unglamorous? Completely. But it is the piece that decides whether banks and payment networks trust the device with credentials. That trust is a certification game, slow and unphotogenic, and it determines whether a phone can carry digital keys, transit cards and, eventually, the digital identity documents governments keep piloting. Win the certifications and the phone stops being a gadget. It becomes a wallet, a key ring and an identity card that happens to make calls.
Then the image signal processor. Flagship photography in 2026 is not won with sensors. Sensors have converged. Lenses have converged. The gap between a good night photograph and a great one increasingly sits in the quality of the silicon cleaning up the signal, and that silicon is the ISP.
Huawei knows this terrain well. It built its premium reputation on camera performance during the years when it could buy best-in-class components from whoever made them. Keeping that reputation on its own ISP is a different sort of test. Quieter. Harder. No benchmark headline covers it, and no single launch event can fake it.
Five scenarios, one operating system
Yu framed the consumer strategy around what Huawei calls a seamless AI life. It is spread across five scenarios: smart travel through the HIMA vehicle alliance, HarmonyOS-powered office work, the smart home, fitness and health, and entertainment. And he was plain that this is the company's long-term strategy for the next ten years — not a seasonal marketing theme that will be quietly replaced when the next phone arrives.
The logic underneath is familiar to anyone watching the platform wars. Phones are a mature market now. Margins are thin, replacement cycles are slower, and the handset alone no longer generates the growth it once did. Cars, homes and wearables generate something else: the data, and the lock-in. Whoever owns the connective tissue between those devices owns the customer relationship. Huawei's bet is that an operating system it controls — HarmonyOS, built after it lost access to Google's services — can be that tissue.
Of the five scenarios, smart travel is the one with money attached today. The arrangement is worth stating plainly, because it mirrors the phone playbook exactly. Huawei does not build cars. It supplies the cockpit systems, the driver-assistance software and the HarmonyOS integration that Chinese automakers sell under the HIMA alliance. Partners carry the manufacturing risk. Huawei owns the operating system and the silicon-adjacent software. A business that began as a press-release curiosity has become a genuine revenue line.
Insurance, paid in yuan
Bets like this are usually announced with adjectives. Yu brought a number instead. Between 2007 and 2025, he told the roundtable, Huawei invested more than 1.8 trillion yuan in research and development. The annual share of revenue going to R&D climbed from roughly a tenth in the early years to more than a fifth recently.
Spending figures at launch events are usually decoration. This one deserves a second look. The figures track with the company's published annual reports over the period, which have consistently shown R&D intensity rising even as smartphone revenue outside China collapsed under export restrictions. And a fifth of revenue on research is a ratio worth pausing on: it belongs to pharmaceutical companies and frontier AI labs, not consumer electronics firms, most of which run far leaner and buy their differentiation from suppliers.
Huawei has inverted that model, deliberately. It has decided the suppliers are the vulnerability. Seen that way, every yuan of the 1.8 trillion is insurance against the next export-control headline — insurance that pays out in Kirin tape-outs, in a modem team that does not need Qualcomm's permission, and in an NPU architecture whose roadmap no foreign regulator can freeze. What the money also bought was less visible and arguably worth more: a patent portfolio, and a bench of chip designers deep enough to keep the Kirin line alive when most analysts expected it to die.
Which brings the story back to the chip. Semiconductor watchers will read the Mate 90 less as a phone launch than as a progress report on China's domestic chip stack. The interesting question was never whether Huawei could design a competitive processor — its HiSilicon unit was doing that before 2019. The question was whether it could get one manufactured at scale, at a modern process node, without the supply chain it used to rely on. A Kirin chip shipping in a flagship phone, in volume, with a claimed 31 percent generational jump, suggests the manufacturing workaround is holding. The same self-reliance logic is playing out far beyond phones, from sovereign AI deployments built on open models to the build-out of enterprise data centers around Nvidia's Blackwell superchips, where control of the hardware layer has become its own strategy.
The Da Vinci NPU, sitting inside that chip, deserves a caveat and a compliment. The caveat first: Huawei's on-device AI demonstrations, like everyone else's, are curated. Nobody demos the questions their assistant fumbles. The compliment: putting a serious neural accelerator in the phone, rather than renting intelligence from a cloud data center, is the architecturally honest answer to the privacy and latency problems cloud assistants keep creating. The wider industry is racing both ways at once — toward ever-larger models thinking longer in distant data centers, some of them now linked by silicon photonics built to shatter bandwidth limits, and toward smaller, faster answers computed on the device in your hand. Both bets can be right. The phone that answers instantly and privately will win the small tasks. The cloud will keep the hard ones.
None of that removes the strategy's central risk, and it is financial, not technical. Research intensity that high has to be fed by profitable volume somewhere. Huawei's most profitable historical engine — premium phones sold globally — is the very business the restrictions were designed to shrink. So the Mate 90 has a commercial job as well as a technical one: to prove the Chinese premium market alone can carry the research budget of a company still behaving like a global one. That is a much harder trick than designing a fast chip. It is also the context in which the 31 percent claim should be read — not as a benchmark, but as evidence offered to the home market that the insurance is working.
The wall outside China
For all the silicon progress, Huawei's oldest problem is still standing where it always was. Hardware parity does not restore the app ecosystem the company lost outside China. A ten-year strategy built on HarmonyOS needs developers to believe the platform will be worth their time in year three, not just year one — in an industry where developers can increasingly reach for open-source models that have overtaken proprietary systems in code generation and a crowded shelf of rival platforms competing for the same hours.
Inside China, where the Mate 90 will do most of its selling, the argument is far easier to make. HarmonyOS has the installed base there. It has the state-adjacent tailwinds. It has a home market that treats buying Huawei as a mildly patriotic act. Outside China, the five-scenario pitch runs into a wall of entrenched platforms — Apple's services bundle, Google's AI subscriptions, Samsung's hardware reach — each telling its own version of the same seamless-life story.
Then there is the question the roundtable format is designed to avoid: yields and cost. A flagship chip made on a constrained domestic process is an engineering triumph and, very likely, an expensive one. Huawei can absorb that cost as the price of strategic independence — once, on a flagship, as a statement. Whether it can absorb it across a full product line, year after year, while also funding a car business, a cloud business and a 1.8-trillion-yuan research habit, is the real test of the strategy Yu described.
So what would prove Friday's confidence is earned? Three things. Independent benchmarks of the Kirin 9050 Pro, once reviewers outside Huawei's orbit get retail units — and not just peak scores, either: sustained performance under thermal load and modem efficiency on weak signals are where in-house silicon historically shows its age, and where a Balong modem designed alongside the main processor could plausibly shine. Mate 90 sales in its first quarter, measured against the Mate 80's run and against Apple's iPhone in the premium tier Huawei cares about. And the developer count for HarmonyOS at Huawei's next developer conference — the least glamorous number of the three, and the one that decides whether the five scenarios become one ecosystem or five separate products with the same logo.
The silicon, on the evidence of this week, is no longer the weakest link. Keep watching those three numbers, and it might soon stop being the story at all.
A phone chip is a statement about an entire supply chain. At the data-center end of the same industry, NVIDIA's Blackwell B200 superchips are dictating how enterprise facilities handle power and cooling, while the interconnect problem is being attacked with light itself — silicon photonics chips shattering bandwidth limits in ultra-scale data centers. Huawei's pitch is that it can build its side of that stack without any of it.
Sources: Dagang News and Sina Hong Kong, both carrying Huawei's media roundtable release of October 3, 2026.
