September 9, 2026 · Industry Watch · Apple’s next decade under hardware leadership
1. Core Thesis
- Fact: Apple’s new CEO John Ternus is a hardware engineer by training (the “president” everyone mentions is the CEO). He took office on September 1, 2026. Simultaneously, Johny Srouji was elevated to Chief Hardware Officer, putting silicon and hardware engineering under a single commander for the first time.
- Direction: Apple is betting on “on-device AI + hardware endpoints.” The center of gravity in the AI race is shifting from cloud-based large models to whoever owns the hardware in users’ hands. Apple is not fighting on cloud parameter counts; it is fighting on the ground it knows best.
- Industry: This is the clearest signal AI hardware companies have had in nearly a decade. On-device AI silicon, AI glasses, AIoT, sensors and foldable supply chains are entering a volume window, but opportunities will diverge sharply. Only those who lock in real positions will eat well.
2. The Facts: Who Are Apple’s New Leaders?
On September 1, 2026, John Ternus — a 25-year Apple hardware engineer — formally took over as CEO, with Tim Cook stepping into Executive Chairman. Shortly after, silicon veteran Srouji was elevated to Chief Hardware Officer. A US$4 trillion company just handed the wheel to two engineers. This is not a routine personnel change. It is Apple’s bet on the next decade of the AI era.
2.1 Key Timeline
| Date | Event |
|---|---|
| April 20, 2026 | Apple announces Ternus will take over as CEO on September 1; board approves unanimously; Cook moves to Executive Chairman |
| September 1, 2026 | Ternus formally takes office as Apple’s eighth CEO; former chairman Levinson becomes Lead Independent Director |
| September 9, 2026 | Ternus hosts the fall keynote as CEO for the first time: foldable iPhone debut and AI-powered Siri go live |
Source: Apple press releases (April 20 and September 2026) · Verified September 9, 2026
2.2 John Ternus: The 25-Year Hardware Veteran
Ternus joined the product design team in 2001 with a mechanical engineering background. He rose from VP of Hardware Engineering (2013) to SVP (2021), overseeing the hardware R&D of iPhone, iPad, Mac, Apple Watch, AirPods and Vision Pro. He led Apple’s full transition off Intel silicon to Apple Silicon and pushed material processes like recycled aluminum and 3D-printed titanium. Apple’s official assessment: “He has the mind of an engineer and the soul of an inventor.”
Tim Cook’s farewell note was even more direct: “Few people truly understand what it takes to build products that change the world — John is one of them.”
2.3 Johny Srouji: The Silicon Father Becomes “Chief Hardware Officer”
In the same week as the CEO transition, Apple did something else: it elevated Srouji (joined 2008, father of the A4 chip, former Intel and IBM executive) to Chief Hardware Officer, putting both the silicon team and the entire hardware engineering organization under one leader. This is the first time Apple has merged “chips” and “hardware” under a single commander. In short, Apple intends to treat silicon and product as one thing.
Strictly speaking, Apple has no “president” role — when people refer to the “new president,” they mean new CEO Ternus. Chief Hardware Officer Srouji is a separate move. Two seats, two technologists from the hardware side — that is the real story of this transition.
3. Why Hardware People? What Is Apple Betting On?
3.1 The Cook Era: A Mount Everest
- Market cap: roughly US$350B to US$4T, a gain of more than 10x
- Annual revenue: from US$108B in FY2011 to US$416B in FY2025, nearly 4x
- Active installed base over 2.5 billion devices across 200+ countries
But the Cook era also left a public weakness: generative AI. Apple’s internal assessment is that it lags Google and OpenAI by roughly two years on generative AI. Siri has been a long-running punchline. Apple Intelligence launched in 2024, and China-region users only got the filing-approved version in 2026. A company that built its reputation on “experience” became the chaser in AI experience.
3.2 The Second Half of the AI Race Is Played on Hardware
The AI race has two halves. The first half was about cloud models: parameters, compute, data centers — Google’s, OpenAI’s, Microsoft’s turf. Apple would burn capital chasing there and still not win. The second half is different: large models no longer live only in servers; they need to “grow eyes, ears, hands and feet” and into the physical world — phones, glasses, earbuds, watches, cars, robots. Whoever builds these devices controls the AI entry point; whoever controls the entry point owns the user data; whoever owns the data trains better models, ships better experiences, sells more hardware. Once that loop spins, it is a moat no one can take away.
That is exactly where Apple’s assets sit: 2.5 billion active devices, in-house A-series and M-series silicon, a closed but smooth ecosystem. The board’s choice, in plain language: stop tailgating on someone else’s track. Go back to the ground you know best and rebuild the fortress.
If we compress this transition into one sentence: for the past fifteen years Apple became “the most profitable” through supply chain and operations; for the next ten years Apple wants to become “the most leading” through silicon and hardware. It is a shift from operations-driven back to product- and deep-tech-driven.
4. What Will Apple Do in the Next 3–10 Years?
A roadmap that explains the Ternus era: no more incremental drip; a dozen new product lines at once.
| Phase | Key Moves |
|---|---|
| 2026 — Opening Year | First foldable iPhone (from ~US$2,099, 2nm A20 Pro, 5.5-inch outer + 7.8-inch inner display); iPhone 18 Pro series (2nm, variable aperture, US$150–200 price increase); release cadence restructured — fall for premium only, standard model pushed to spring 2027; AI Siri launches with iOS 27; Apple Intelligence rolls out in China (with Alibaba Qwen and Baidu); M6 silicon (world’s first 2nm PC chip, already shipping with Mac mini in August) |
| 2027 — Category Expansion | AI glasses (codename N50, positioned as an AI sidekick to iPhone); AirPods with cameras (giving Siri “eyes”); smart home hub, touchscreen MacBook and other new products; M7 silicon family (from H1 2027, memory bandwidth up to 240GB/s) |
| 2028–2029 — Depth | M7 Ultra: up to 1.5TB unified memory, targeting trillion-parameter models fully resident on-device; in-house AI server silicon (the “Baltra” chip with Broadcom hits volume in 2026, server variant ready ~2029); desktop robots, AR glasses and other new form-factor experiments; one chip architecture stitching together an “edge–fog–cloud” AI closed loop |
Source: Bloomberg’s Mark Gurman, Apple press releases and public reports · September 2026 · Not a committed schedule
5. Apple’s AI Strategy Dissected: On-Device First, Cloud as “Plug-in”
Apple is not chasing trillion-parameter bragging rights. It splits AI into three layers and lets each layer do its own job.
5.1 Three-Layer Architecture
5.1.1 Layer 1: On-Device Models (Daily Tasks · Fully Local)
Apple’s AFM Core model (a 20B-parameter sparse model that activates only 1B–4B at a time) runs directly on iPhone and Mac silicon. Voice, summarization, photo edits and other everyday tasks never leave the device. This is Apple’s privacy banner — and the reason on-device compute exists.
5.1.2 Layer 2: Private Cloud (Mid-Tasks · Self-Controlled)
Mid-complexity requests are routed to Apple-built Apple Silicon servers — a private cloud that is auditable and does not pass data to third parties. No Google involved.
5.1.3 Layer 3: Cloud Pro (Heavy Tasks · Borrowed Compute)
The heaviest inference runs on Google Cloud’s GPUs using Google’s Gemini (~1.2T parameters) “distilled” by Apple. Apple pays roughly US$1B per year for this license. Apple is not grinding it out on the cloud — it is renting brains.
5.2 Three Key Moves
- Silicon roadmap goes “All in AI”: Apple unusually cut the high-end M6 SKUs (Pro/Max/Ultra) and concentrated resources straight onto M7. M7 is designed as “an AI-first chip” — the Neural Engine is overhauled and memory bandwidth is pushed to 240GB/s. Industry analysts called this the clearest signal yet that AI has replaced CPU and graphics as the organizing principle of Apple’s silicon roadmap.
- Cash starts flowing toward AI: Apple abandoned its “net cash neutral” target and freed hundreds of billions for AI R&D and M&A. It plans to triple the AI R&D team by 2027 and is already in talks with silicon startups about acquisitions.
- Siri reborn + ecosystem openness: The new Siri jumps from “voice assistant” to system-level AI agent — it can see the screen, chain tasks across apps, and book tickets or sync calendars by itself. In an unprecedented move, Siri also lets users switch to third-party models like Gemini, Claude and ChatGPT. Apple has shifted from “closed in-house” to “open platform,” admitting it cannot win on every dimension.
Apple’s play, in plain language: grow your own crop on your own land (on-device silicon and models), buy what you cannot grow (Google Cloud), but never let the seeds (silicon) out of your hand. Short term, rent to patch weaknesses; long term, silicon and hardware build the moat.
6. The AI Hardware Industry: The Window Is Already Open
Apple’s leadership change is not just about Apple. It is the starting gun for the entire on-device AI supply chain. Look at the numbers already moving. In Q1 2026, cloud AI chip shipments grew only 12% year-over-year, while on-device and edge AI chips grew 45%, and IoT lightweight on-device chips surged 110%. Money is flowing from cloud toward device. As one of the world’s largest end-hardware companies, Apple personally betting on on-device AI is the heaviest weight it can put on this track.
On volume, China’s AI phones will grow from 118M units in 2025 (40.7% penetration) to a projected 147M in 2026 (53% penetration) — the crossing-50% inflection has arrived. In dollars, the global AI terminal SoC market is projected to grow from ~US$45B in 2025 to US$143.8B in 2030, ~23.6% CAGR. Both numbers tell the same story: AI capability is moving from “flagship selling point” to “mid-range default,” and hardware value is being structurally raised.
By category, AI PC penetration (global, 2026) is expected to reach 55%, AI phone penetration (China, 2026) at 53%; AI glasses are crossing from millions to tens of millions (2026 global shipments forecast at 15–23.68M units). After Apple enters the foldable category, it is expected to take 25% global share in 2026, second only to Samsung (~32%). Every new category that opens up means new demand for silicon, sensors, optics and structural components.
6.1 Three “Magnitude” Numbers
- 147M units: China’s 2026 AI phone shipments (+31.6% YoY), penetration up from 40.7% to 53%
- 30%–50%: the price uplift of flagship phone SoCs versus non-AI era — NPU has gone from “optional” to “standard”
- 23.68M units: 2026 global AI glasses shipment forecast (IDC) — the fastest new category to jump from millions to tens of millions
7. Supply Chain Opportunity Map: Who Is Locked In, Who Gets Paid
Apple’s entry opens four layers of opportunity simultaneously: on-device AI silicon, AIoT, new form-factor hardware and upstream supply chain.
| Segment | Opportunity | Key Players (public information) |
|---|---|---|
| On-Device AI SoC & AIoT Silicon | The highest-certainty golden segment: NPU becomes standard, SoC ASP rises 30–50%, 8–10 year lifecycle; mid-range compute (US$20–40) is the volume sweet spot | Apple (in-house A/M series), Qualcomm, MediaTek, HiSilicon; China on-device: Rockchip (2025 revenue RMB 4.4B, +40.4%), Amlogic, Bestechnic, Espressif, Allwinner |
| New AI Form-Factor Hardware | The most elastic: AI glasses (15–23.68M units), AI toys (global 2030 forecast US$35.1B), AI speakers, desktop robots — 0-to-1 incremental markets | Meta (Ray-Ban main driver), Xiaomi, RayNeo, Rokid; ByteDance (Xianyanbao), BubblePal sparked the AI toy chain |
| Compute & Memory | High-bandwidth memory becomes the bottleneck: M7 targets 240GB/s bandwidth and 1.5TB unified memory; on-device LLMs drive memory and storage upgrades | TSMC (2nm capacity), Broadcom (Apple Baltra server chip partnership through 2031), Samsung/SK Hynix/Micron (HBM and LPDDR) |
| Sensing & Structural | Cameras (including cameras on earbuds), sensors, mechanical apertures, hinges (liquid metal), titanium structural parts, batteries (foldables 5400–5800mAh dual-cell) | Apple in-house + Chinese supply chain (optics, precision structural, hinge makers) |
| Domestic Substitution | Premium on-device AI silicon domestic share is below 20%; policy + security + supply chain drivers could lift it from 20% toward 50% in 2026–2028 | NPU IP and customization: VeriSilicon etc.; ADAS silicon: Horizon Robotics (Journey 6 covers 10–560 TOPS) |
Source: public information (IDC, JW Insights, corporate filings etc.) · September 2026
Opportunity verdict (de-AI’d version): The steadiest is on-device AI silicon and AIoT — volume and price both rising, demand is inelastic, like selling “infrastructure materials.” The most exciting is AI glasses and AI toys — these are 0-to-1 categories; if you bet right, it’s a double click. The most demanding is domestic substitution — large space but requires real capability; a rebound off a low base is not the same as high-quality growth.