iPhone 18 Pro, iPhone Duo, and the AI Problem Apple Still Hasn't Solved

Yesterday Apple pushed Siri AI into public beta. This morning it cut trade-in values on every iPhone it accepts. Between those two events sits the September 9 keynote, the first one John Ternus has run as CEO, where Apple announced the iPhone 18 Pro, the iPhone 18 Pro Max, and the foldable it calls iPhone Duo.

I build AI products for a living and I have carried an iPhone since the 3G. So I watched this one with two questions in mind. First, is the hardware worth the money? Second, after two years of promises, does Apple finally have an AI story that is its own? The answer to the first is mostly yes. The answer to the second is still no, and the reasons are more structural than a single delayed feature.

This is the long version: the phones, the foldable's trade-offs, the trade-in math, and a section on why the richest hardware company on earth keeps showing up to the AI fight with someone else's model.

iPhone 18 Pro: the best iterative update Apple has made in years

Nobody is going to call the iPhone 18 Pro a redesign, and some reviewers have already used the word "boring." I think that undersells it. The changes are inside, and they are the kind that matter for the next three years of owning the thing.

The headline is the A20 Pro, the first Apple chip on a 2-nanometer process. Apple claims the six-core CPU is up to 20 percent faster than the A19 Pro and the seven-core GPU up to 40 percent faster. The number I care about is the Neural Engine: a dual 16-core design with roughly twice the AI compute of last year, plus 50 percent more memory bandwidth. RAM stays at 12GB. That combination tells you what the chip was built for. It is an on-device inference machine first, and a gaming chip second.

The other quiet change is connectivity. Most iPhone 18 Pro models use Apple's own C2 modem, which finally adds mmWave 5G in the US and, by Apple's claim, uses about 15 percent less power than the C1X. The exception is the US Pro Max, which still ships with a Qualcomm Snapdragon X80. Alongside it sits the N1 chip for Wi-Fi 7, Bluetooth 6, and Thread. Apple now designs the CPU, GPU, Neural Engine, modem, and wireless radio in its flagship phone. Keep that in mind for the AI section, because it is the part that makes Apple's software position so strange.

The camera is where the money went. The 48MP main sensor is larger, and for the first time it has a mechanical variable aperture, six laser-cut blades stepping between f/1.48 and f/4.0. Apple says it gathers around 50 percent more light in low-light conditions. The telephoto stays at 48MP with 4x optical and a 12MP 8x crop, and the front camera moves to 18MP. Pro controls now expose aperture, shutter speed, and white balance directly, which photographers have asked for since the first Pro model.

The rest of the spec sheet, quickly:

• 6.3-inch and 6.9-inch ProMotion OLED panels, 3,000 nits outdoors, a smaller Dynamic Island that can show three Live Activities.
• Up to 36 hours of video playback on the Pro and 45 on the Pro Max, with 50 percent charge in about 15 minutes on a 60W adapter. The best battery figures require the eSIM-only configuration.
• Weight went up: 211 grams for the Pro and 249 for the Pro Max.
• Colors are Black, Silver, Burgundy, and Glacier. Storage starts at 256GB and now goes to 2TB even on the smaller model.
• $1,199 for the Pro and $1,299 for the Pro Max. Pre-orders opened September 12 and the phones land in stores this Friday, September 18.

My take: if you are on an iPhone 15 Pro or older, this is a clear upgrade. If you have a 17 Pro, the only reasons to move are the variable aperture and the Neural Engine, and the second one only matters if Apple ships software that uses it. More on that below.

iPhone Duo: Apple's first foldable, and what it gives up

Ternus called the Duo "the most transformational change to iPhone since the original." That is keynote language, but the product underneath it is more careful than the quote suggests.

The Duo is a book-style foldable. Closed, it has a 5.4-inch outer display that Apple says gives you about 90 percent of the screen area of an iPhone 18 Pro. Open, you get a 7.6-inch inner display, roughly 80 percent larger than the 18 Pro's. Both panels are Super Retina XDR with ProMotion, Always-On, 3,000 nits, and a nano-texture finish, and they share the same aspect ratio so apps scale proportionally instead of reflowing. The inner panel uses a micro-lens texture made with maskless laser lithography, which is Apple's answer to the crease. Reviewers who have handled it say the crease is reduced, not gone.

The frame is Grade 5 titanium. The hinge has more than 100 precision components and a 3D-printed cover made from recycled titanium. The whole thing is IP68 to six meters, which is unusual for a foldable. Unfolded it is 5.2mm thick, which Apple is calling the thinnest iPhone ever. Folded it is 11.3mm, and it weighs 254 grams.

That weight is the first trade-off. Samsung's Galaxy Z Fold 8 is 201 grams and 4.5mm unfolded, and the Fold 8 Ultra is 215 grams and 4.1mm. The Duo is thicker and heavier than nearly every Android foldable on sale, and $100 more than the base Fold 8 at $1,999 for 256GB. Apple spent that mass on durability and on two batteries, one per half. The payoff is up to 31 hours of video on the inner display, 44 on the outer, and 24 when you split time between them. Charging is 50 percent in about 20 minutes wired.

The other trade-offs are the ones that will surprise people who expect a Pro phone that folds:

• No Face ID. Touch ID is built into the side button, and you can unlock with an Apple Watch. The inner FaceTime camera sits under the display.
• No telephoto. You get the 48MP Fusion main and the 48MP ultra wide from the 18 Pro, with a 2x optical-quality crop. No 4x, no 8x, no variable aperture.
• No Action button, and eSIM only worldwide.
• Same A20 Pro as the 18 Pro, with a custom vapor chamber that Apple says sustains 35 percent more performance than the 17 Pro.

Software is where Apple has clearly done real work. The Duo ships on iOS 27.1 with the first true Split View on an iPhone, a dock and Lock Screen pushed to the sides when open, a vertical Dynamic Island, a redesigned StandBy that works unplugged on either screen, and Apple Pencil support coming later over USB-C. Camera features like Duo Preview, which shows your subject the shot on the outer screen while you frame it with the inner one, are the kind of thing only a two-screen phone can do.

Pre-orders open October 16 and it ships October 23 in Star White and Night Sky, with storage up to 2TB. Counterpoint thinks it could take up to a quarter of the foldable market by year end, which sounds dramatic until you remember foldables are under 2 percent of smartphone shipments. Apple is entering a small category late, at the top of its price range, with a heavier device that has fewer cameras. That is either a mistake or exactly the iPad playbook: arrive late, make the software feel finished, and let the market catch up. I lean toward the second, but not with my own $2,000, not in the first generation.

Macworld's line is the right one: buy the Duo because you want the folding experience, not because it is the best iPhone. The best iPhone is the 18 Pro Max.

Why is Apple still so weak in AI?

This is the section I actually wanted to write.

Rewind to June 2024. Apple stood on stage at WWDC and announced Apple Intelligence and a new Siri that understood personal context, could see your screen, and could take actions inside apps. It ran ads for it. It sold iPhone 16s on the back of it. That Siri did not ship in 2024. It did not ship in 2025. Apple pushed it beyond iOS 26.4 in February of this year because of, in the reports at the time, bugs it could not close.

What did ship yesterday, September 14, is called Siri AI, and it does most of what the 2024 demo promised: personal context across Messages, Mail, and Photos, on-screen awareness, actions inside third-party apps like WhatsApp and Outlook, a standalone Siri app with synced conversation history. It is a beta. It is English only until October. It is not available in the EU on iPhone, iPad, or Watch, and China is waiting on regulators. And the models behind it are, in Apple's own careful phrasing, custom-built in collaboration with Google's Gemini, under a deal reported in January at around $1 billion a year.

Read that again. The company that designs its own CPU, GPU, Neural Engine, modem, and radio, and that just put twice the AI compute into the A20 Pro, could not close the gap on the model itself. It bought the capability from the company it competes with in search, maps, and phones. When the WWDC keynote in June showed this off, the stock fell about 2 percent on the day, and KeyBanc's Brandon Nispel called the updates "lacking" with "no clear signs of monetization." I do not think the market was wrong.

So why? Here is my read, from someone who has built machine-learning teams inside startups and watched large companies try to do the same.

1. Apple optimized for the wrong constraint. Apple's AI strategy was built around privacy and on-device execution before it was built around capability. Those are good values, and Private Cloud Compute is genuinely clever engineering. But frontier models in 2024 and 2025 were won by whoever could train the largest models on the most compute, and Apple deliberately did not build that infrastructure. Google, OpenAI, Anthropic, and Meta each spent more on training compute in a year than Apple spent on its entire AI effort. You cannot distill a great small model from a mediocre large one, and Apple did not have a great large one.

2. It lost the people who could have fixed that. Ruoming Pang, who ran the Apple Foundation Models team, left for Meta in the summer of 2025. Roughly a dozen researchers and executives followed over the next six months, including several more from the foundation models group. John Giannandrea, the SVP who ran AI strategy since 2018, was moved aside in favor of Craig Federighi and retired this spring. Apple hired Amar Subramanya from Microsoft to rebuild, but you do not rebuild a frontier lab in nine months. A talent war is fought with mission and compute as much as money, and Apple was short on both.

3. The hardware and the software are on different clocks. The A20 Pro's Neural Engine is a real advantage. Nobody else ships that much on-device inference in a phone at Apple's volume. But silicon is a three-year pipeline and software is a three-month one, and Apple's model team could not keep up with its own chip team. That is why every iPhone since the 16 has been sold on AI features that arrived a year or two later, or not at all. Nispel's other complaint, that even the advanced features require an iPhone 17 or newer, is the same problem from the customer's side.

4. Apple's culture punishes shipping unfinished things. Google and OpenAI ship models that are wrong 10 percent of the time and fix them in public. Apple's entire brand is that the thing works when you take it out of the box. That instinct is right for phones and wrong for the current state of AI, where a product that is 80 percent right and improving weekly beats one that is 95 percent right and two years late. The 2024 Siri demo was Apple trying to do the Google thing. The two-year gap was Apple remembering it is Apple.

5. The keynote told on itself. The slide that stuck with me from September 9 was "2.5 billion requests per day." That is a Siri usage number. It is a distribution number, not a capability number. Apple has the largest installed base of any AI assistant on earth and it is leading with reach because it cannot lead with quality yet.

None of this means Apple loses. Apple has done this before. It bought its way into maps, into chips, into the modem. The pattern is arrive late, lease the capability, integrate so tightly the late start stops mattering, then internalize. The Gemini deal is the leasing step. The A20 Pro and the Duo's Split View and Live Rewind are the integration step. The question is whether Apple can do the last step with a model team that is a fraction of the size it was eighteen months ago, while paying a competitor a billion dollars a year for the privilege of catching up.

My prediction: Apple's on-device models will be excellent within two years, because that is a hardware-adjacent problem and Apple is good at those. Its cloud models will stay rented. And Siri will get most of the way to what was demoed in 2024, in 2027.

What to do with your old iPhone before Friday

Now the part that costs you money if you skip it.

This morning, September 15, Apple cut trade-in values on every iPhone in its program. The iPhone 16 Pro dropped from $630 to $510, a 19 percent cut. The 16 Pro Max fell $110 to $610, a month after Apple had raised it. The iPhone 17 Pro and 17 Pro Max appeared on the list for the first time at $785 and $885. This is not a one-off. Across the iPhone 15, 16, and 17 launches, about three quarters of models tracked lost value between announcement day and release day, and Apple's own program is usually the floor, not the ceiling.

The floor is the problem. Apple's number is convenient and it is store credit against a new phone. The independent buyback market is a different market, and the site I send people to is cashmycell.com. Full disclosure first: CashMyCell is a startup I have worked with closely, so I know the team and how the product is built, and you should weigh my recommendation with that in mind. It is also why I am comfortable putting my name next to it.

CashMyCell is a price comparison marketplace, not a buyer itself. It pulls live offers from multiple vetted buyback companies, shows them side by side, and lets you pick. When I checked while writing this, it listed an iPhone 16 Pro at up to $870 from third-party buyers, against Apple's $510 as of today. That is not a subtle gap. The site's own claim is that the same phone can differ by more than $100 between buyers, and the numbers above suggest that is conservative. Payment is real money over PayPal, Venmo, Zelle, or bank transfer, the buyer sends a free insured shipping kit, and payout is typically within 24 to 48 hours of the device passing inspection. Every buyer on the platform is vetted on background, BBB rating, and customer reviews before it is listed, which is the part I have seen up close and the reason I trust the offers on it. It also has a depreciation calculator that shows what your specific model has lost since launch, which is a useful reality check before you decide whether to sell at all.

The rule I follow, and the same rule CashMyCell gives on its own site: sell before the next generation ships, not after. The value cliff is between announcement and release. We are inside that window right now. The 18 Pro arrives Friday. The Duo pre-orders open October 16. If you are upgrading to either, get your quotes this week, lock the best one, and ship the old phone the day the new one arrives.

Three practical notes from doing this a dozen times:

1. Compare before you accept Apple's offer. Apple is a fine fallback and the safest option if you do not want to deal with shipping. It is rarely the best number.
2. Condition grading is where buyers make their margin. Be honest in the quote and photograph the phone before it ships. A locked quote from a vetted buyer with a real returns policy is worth more than a slightly higher one from someone you cannot find on the BBB.
3. Wipe it yourself. Sign out of iCloud, turn off Find My, erase all content and settings. Do not rely on the buyer's data-wiping process, even a good one.

The short version

• The iPhone 18 Pro is an iterative update with a genuinely new chip, camera, and modem. If you are on a 15 Pro or older, buy it. $1,199 and $1,299, in stores September 18.
• The iPhone Duo is a durable, heavy, well-built first foldable that gives up Face ID, the telephoto, and the Action button, and costs $1,999. Buy it for the form factor or wait for the second one. Ships October 23.
• Apple's AI is finally usable and still not Apple's. Siri AI is a beta, English-only, EU-excluded, and built with Google's models, two years after it was promised. The hardware is ahead of the software and the model team is smaller than it was. Expect on-device to get good and cloud to stay rented.
• Your old phone lost value this morning and loses more on Friday. Compare offers on cashmycell.com before you take Apple's number. I have worked closely with that team and I would bet on it.

I will revisit the Duo once I have used one for a month, and I will revisit Siri AI when it leaves beta. If Apple's own models close the gap before then, I will say so, and I will be glad to.