LEDE ON
Need a state-of-the-art semiconductor produced for your camera? Due to the process size race, there are only three companies that can really make super advanced chips for you: TSMC, Samsung, and Intel. Worse news? 70% of those advanced chips are currently being produced by TSMC (source: Taipei Times), and according to sources their fab orders are already full through the end of 2028.
If you don’t need super small process size, then SMIC (China), United Microelectronics (Taiwan), GlobalFoundaries (US), Huang Group (China), Tower Semiconductor (Japan), Vanguard International (Taiwan), Powerchip Semiconductor (Taiwan), or NexChip (China) may be able to help you, but they, too, have little extra fab utilization they can provide in the near term.
So how does this impact camera companies? Specifically in the BIONZ, DIGIC, or EXPEED. As we demand more intelligence and performance in the camera, these chips are becoming to cameras as Apple Silicon is to the iPhone, iPad, and iMac, i.e. critical and requiring smaller process sizes. Meanwhile, the image sensor fabs are also running at or near capacity, though this is mostly for mobile, auto, and security clients.
If you miss a step and need to redesign a processor or image sensor, you pay the price: you’re at the whim of the overscheduled fabs, and the fab folk aren’t paying a lot of attention to the dedicated camera industry. Even Canon with their own image sensor fab has the issue that they can’t just put new sensors into production without impacting other models. I’m aware of three products now that have hit this redesign and fab capacity problem and have gotten delayed.
——————————
Commentary
What’s Inevitable?
One of my jobs for many years (and to some degree still today) was looking five and ten years out. In the past I’ve mentioned that this was mostly looking at new technologies in research and labs but not yet in production, estimating when they’d be ready for production, and also figuring out the user problem they could solve that wasn’t currently solved. I did this both for software and for hardware during my Silicon Valley career, and I’ve proven pretty prescient in that respect.
But there’s another aspect to this that doesn’t get talked about much, and that’s what happens on the iteration cycle, and why. Generally, that too, tries to solve problems, but not always user problems, as often iteration mostly solves manufacturer or cost problems.
For example, when I predicted well over a decade ago that mirrorless would take over from DSLRs—my original prediction date missed the mark by 12 months, a 10% miss—it was not really because such cameras solved user problems (though they do), but instead because it solved manufacturing and engineering problems. I saw mirrorless as inevitable, regardless as to whether any big user problem got solved, because the Japanese camera companies need constant improvement in cost of production to keep prices in a realm where they still have a consumer market (typically US$1000 to US$3000 these days, though that price range was lower ten years ago).
Mirrorless eradicated parts (e.g. mirrors, prism, and eventually shutters). Mirrorless eradicated many manufacturing alignment steps. Those two things alone were huge cost savings and thus were inevitable.
So the question you should be asking these days is what’s still inevitable in the camera-making industry? I obviously have a few thoughts about that, so let me throw out some likely possibilities:
- Mechanical shutters disappear. This has already started (with the Z9), but it will become more and more prevalent, as it means fewer parts and fewer manufacturing processes, plus fewer mechanical parts that can wear out. Parts and process reduction means lower cost to produce. Eliminating mechanical process is a veiled user benefit (camera less likely to fail). But getting rid of shutters is somewhat tied to another inevitability:
- Rolling shutter disappears. The Holy Grail here is global shutter. The reason why that’s a goal and inevitable is that it solves a bunch of hairy and problematic internal timing and data management problems. “Here’s the data” is a lot easier to deal with then “here’s some of the data, more coming.” Sony moved first here with the A9 III, and this has been a “win” for a limited client set. Eventually it will be a win for everyone.
- Chips disappear. We’ve seen attempts at this with some stacking of memory or partial processing behind the image sensor, but ultimately it’s likely that we end up with a single chip doing image sensing, memory handling, and all sorts of processing (i.e. Stacked, but well beyond the current implementations). This one is tricky, as it’s a huge investment to do in a small market. However, the ultimate payoffs of doing so are going to attract one of the companies to pursue this approach, and then all will have to follow. Sony is probably best positioned to do this, but don’t count out a surprise competitor.
- Viewfinders disappear. This will be a highly contentious thing because tunnel view is so cherished by those that grew up with it. But there’s a series of intersecting things happening behind the scenes now that make traditional viewfinders' disappearance inevitable. First, consider this (very real and currently being pursued) possibility: your mobile phone gets smaller by dropping the screen, and the UI for it transitions to glasses. I’ve been playing with two AI+screen glasses lately and it’s pretty clear to me that this will be (one of) the way(s) we work in the future. I can already do things I didn’t used to be able to do, and the young will almost certainly instantly gravitate to this idea. Moreover, in the meantime, cameras such as the ZR show that a really, really good LCD is perfectly acceptable, and cameras such as the Insta360 Luna show that having that LCD detach opens up new possibilities, too.
I could go on (and probably will at some point), but I’ve kept the list simple and focused here. The key word is “disappear.” The more parts and processes that the camera companies can remove in the creation of a new camera for you, the more they can keep the cost of producing products down, which both benefits the camera makers as well as the users (today’s US$2000 cameras are way more simple to produce than those of 20 years ago, plus they do more, are arguably better at what they do, don’t need as much repair, and so on.
The challenge for the Japanese companies is that the Chinese companies are skipping steps and getting “there” faster. The DJI and Insta360 progression that has rendered GoPro’s dominance a thing of the past and put GoPro’s entire business in jeopardy could be repeated by companies such as Viltrox eventually entering the camera market. A new entrant such as Viltrox would have no skin in the game for parts and designs that are in current use; they could simply skip to the inevitable (and my advice to them would be to do just that).
——————————
Commentary
Not Understanding the Numbers
I continue to get attacked on forums by people who don’t actually look at numbers, let alone interpret them well. One recent repeated comment has been that Nikon doesn’t really sell many APS-C (DX) cameras. Yet Nikon’s own financial statements point out that one of the reasons why the Imaging division's overall revenue and profit lagged a bit last year was because of two specific cameras (Z5II and Z50II), which sold at lower price points than the cameras introduced in previous years. So a low-cost DX camera—which was part of one person’s argument in forum discussions as they thought that Nikon wasn’t doing much in the low-cost market anymore—clearly rose its head and bit Nikon simply by being popular.
Here in the US I can put Z DX unit sales so far over the 200k unit volume mark. True, Z FX lifetime in the US is now over 500k, but one fifth of those predate the release of any DX camera, and that unit volume is spread over seven FX models (and multiple generations) instead of three DX (and only one DX model has had a second generation so far). Dealers tell me DX is still selling well, but again, there are only three DX models to sell, and two of them are getting long in the tooth.
No doubt that DX isn’t as important to Nikon as FX is, but that’s also self-realized. Simply put, since 2009 or so, Nikon has put most of their marketing and sales energy into FX, and they doubled down on that with the Z System. Sometimes you get what you seek ;~).
But here’s the thing: Nikon lives in a market share and unit volume no-man’s land at the moment. They’ve made no real in-roads into Canon and Sony in the past year, and at ~900k total units a year, they’re in Ries and Trout's “marginal third position.” New FX cameras aren’t going to increase the unit volume significantly now, no matter how good they are, and thus parts-sharing costs are stuck in a time warp for Nikon. One way out of that would be to come up with two or more desirable DX cameras. An updated Z30 or Zfc, plus a new higher-end unit would solve a lot of problems for Nikon right now.
As I reported earlier, CIPA’s 2026 numbers so far show full frame mirrorless is in a significant downturn for volume and a modest drop in dollars (yen) brought in the door. I don’t think any of the Canikony trio is going to solve their current dilemma with “more great full frame cameras.” They’ve sucked the dollars out of that clientele and there’s clear Update Fatigue now. The customer target I see not getting everything they’d like is the young and entry market, and they're much more price sensitive, and thus more likely to consider DX (APS-C).
Finally, many seem to argue that “DX is only a portion of Nikon’s sales, so it’s not important to create new DX models.” In the overall market, though, full frame (FX) volume is only about 60% of smaller sensor volume right now [source: CIPA]. Nikon is losing traction in that smaller sensor market, and this also explains their problem in gaining market share. Nikon needs additional (or updated) smaller sensor products to regain a footing, particularly against Fujifilm, which is the number four competitor in terms of market share and trying to unseat Nikon.
——————————
Commentary
AI as UI
Adobe’s announcement this week of adding an AI agent that interacts with products such as Photoshop is another example of something I call AI as UI. Instead of using tools directly, you use them indirectly via chat (or possibly voice chat). Want to remove a background, then you type “remove the background” instead of finding the “remove background” button or menu choice.
I call this backseat driving, because when the result isn’t exactly what you wanted, you’ll need to be “more chatty.” Adobe points out that you can string tasks together, as in “make this portrait my style but substitute a plain background.” My experience with AI and LLMs is that the more you try to string together, the more likely you spend longer doing something, because the devil is in the details, and having to consistently explain your intention makes it incumbent on you to make that clear and precise. Moreover, when you start using standard terms, such as make it brighter, AI always devolves towards the mean, not the (creative) exception. I actually experiment a lot when I process images; I’m not trying to make them look all the same. That said, I can see where someone who does production photography and wants “sameness” might gravitate towards what essentially become AI-driven macros.
The “win” for Adobe is another product they can sell (the AI assistant is part of Firefly, which is another US$10/month for the minimum AI credits and model access, more for more). The “win” for you? I’m not sure. We’ve moved to an SaaS (software as a service) model for the most part, and now we’re moving towards SaaS2 (service as a service), where if you want to do something you’ll pay someone for the privilege.
While I’m a proponent of AI (used correctly and run locally), I’m not a fan of the behemoth tech companies trying to elbow each other to be the first “we covered the planet in servers” winner. xAI (SpaceX), Adobe, Amazon, Anthropic, Google, Microsoft, Nvidia, and OpenAI have the biggest elbows right now, and are jabbing away at each other (Apple is more close to my position).
Wait, you didn’t know you could run AI locally? If you have a recent Apple Silicon Mac with more than the base RAM, yes, you can, and in many different forms (including Apple’s own upcoming on-device one in macOS Golden Gate). Personally, I was vibe programming in Botswana last month with no Internet connection, all local on my MacBook Pro 14”. Yes, there were a few times where I’d rather have had “more power” from a server farm, but quite frankly, my little MacBook Pro provides most of what I need an AI for these days. And for free (well, okay, the MacBook Pro M5 with 64GB RAM wasn’t free, but I use that for other things, too ;~).
Adobe wants you to buy their agent and AI credits to tell their software what to do. Maybe we should instead ask them to just make it easier for us to do what we want in the software. If “remove the background” via AI UI is useful to some, it’s because Photoshop is now complex enough that a number one support question might be “how can I remove the background?”
I’ve been a proponent of UI that follows Alan Cooper’s direct manipulation strategy since he first described it to me in the late 1980’s (see his books About Face and The Inmates Are Running the Asylum). AI as UI is not particularly direct, even though it might feel that way to some. As they might say in parts of the US, “I’m agin it.”