David Drinkwater on Tiny Chinese chipmaker teaches giant Apple a lesson - 'From my experience (and I have done wafer costing) the starting wafer is a small portion of the total cost of the die, so I don’t think there is much to make up there. The hit that CXMT takes is in their photolithography. The reason that they have to start 30% more wafers is that the individual device sizes (transistors, resistors, etc) that build up the finished functional device (a memory chip) are all larger: their photo tools can’t print the small features small enough. It literally takes more silicon surface area to make a device. Costing is evaluated on the basis of what it costs to process one wafer, and then divided by the number of die on the wafer. (This part is neither rocket surgery nor semiconductor design engineering, so I feel pretty safe explaining it, so long as I don’t include actual $$$ numbers.) I also feel safe saying that the actual silicon in one of my old projects was only 3% of the cost of the finished wafer. After that, we test and trim this particular wafer at the die level, adding another 50% to the cost of the finished die, reducing the silicon to only 2% of the total cost. So there is not a lot of water to be squeezed from that rock, the cost of the silicon starting wafer. I believe my neighbors to the north (Global Wafer is in Sherman. I work in Richardson, and live “even further south” in Dallas itself.) *only* produce the starting wafers. So they can only offer a very small piece of die cost reduction. Hope this helps.'
on How Apple wins AI's endgame - 'When it comes to flagship models — Claude Fable 5 (Max) and GPT 5.6 Sol (Max) — Claude is marginally more accurate according to the AA-Omniscience Accuracy benchmark. The scores stand at 61 percent and 59 percent, respectively. Because the difference is so marginal, you’ll rarely notice it in day-to-day usage. But, unless you’re tokenmaxxing, you’ll be using mid-tier models for most tasks. On Claude, this is Sonnet 5, and on ChatGPT, 5.6 Terra. Here, the scales are tipped in ChatGPT’s favor: ChatGPT 5.6 Terra (Max) scores 46 percent, whereas Claude Sonnet 5 (Max) is significantly lower at 38 percent. Another important factor when measuring accuracy is the tendency of the model to hallucinate. Ideally, if an LLM doesn’t know the answer to something, it should flat out refuse to answer it instead of making stuff up, i.e., hallucinating. The benchmark for this is AA-Omniscience Hallucination Rate, in which Claude has a significant leg up against ChatGPT. A lower score is better in this benchmark, and Claude’s Fable 5 model scores 55 percent compared to ChatGPT 5.6 Sol’s score of 89 percent. The difference is even more stark in the mid-tier models: Claude Sonnet 5 scores 37 percent, whereas ChatGPT 5.6 Terra scores 85 percent. Source: https://www.engadget.com/2229993/claude-vs-chatgpt-how-ai-assistants-differ/ My thought: And what surprises me more, is that people are actually -paying money- for this level of “quality”….'
on Saturday Apple video: Steve Jobs and the Apple story - 'This documentary looks great in the beginning but it turns into a highly biased hatchet job. I’d never heard of Bob Belleville before; he came from Xerox and worked on the Mac team for a short time. Yet he seems to have become the global expert on how Apple has devolved into a hideous “monster.” I wish one of these documentarians would stop reifying people with axes-to-grind, and dove instead into understanding the ethical issues. These issues (technology addiction, worker exploitation, workplace safety, and others) are very serious and important for the public to understand and engage in. IMHO, documentaries like this are sensationalistic, simplistic, and do a great disservice. Instead of helping make the world better in some way, they just spread anger and enmity.'
on How Apple wins AI's endgame - 'The problem for everyone is that AI is an insane business as it exists today. If it were based on real products and real revenue then the big five would not be spending like this. Customers are not knocking down doors trying to obtain AI. Most end users are skeptical about AI. It is the billionaires who think that AI will give them god-like powers who are spending like there is no tomorrow. They all feel that this is like a game of musical chairs with one winner and they don’t want to be the loser. This game will go on till they run into some sort of natural limits. Maybe the bond market will start charging them much higher interest rates. Maybe there is not enough water, coal and uranium to power their data centers. If NVIDIA stops expanding and just replaces existing product that will hurt the stock. A lot of this seems leveraged and once the balloon starts deflating it will be painful for some. NVIDIA will likely shrink, but not go away. There is always a market for high speed computation, just maybe not at the $5T level.'
on Apple-AI enthusiast Errol Brandt looks forward to Oracle's comeuppance - 'Mark: Thank you! I will reach out early in the week! I very much appreciate your participation in the Apple 3.0 comment stream.'
on Tim Bajarin's Mt. Rushmore of tech journalism - 'PED: Any chance you would share with us his longer list? I agree wholeheartedly with respect to Mossberg and Pogue; the other two had far less of an impact on me personally. Hopefully Jason Snell is on the longer list too! YMMV…'
on How Apple wins AI's endgame - 'It is indeed interesting how history tends to repeat itself. In the past, we had IBM main frames dominating, followed by the transitioning to Apple personal computers/iPhones. Now we have Nvidia dominating IA cloud computing, as we watch the transitioning to Apple desktop/iPhone AI computing. After a while, it gets confusing as to who’s actually the hare and who’s the turtle.'
on Omdia: Apple still rules the shrinking tablet market; Lenova is growing fast - 'Despite iPad shipments dropping 8% YoY, Apple is still comfortably ahead of the next three vendors (Samsung, Lenovo, Xiaomi) combined and more than double Samsung’s total. From Q3 FY2026 Apple earnings call for iPad: “iPad rev. was $6.2B, down 6% year-over-year, driven by continued difficult compare against launch of A16 powered iPad in the prior year. At the same time, iPad installed base reached a new all-time high, & over half of customers who purchased an iPad were new to the product.” 1. no new product introduction spike in quarter to spur sales. 2. as noted, > half are NEW iPad buyers who never had an iPad before, some users are new to tablets or Android tablet switcher or current iOS user buying a tablet for the first time are choosing iPads over competition even w/iPads priced in the premium price range. This grows iPad install base, as does iPad longevity, reliability & long term support of current & passed down iPads. 3. literally, > one out of every 3 tablet sales is an iPad, even in a declining tablet market. Worldwide. Think about that. 4. Average Sales Price (ASP) is important: Apple iPad $459 (Base iPad sells best but still decent iPad Air & Pro sales) Samsung $320-340 est. (Restricted cheaper A series tab sales -> prioirtized more expensive Tab S to keep profits up) Lenovo $140-175 (Aimed at & sold into international education market at high volume, low margin contracts. Lowest ASP of top 5) Huawei $260-310 (Strong sales in “luxury” tablet China domestic market, small int’l sales, reduced cheap tablet sales) Xiaomi $180-220 (Sub-$250 focus in Asia Pacific & China) An iPad sale ≠≠ an Android tablet sale, but iPad’s >>volume x >ASP = >> Revenue by far.'
on How Apple wins AI's endgame - 'I think people seem a little bit too eager to dismiss some of these potential threats.'
on How Apple wins AI's endgame - 'John, great observation. I wonder what the cost comparison is between the Nvidia and Apple solutions and when will they be available in meaningful numbers to foster widespread adoption?'
on How Apple wins AI's endgame - '” questions linger as to what could happen once custom silicon (think ASICs) looks to displace GPUs in the data centers of tomorrow.” Jensen may be play chess, but his primary opponent is a Grandmaster. ”Broadcom and Apple Baltric ASIC partnership “ Who is leading who in this two man race?'
on Tim Bajarin's Mt. Rushmore of tech journalism - 'That list seems ok for 20th century tech journalism. But, we are more than a quarter of the way through the 21th century, and things are a lot different. For the first place on the current list, I nominate PED as publisher of his own blog. I think he has created something really great, and quite different from anything in the 20th century.'
on How Apple wins AI's endgame - 'This caught my eye from the article – “But here’s my favorite piece of evidence, the tell I’d point to if I could only pick one: watch what NVIDIA sells now. The DGX Spark — Jensen’s new “personal AI supercomputer” for your desk — is a small quiet box with 128GB of unified memory shared between CPU and GPU. Sound familiar? It’s a Mac Studio. NVIDIA, the company whose entire empire is built on discrete GPUs with discrete VRAM in centralized racks, is now shipping Apple’s architecture, at Apple’s form factor, aimed at Apple’s use case — and even Jensen has started musing that “an AI supercomputer could become common in the home,” name-checking Apple’s M-series as laying the groundwork. The king of the datacenter is hedging into my thesis with his own product line. When your opponent starts copying your homework mid-argument, the argument is over.”'
on How Apple wins AI's endgame - 'Thanks for posting Ped. It’s worth reading the full piece. “If Apple has its way, hyperscaling as the default architecture of AI is dead. Not dead like it disappears — dead like the mainframe: still humming in the basement for training runs and big enterprise, while the actual life of computing happens somewhere else”'
on How Apple wins AI's endgame - 'I don’t think for one second that Nokia or BlackBerry RIM would agree with those views! And yes, Microsoft survives by serving corporates, but its valuation for consumer personal products must be near zero by now. Nvidia is a dead man walking, and quite possibly OpenAI is as well. Dead as in a shadow of their former selves.'
