Apples Critical Weakness

TITLE: Apple’s critical weakness CHANNEL: Kiraa DATE: 2026-04-05 ---TRANSCRIPT--- I’ve said repeatedly on this channel that Apple will be the eventual winner of the AI race, even if that’s not obvious today. I know a lot of people disagree with that view because unlike other tech companies, Apple is not pushing the frontier models or pouring billions dollars into cloud infrastructure. But the truth is that Apple doesn’t need to be good at those things. Frontier models and cloud infrastructure will, at least in my view, end up becoming commodity resources which the market will provide. Apple instead focuses on something far more valuable, which is the relationship with the user. But the criticism about Apple is not completely without merit because there are some pretty significant software gaps which hold Apple back. The first one, which I’ve already covered, is CUDA, which is a language that powers the Nvidia GPUs. The developers who learn to use CUDA to build games or AI models are highly unlikely to switch over to Apple hardware because it means learning a whole new language. Now, of course, Apple has developed its own metal framework which addresses the gap and, in my view, is arguably easier, but it’s still a major barrier for developers. But there is one other which I think is a much bigger deal, and that’s Python. And if Apple can address the Python issue, it will open up a whole new set of of applications which could give Apple a superpower in high-end enterprise use cases. Apple silicon is, at least on paper, the most elegant architecture for serious analytical workflows. It has unified memory which makes data available seamlessly for CPU, GPU, and the neural engine. And for data science, the world that I live in, that should be a game changer. And yet almost nobody in the data science world takes Apple seriously. Now, to be fair, Apple’s MLX framework is a great step in the right direction, but it’s focused on machine learning and training and inference, not about the bread and butter data wrangling and analytical processing that takes up most of a data scientist’s actual day. So, in this video, I want to walk through why Apple, should at least in my view, be targeting Python to further strengthen its lead in the world of AI. In a traditional data science setup, you’ve got a CPU doing general computation and a GPU bolted on separately for parallel workloads. And then you got data constantly shuttling backwards and forwards. And of course, every time data moves between CPU and GPU, you’re waiting. And that transfer time adds up really quickly when you’re processing millions of rows of financial data. Of course, Apple silicon doesn’t have that problem because the CPU, GPU, and neural engine all share the same unified memory pool. And on top of that, you can get a Mac Studio with 192 gig of unified memory without spending tens of thousands of dollars on specialized hardware. So, the architecture is genuinely superior for analytical workflows. So, the question is, why aren’t people in business using it? And for me, the answer is one word, Python. And that’s because the entire data science ecosystem runs on Python. Python has become the lingua franca of the AI world. But Python has a big problem, it’s painfully slow. It’s painfully slow because it’s an interpreted language. It was never designed for performance. It was designed for readability and ease of use, which is fine if you’re writing a web scraper, but not fine if you’re processing 10 million financial transactions looking for anomalies. Python’s pretty flexible and easy enough to learn, and there are hundreds of thousands of libraries available that allow developers to build systems quickly. But the data science community knows its limitation, and they’ve worked around it by creating libraries that are actually written in very powerful languages like C and Fortran under the hood. These languages are really sophisticated and hard to write, so by putting a Python wrapper around them makes them really easy to use. This, at least in theory, reassures people that you know, Python’s not so bad, it’s like the best of both worlds. But even with those optimizations, you’re still dealing with the Python overhead. Python is not designed to run on multiple cores. Now, if you’ve got a basic Windows laptop, that probably doesn’t really matter. Python’s like a small car driving on a highway between cities. You’ll get there eventually, it’s fine. But if you’ve got a MacBook with an M-series chip, your machine is barely even breaking a sweat when it’s running Python. It’s like buying a Lamborghini and never taking it out of first gear. Now, for those of you who aren’t software developers, think of cores like pistons in an engine. A Mac Studio has dozens of pistons, they’re all designed to fire together. But Python can really only fire one piston at a time. So, you bought a team of V12 engine, and the software is running like a single cylinder lawnmower. Now, the real barrier to Apple dominating data science isn’t the hardware because we know that Apple solved the hardware issue years ago. The real barrier is the libraries that everyone depends on, and they’re locked into a language that can’t take advantage of what Apple built. But that’s where Swift comes in. Swift is a compiled language, which makes it almost as powerful as C without all the complexity. And it’s designed to work natively with Apple’s hardware with direct access to metal for GPU acceleration and the neural engine for machine learning workloads. So, when you write Swift code, you’re interacting directly with the chip, which means as a developer, you’ve got an enormous amount of power available at your fingertips which you can tweak and optimize as you need. Now, I’m not just theorizing here. Uh with my team, we’ve been developing Swift native data science libraries to specifically prove what’s possible when you eliminate Python from the equation and let Apple’s hardware do what it was designed to do. And to be honest, the results are staggering. And to show you what I mean, I built a small iPad application to give you a direct comparison. So, on one side, I’ve got the results from running Python on my Mac M4 laptop, and on the other, I’m running a metal GPU-enhanced Swift application over the same data. I’ve already run the Python side on my laptop, and now I’m running the same functionality using Swift with metal acceleration. I’ve designed this deliberately as an iPad application because it turns out that running Python on an iPad can be quite a challenge. So, what I’ve done instead is rewrite the functionality of the Python script, and it turns out that Swift was more than 80 times faster. Now, this is an extreme example, and of course, if you understand Python well, you can do all sorts of clever optimizations for specific tasks. But one thing is for sure, it will never be quite as fast as Swift for an overall workflow. That’s because Swift takes advantage of the hardware and Python does not. So, to put that into terms that a CFO would understand, if a Python job takes 3 days to finish, the same job in Swift finishes in 1 hour. 3 days versus 1 hour. And that’s not an incremental improvement, that’s a completely different capability. With that kind of acceleration, jobs that were impractical can now become routine. It means real-time processing of data sets that Python users would have to batch overnight. It means that a single Mac Studio on your desk could replace a cluster of cloud instances that you’re paying for by the hour. So, if Swift on Apple silicon is so much faster, why isn’t everyone already using it? Well, because the libraries don’t exist yet. That’s the gap. Python has had decades of community development, thousands of contributors building critical libraries such as Pandas, NumPy, and building out the entire analytical tool kit that data scientists rely on every day. Swift has almost none of that for data science, and that’s the problem I’m working on solving. For Kira, I’ve had to build a lot of these libraries myself, and there’s a few of these that I’m thinking about making open source. The one I’m most keen about is an Apple native port of of Pandas, which is the data science libraries that most data scientists rely on. I’ve rebuilt those using the low-level C libraries, but instead linked that with metal GPU acceleration. It’s still early, and there’s a long way to go, but the performance results demonstrate something that I think people should take seriously. But contrary to what some of the comments think, I actually have no relationship with Apple. I don’t even know anyone who currently works there, and I have no special access or insights into their plans. But if I did have a chance to speak to someone inside Apple, I would tell them there’s a massive opportunity for Apple to take a leading position in the world of data science. This hardware is so good that it would make a meaningful difference in the speed at which data can be analyzed, and that would be an absolute game changer for the world of business analytics. This is something, in my view, Apple should support and encourage. Addressing a few of these critical Python libraries would make it significantly easier for developers to build high-performance data science applications using Swift metal, which, as you can see, offers some seriously impressive performance gains. This would open up a whole world of iPad native applications which people could use on the go, out in the field. It seems so strange to me that Apple has all these great video and audio applications which run extremely well on Apple hardware, and yet nobody is using it for data science. Even if we could build just a fraction of what Python has running natively on Apple silicon, the performance differential is so large that it changes the economics of processing data entirely. Smaller companies using local hardware instead of renting cloud compute. Real-time analytics instead of overnight batch jobs. And deterministic, auditable processing instead of agents running amok and spitting out rubbish. Apple has the best hardware for this work, and Swift is the right language to unlock it. The only thing missing is the ecosystem. And ecosystems are built by people who see the opportunity and start writing the code. So, if you’re a Swift developer interested in data science, or a data scientist frustrated by Python’s limitations, drop a comment. I’d love to hear from you. And once again, thank you for watching my videos. I’ve now opened up memberships to supporters who like what we do and would be happy to buy me a coffee. Thank you again, and see you in the next one.