🎙️ AI 日報 Podcast — Apple 開放 Siri、Arm 跨足自研晶片、OpenAI 關閉 Sora

Opening Remarks

Welcome to today's "Mark's Tech Insights" for March 27, 2026! The tech world is absolutely buzzing today. Apple has finally come to its senses and is opening up Siri, Arm is breaking a 35-year tradition to make its own chips, and most shockingly, OpenAI is shutting down the viral sensation Sora? Without further ado, let's dive right into today's top stories!

Today's Top Stories

1. Major Overhaul in Apple iOS 27: Siri Will Be Fully Opened to Support Third-Party AI Assistants

Source: Bloomberg (Original Article)
Summary: Apple is preparing for an epic overhaul of Siri in iOS 27. In the future, users will be able to call on other AI assistants like Claude and Gemini directly through Siri, no longer being locked into just ChatGPT. This means the iPhone will transform into an open AI platform, potentially taking a cut of AI subscription profits.
The View from Taiwan: App developers in Taiwan need to pay attention. The future traffic gateway might not be the App Store, but rather who can most seamlessly integrate their services as APIs into Siri's multimodal ecosystem.
Discussion Points:
- Is Apple admitting its own foundational models can't compete?
- Will the infamous "Apple Tax" (30% cut) be applied to all AI subscription services?
Suggested Script:

Imagine this: in the future, you'll say "Hey Siri" to your iPhone, and it won't just call ChatGPT—it can also summon Claude or Gemini to help you out! Apple is really making a huge move with iOS 27. It's actually quite clever; Apple knows it can't win by brute-forcing its own foundational models, so it's turning itself into the "portal for the AI world" instead. For us software developers in Taiwan, this is a strong signal that the future battlefield might not be the App Store, but who can plug their services into Siri. But I'm guessing that, deep down, Apple is still playing the long game with its commission cut, turning the iPhone into an AI tollbooth. A truly brilliant move!

2. Bombshell! Arm Launches Its First In-House AI Chip in 35 Years, Meta Snatches Up First Orders

Source: TechCrunch (Original Article)
Summary: Arm, which has always sold only silicon intellectual property (IP), has broken its 35-year tradition by personally launching the Arm AGI CPU, aimed at data center inference. Its performance-per-watt is double that of traditional x86, and Meta has already become its first major customer.
The View from Taiwan: Arm's shift from pure IP licensing to physical chip design will directly increase its wafer orders for TSMC's advanced processes. But at the same time, this puts IC design houses like MediaTek, which rely on the Arm architecture, in the delicate position of being both a customer and a competitor.
Discussion Points:
- Why is Arm breaking its unwritten rule of "not competing with customers" at this moment?
- Has chip power consumption become the biggest pain point in AI development?
Suggested Script:

This news is absolutely nuclear for Taiwan's semiconductor industry! For the past 35 years, Arm has just been selling blueprints to everyone, and now it's jumping in to make the chips itself—and Meta has already snapped up the first batch. What does this signal? It means the power consumption of AI compute is just terrifying, forcing Arm to get its own hands dirty to demonstrate how to build ultra-low-power CPUs. For Taiwan, this is a huge boon for TSMC, with endless orders on the horizon. But long-time customers like MediaTek are probably having mixed feelings right now. After all, their former "arms dealer" has now become a rival competing for the same pie in the data center battlefield.

3. A Major Strategic Pivot? OpenAI Shuts Down Sora Video Generation App, Diverts All Compute to Robotics

Source: Promethean AI (Original Article)
Summary: OpenAI has confirmed it will shut down the controversial video generation tool, Sora, and transfer all the massive computing resources freed up to its expanding robotics research program, signaling a major shift in its strategic focus.
The View from Taiwan: This is a huge boon for Taiwan's hardware supply chain! As OpenAI shifts its focus from virtual videos to physical robots, the demand for edge computing chips, precision servo motors, and sensors will explode—these are precisely the greatest strengths of Taiwan's manufacturing industry.
Discussion Points:
- Was Sora shut down because copyright issues were too difficult to solve, or simply because it was "burning cash and hard to monetize"?
- Is Embodied AI the only path to AGI?
Suggested Script:

Does everyone remember how mind-blowing Sora was when it first came out? Well, OpenAI just shut it down, just like that! They've confirmed they're discontinuing Sora and throwing all that freed-up, massive compute power into building physical robots. Personally, I think this isn't just about copyright issues. The harsher reality is that the business model for video generation currently can't sustain the astonishing amount of compute it burns through. But looking at it from another angle, this is fantastic news for Taiwan's tech industry! When a software giant like OpenAI decides to move into the physical world, the precision motors, edge computing, and sensors needed for AI robots—aren't all of those the strengths of our Taiwanese supply chain? The hardware dividend of the next decade might be right here!

4. AI Finally Grows Hands and Feet: Claude Can Now Automatically Click Your Mouse and Operate Your Computer

Source: Promethean AI (Original Article)
Summary: Anthropic officially rolls out its "computer control" feature to users. Claude can now autonomously control the mouse, click, scroll web pages, and even navigate the desktop to complete complex tasks, marking the entry of AI agents into real-world operating environments.
The View from Taiwan: For Taiwanese small and medium-sized enterprises eager for digital transformation, this will be the ultimate RPA (Robotic Process Automation) tool; but for corporate IT departments, managing the AI's operational permissions will become a brand-new cybersecurity challenge.
Discussion Points:
- Are we really ready to let AI completely take over our desktops?
- If an AI accidentally deletes the company database, who is held responsible?
Suggested Script:

Folks, AI has truly grown virtual hands and feet! The new feature just released by Anthropic lets Claude directly take over your mouse and keyboard, clicking and finding information on the screen by itself. It sounds super futuristic, but it's also a bit creepy, right? In the future, you'll just have to say, "Compile last month's report and send it to my boss," and it will open Excel, pull the data, and send the email all by itself. For many companies in Taiwan that still rely on manual labor for repetitive clerical work, this is a massive productivity explosion. But for all you friends in IT departments, your cybersecurity nightmare is about to begin. Figuring out how to make sure this AI doesn't "accidentally" click on a confidential file will be the biggest headache from now on.

5. You Still Need a Brain to Code! Cursor CEO Warns Developers Against Over-relying on "Vibe Coding"

Source: Cambridge Today / National Today (Original Article)
Summary: The CEO of the well-known AI coding tool Cursor has publicly warned that developers should not engage in "Vibe Coding"—that is, blindly using AI-generated code without understanding the underlying architecture. This leads to software that is fragile and breaks easily in production.
The View from Taiwan: Taiwan has a vast pool of software outsourcing and development talent. This news is a reminder to the industry: the value of senior engineers who "understand architecture" will only increase, while junior engineers who only know how to copy-paste prompts will face the risk of being replaced.
Discussion Points:
- In the age of AI, should engineering interviews focus more on "architectural design" rather than "rote memorization of algorithms"?
- How should team leads review project code that is largely generated by AI?
Suggested Script:

This news is super interesting—even the CEO of Cursor, arguably the most powerful AI coding tool out there, is pumping the brakes himself! He's warning everyone not to do "Vibe Coding," which is letting AI randomly generate code based on a "vibe" while you don't even understand the underlying logic yourself. I couldn't agree more! Right now, a lot of junior engineers in Taiwan, when they hit a bug, just mindlessly paste it to an AI to fix, and the project gets more and more convoluted. AI-assisted coding is fast, for sure, but if you don't have a solid foundation in architectural design, the stuff you build will absolutely crash the moment it goes live and gets some traffic. So, my fellow engineers, AI can do the typing for you, but you still have to provide the brain!

6. Open Source Model Breakthrough! Google's TurboQuant Slashes Large Model Memory Usage by 6x

Source: Open Source For You (Original Article)
Summary: Google has introduced TurboQuant, a technology that can extremely compress a language model's cache memory, reducing memory consumption by nearly 6x with almost no loss in quality. And it can be directly applied to existing open-source models without retraining.
The View from Taiwan: This is a key technology for driving the adoption of "AI PCs." PC brands in Taiwan, like ASUS and Acer, can now use standard-spec memory hardware to run powerful local AI models smoothly on users' laptops, significantly boosting their products' selling points.
Discussion Points:
- As memory optimization techniques continue to advance, will they weaken the market demand for high-end AI chips?
- In the future, will "Local AI" replace some cloud services due to its privacy advantages?
Suggested Script:

Lastly, let's talk about a more hardcore but super practical technical breakthrough. Google has introduced TurboQuant. Simply put, it has slashed the memory that large language models consume during operation by nearly 6 times, and it doesn't even require retraining! For our major PC brands in Taiwan that are pushing "AI PCs," this is like a gift from heaven. Before, to run a large model on a laptop, you'd need at least 32GB or 64GB of RAM, which is crazy expensive. Now, with this technology, a standard business laptop might be able to run a pretty powerful AI smoothly. For privacy and speed, AI computing is really going to shift more and more towards our personal devices!

Closing Paragraph

Today's AI news really covered everything from software and hardware to the developer ecosystem, especially seeing OpenAI pivot to physical robotics—it feels like we're one step closer to a sci-fi movie plot! If you enjoyed today's content, don't forget to subscribe and share it with your fellow tech enthusiasts. See you next time, bye!