🎙️ AI 日報 Podcast，AI裁員潮、開源模型爭霸、川普AI新政

Introduction

Hello everyone, welcome to "Mark's Tech Insights." I'm Mark. Today, we're going to talk about some highly impactful AI trends, including tech giants openly blaming layoffs on AI. Ironically, the latest research indicates these automation-driven layoffs haven't actually made companies any money yet. We'll also take a look at the latest geopolitical chess match between the US and China regarding open-source models and regulatory policies. It's an action-packed episode, so don't miss out.

Today's Top Stories

1. AI Layoffs Are Getting Real! Led by Oracle, US Tech Stops Sugarcoating

Source: Bloomberg (Original Link), Asanify (Original Link), TechCrunch (Original Link)
Summary: The US tech industry has hit a massive wave of layoffs in 2026. This time, major corporations are ditching the euphemisms and directly attributing job cuts to AI adoption in official regulatory filings and earnings calls. Oracle slashed 21,000 jobs over the past year, with Amazon and Meta following suit. Total AI-driven layoffs across the tech sector have now climbed into the tens of thousands.
Taiwan Perspective: Although Taiwan's tech industry remains heavily centered on hardware manufacturing, local branches of foreign MNCs and high-level local software talent are starting to feel the squeeze of AI-driven organizational flattening. This serves as a wake-up call for Taiwanese software engineers to accelerate their transition into "AI-collaborative talent."
Key Discussion Points:
1. Is corporate America openly blaming AI for layoffs because the tech is genuinely mature enough to replace humans, or is it just a convenient excuse to appease Wall Street shareholders?
2. As AI becomes the go-to scapegoat for layoffs, how can individual contributors rebuild their professional moats?
Suggested Script: Have you noticed how brutally honest tech layoff announcements have gotten lately? In the past, companies hid behind vague terms like "restructuring" or "macroeconomic headwinds." But now, legacy giants like Oracle and Amazon are flat-out stating in official filings: "We adopted AI, so we don't need this many people anymore." This is ringing alarm bells in Taiwan's tech community. We can't just treat AI as a cool toy anymore; it's actively taking jobs. But is this because AI is actually that powerful, or are companies just using it as a convenient excuse to trim fat and boost their balance sheets? It's definitely something worth thinking about.

2. Do Layoffs Actually Save Money? Gartner Report Shows AI Automation Hasn't Paid Off Yet

Source: Fortune (Original Link)
Summary: A new study from market research firm Gartner reveals that while up to 80% of companies have laid off staff due to AI implementation, these moves show zero correlation with improved return on investment (ROI). The study bursts the bubble of "AI layoffs equal operational efficiency," pointing out that many enterprises are rushing to fire employees before AI technology has actually delivered real productivity gains.
Taiwan Perspective: Taiwanese SMEs often suffer from a "bandwagon" mentality during digital transformation. This report serves as a timely warning for local management: blindly adopting AI and cutting headcount will only lead to internal knowledge gaps rather than boosting competitiveness.
Key Discussion Points:
1. Why do corporate executives suffer from the cognitive bias that AI can instantly replace human labor?
2. Before AI's ROI becomes clear, how should companies strike a balance between headcount allocation and technology adoption?
Suggested Script: This Gartner report is incredibly ironic—it's a massive reality check for managers rushing to lay off staff in the name of AI. The report points out that 80% of companies cutting heads haven't actually made more money from AI. It's like letting your horses go before you've even finished building the car engine. In Taiwan, we often see bosses jumping on whatever trend is hot overseas, thinking that buying an AI license means they can immediately slash payroll. In reality, the cost of maintaining, debugging, and integrating AI systems into existing workflows is astronomical. If you rush into layoffs without a proper plan, you'll end up with the worst of both worlds.

3. Trump Signs New AI Executive Order: Embracing Innovation, Rejecting Mandatory Licensing

Source: The White House (Original Link)
Summary: The White House has issued a new AI executive order aimed at driving innovation and security in frontier AI models. The order establishes a voluntary cooperative framework, asking developers to provide the federal government with up to a 30-day early evaluation window before publicly releasing software. Crucially, it explicitly bans any mandatory licensing or pre-clearance regimes for AI development, signaling a strong deregulatory stance.
Taiwan Perspective: The US easing up on AI regulation is great news for Taiwanese AI startups. It means the US-led platform and technology ecosystem will continue to iterate rapidly, allowing Taiwanese players to integrate more freely with the US ecosystem.
Key Discussion Points:
1. Can a "voluntary" 30-day review period actually mitigate the national security risks posed by AI?
2. How will a deregulated AI policy widen the gap between the US and the EU's strict regulatory framework?
Suggested Script: The executive order signed by the Trump administration has a very distinct style. Simply put, it's: "Big Brother wants a quick look, but we won't block you." The government is asking developers to give them a 30-day heads-up before releasing powerful models, but they are strictly avoiding any tedious licensing processes. This is great news for a tech industry that thrives on speed. For developers and startups in Taiwan, this hands-off approach from the US means the pace of innovation will accelerate, and we'll have access to more US-based open-source tools. However, it also puts the onus on us to self-regulate when it comes to security and compliance.

4. The King of Price-to-Performance? GLM-5.2 Open-Weight Model Beats GPT-5.5 at One-Sixth the Cost

Source: Labellerr (Original Link)
Summary: Beijing-based Zhipu AI has released its new open-weight model, GLM-5.2, boasting 753 billion parameters and a massive 1-million-token context window. In the FrontierSWE benchmark, the model not only outperformed OpenAI's GPT-5.5, but its running costs were also just one-sixth of its rival's, leveraging innovative IndexShare technology to drastically reduce compute resource consumption.
Taiwan Perspective: Faced with the aggressive rise of highly cost-effective Chinese open-source models, Taiwanese enterprises evaluating commercial applications must carefully balance cost advantages against geopolitical and cybersecurity risks.
Key Discussion Points:
1. Does an open-source model outperforming closed-source giants in both capability and cost mean that AI moats are disappearing?
2. How should Taiwanese companies establish secure and compliant evaluation frameworks when dealing with highly cost-effective models of Chinese origin?
Suggested Script: Zhipu AI's release of GLM-5.2 is absolutely jaw-dropping. Not only does it beat GPT-5.5 in performance, but it does so at just one-sixth of the cost. For developers building applications, that price point is incredibly tempting. However, in Taiwan, we inevitably have a red line when it comes to using models with Chinese backgrounds, especially regarding cybersecurity and geopolitical risks. This serves as a reminder that while open-source tech is great, figuring out how to design "de-risked" architectures—or shifting our focus to other US-backed open-source models like Llama—is a practical challenge the Taiwanese tech industry must face.

5. The Ultimate Battle for Developers! Microsoft and Google Target Anthropic to Dominate the AI Coding Market

Source: CNBC (Original Link)
Summary: AI-assisted coding has become the most fiercely contested battleground among tech giants. Microsoft and Google are throwing everything they have at challenging the dominance of Anthropic's Claude Code within the developer community. Analysts point out that whoever controls the developer's daily workflow will capture the ultimate high ground in the next wave of enterprise AI adoption.
Taiwan Perspective: Taiwan boasts a massive community of software and hardware engineers. This war over AI coding tools will directly reshape daily development workflows and R&D efficiency across Taiwan's tech sector.
Key Discussion Points:
1. Why does controlling the "developer workflow" equate to controlling the future of enterprise AI?
2. Claude Code's strengths lie in precision and contextual understanding. How can Microsoft and Google leverage their ecosystem advantages to fight back?
Suggested Script: Right now, the hottest battleground in AI isn't chatbots—it's AI coding. Anyone who has used Claude Code knows how mind-blowing it is; it's like having a senior engineer sitting right next to you, writing code. Now, Microsoft and Google are panicking, rolling out major updates at their respective conferences to claw back control. For us engineers in Taiwan, this competition is a massive win. But as these giants lower the barrier to coding further and further, we need to ask ourselves: is our value limited to just "understanding code," or can we bring better system architecture and business logic to the table?

6. EU AI Act Takes Full Effect in August! High-Risk AI Compliance Challenges Loom

Source: Gunderson Dettmer (Original Link), Hunton Andrews Kurth (Original Link)
Summary: The European Union's Artificial Intelligence Act (EU AI Act) will officially take full effect on August 2, 2026. At that point, transparency and risk management requirements for high-risk AI systems (such as recruitment, credit scoring, and critical infrastructure) will become legally binding. Meanwhile, Colorado's AI Act in the US has undergone major revisions, stripping out the original duty of care and algorithmic discrimination provisions while delaying its effective date.
Taiwan Perspective: With many of Taiwan's export-oriented enterprises and tech giants selling products to Europe, they must immediately establish AI risk assessment mechanisms that comply with EU standards. Otherwise, they risk astronomical fines and product bans.
Key Discussion Points:
1. How will the EU's strict enforcement contrasted with the loosening of US state-level bills affect global corporate AI deployment strategies?
2. How can resource-constrained Taiwanese SMEs cost-effectively achieve compliance with the EU AI Act?
Suggested Script: The EU AI Act is about to get very real. Starting August 2, if your AI system impacts hiring, credit scoring, or even critical infrastructure, and you want to sell to Europe, you must pass incredibly strict audits. This is the polar opposite of the laissez-faire attitude in the US. Many Taiwanese SaaS providers or hardware-embedded AI vendors cannot afford to take chances. You need to start auditing your algorithms for bias and verifying your data sources right now. Compliance is painful, but if we can secure this EU passport early, we can leave our competitors in the dust on the global stage.

Outro

Alright, that's all for today. From the reality behind tech giants' AI layoffs to the geopolitical and regulatory chess match between the US and China, we can see that AI development has moved past pure hype and into the realm of real-world business, law, and career survival. I hope today's episode gave you some valuable insights. If you enjoyed the show, don't forget to subscribe and share it with your tech-minded friends. See you in the next episode. Bye!