🎙️ AI 日報 Podcast，Copilot帳單風暴、Anthropic估值超車、開源模型大洗牌

Intro

Good morning, everyone! Welcome to Mark's Tech Insights. I'm Mark. Today, we're going to talk about the recent billing storm that has software engineers and startup founders shaking in their boots, as well as the latest wild funding wars in the AI space—where Anthropic's valuation has officially overtaken OpenAI. We've got an exciting lineup today, so let's dive right in.

Today's Top Stories

1. GitHub Copilot Billing Shock: Automated Agent Users Face 10x+ Cost Surges

Source: TechTimes (https://www.techtimes.com/articles/319340/20260629/github-copilot-billing-shock-confirmed-agentic-users-face-10x-cost-surge.htm）)
Summary: Starting in June, GitHub Copilot transitioned to a token-based pricing model for its advanced features. This has caused bills to skyrocket for developers running automated agents, with some reporting jumps from $29 / m o n t h t o$ 750, and others even hitting $3,000. While standard code autocompletion remains unlimited, advanced model chat and automated code reviews are now metered by usage, sparking a massive backlash in the developer community.
Taiwan Perspective: Many software startups and outsourcing teams in Taiwan rely heavily on Copilot to boost productivity. This pricing shift will force companies looking to implement agentic automated workflows to re-evaluate their costs. It might even push developers back to integrating raw APIs directly or searching for cheaper alternatives.
Key Discussion Points:
1. The end of the all-you-can-eat era: Will this become the new normal for AI subscription services?
2. How can developers optimize agent prompts to prevent automated tools from burning through tokens endlessly?
Script Suggestion: This is honestly insane. Everyone used to treat Copilot as an all-you-can-eat buffet. But now, the moment you let an agent automatically run code reviews or fix bugs, your bill can easily jump 10x or even 50x. Many startup teams in Taiwan operate on tight budgets, and this pricing change might force them to alter their development habits just to avoid crying over next month's bill. No wonder everyone is scrambling to find cheaper API alternatives right now.

2. Anthropic Confidentially Files for IPO, Valuation Nears $1 Trillion After Series H, Overtaking OpenAI

Source: Fortune (https://fortune.com/2026/06/01/anthropic-confidentially-files-ipo-965-billion-valuation/）)
Summary: Anthropic confidentially filed for an IPO with the US SEC in early June, with plans to go public this October. Following its latest $65 b i l l i o n f u n d i n g r o u n d, i t s v a l u a t i o n h a s r e a c h e d a s t a g g e r i n g$ 965 billion, officially overtaking OpenAI's $852 billion to become the world's most valuable independent AI unicorn.
Taiwan Perspective: This indicates that the market is placing a massive premium on AI pathways that prioritize safety and alignment. As Taiwan develops sovereign AI or enterprise-facing applications, we might want to look at this business model that leverages safety and compliance as its core selling points.
Key Discussion Points:
1. Why have Anthropic's revenue and valuation grown at such an astronomical rate?
2. With its valuation surpassing OpenAI, does this mean the leader of the AI market is about to change hands?
Script Suggestion: This is absolutely the biggest bombshell in tech this year. Anthropic actually overtook OpenAI in valuation, which would have been almost unthinkable two years ago. It shows that Wall Street and VCs have immense confidence in the commercialization potential of the Claude model family, especially with their excellent positioning around security in the enterprise market. If the October IPO goes through, it will be one of the largest listings in tech history. We definitely need to keep a close eye on this.

Source: TechTimes (https://www.techtimes.com/articles/317352/20260529/chinese-ai-models-lead-openrouter-traffic-coding-gains-come-china-data-risk.htm）)
Summary: According to the latest data from OpenRouter, open-source models developed in China now account for a whopping 61% of token consumption among the platform's top 10 models, up from just 2% at the end of 2024. GLM, MiniMax, and Kimi are the most popular. To cut costs, many US companies are shifting their agent workloads to these Chinese alternatives, which are 80% to 90% cheaper.
Taiwan Perspective: When choosing these low-cost models, Taiwanese enterprises must prioritize information security and data residency. Especially when handling highly sensitive source code or customer data, we cannot blindly adopt them just because of the cheap price tag.
Key Discussion Points:
1. Is cheap really a cure-all? How should enterprises balance cost savings against cybersecurity risks?
2. Will the aggressive rise of Chinese open-source models force major US providers to slash prices to compete?
Script Suggestion: These numbers are mind-blowing. Over 60% of the token consumption on OpenRouter is now going to Chinese open-source models. This actually ties right back to the Copilot price hikes we just talked about. As APIs from top-tier US providers get more expensive, developers looking to keep their agents running are naturally gravitating toward alternatives that are 90% cheaper. However, for developers in Taiwan, security risks have to be the top priority. Sending your code and sensitive data to these platforms raises compliance issues that are no joke.

4. Zhipu AI Launches Open-Source GLM-5.2, Outperforming GPT-5.5 at One-Sixth the Cost

Source: Trending Topics (https://www.trendingtopics.eu/glm-5-2-chinas-zhipu-ai-beats-even-googles-top-models-with-its-new-open-llm/）)
Summary: China's Zhipu AI has released GLM-5.2, a brand-new open-source model under the MIT license featuring a 750-billion-parameter sparse Mixture of Experts (MoE) architecture. It clinched first place on SWE-bench Pro, which evaluates coding capabilities. Furthermore, its API is priced at just $0.95 t o$ 2 per million tokens—80% to 90% cheaper than comparable models from OpenAI and Anthropic.
Taiwan Perspective: The open-source nature and rock-bottom pricing of GLM-5.2 make it highly attractive to budget-constrained academic research institutions and small-to-medium development teams in Taiwan. However, they must still remain mindful of technical compatibility and potential geopolitical risks.
Key Discussion Points:
1. How have open-source models managed to catch up to, or even surpass, top-tier closed-source models in such a short time?
2. Will this extreme price war accelerate the mainstream adoption of AI applications?
Script Suggestion: Speaking of looking for cheaper alternatives, here we go. Zhipu AI's new GLM-5.2 isn't just cheap; it actually took the top spot among open-source models for coding capabilities, and it even natively supports the MCP (Model Context Protocol). The temptation of getting top-tier performance at one-sixth of the cost is an absolute lifesaver for agent projects that burn through tokens. Still, we need to stay rational—while enjoying the benefits of open source, we must also properly assess the risks.

5. AI Inference Chip Challenger Cerebras Goes Public, Surging 68% on Day One in Year's Biggest IPO

Source: CNBC (https://www.cnbc.com/2026/05/14/cerebras-cbrs-stock-trade-nasdaq-ipo.html）)
Summary: AI inference chipmaker Cerebras Systems officially listed on Nasdaq, with its stock soaring 68.2% on its first day, pushing its market cap close to $95 b i l l i o n . R a i s i n g a m a s s i v e$ 5.55 billion, this listing is the largest IPO of 2026, highlighting Wall Street's intense hunger for AI hardware alternatives to Nvidia.
Taiwan Perspective: Cerebras's wafer-scale chip technology relies heavily on advanced packaging. This undoubtedly represents another massive business opportunity—and capacity challenge—for Taiwan's semiconductor supply chain, particularly TSMC's CoWoS capacity.
Key Discussion Points:
1. Does Cerebras's wafer-scale engine technology actually stand a chance of threatening Nvidia's dominance in the inference market?
2. What role does Taiwan's semiconductor supply chain play in the rise of this non-Nvidia camp?
Script Suggestion: Everyone has been asking: who else besides Nvidia can get a piece of the AI compute pie? Cerebras's wild performance on Nasdaq just gave us the answer. Their crazy idea of using an entire wafer to build a single chip actually delivers mind-blowing advantages in inference speed. Plus, their client list surprisingly includes OpenAI, Amazon, and Meta. This is fantastic news for Taiwan's semiconductor industry, because turning such extreme chip designs into reality ultimately depends on Taiwan's world-class advanced nodes and packaging technologies.

6. Q1 2026 VC Funding Hits Record High, with AI Devouring 80% of Global Capital

Source: Crunchbase News (https://news.crunchbase.com/venture/record-breaking-funding-ai-global-q1-2026/）)
Summary: Global venture capital funding broke records in Q1 2026, surpassing $300 billion, with a staggering 80% of that capital flowing into AI-related companies. Funding rounds for just four giants—OpenAI, Anthropic, xAI, and Waymo—accounted for 63% of all global VC funding, with the vast majority of capital concentrated in US-headquartered companies.
Taiwan Perspective: With capital heavily concentrated in a handful of US AI giants, Taiwanese startups looking for international VC funding must think about how to position themselves within these giants' ecosystems. Otherwise, it will be incredibly difficult to capture a slice of this funding boom.
Key Discussion Points:
1. Is such extreme concentration of capital in just a few companies healthy for the long-term development of the tech ecosystem?
2. How can Taiwanese startups survive and carve out a niche in such an extreme funding environment?
Script Suggestion: Finally, let's look at where the big money is going—and these numbers are absolutely mind-boggling. Over the course of a single quarter, 80% of global VC funding went straight into AI, with just four companies, including OpenAI and Anthropic, taking the lion's share. This shows that the AI race has turned into an incredibly capital-intensive arms race. For Taiwanese startups, competing directly on foundational models against these monsters is out of the question. Our opportunity lies in leveraging the compute power these giants unleash to build vertical applications. Riding their momentum is the way to go.

Outro

We've covered a lot of ground today—from Copilot's billing pain points and the cost wars between US and Chinese open-source models, to Wall Street's massive bets on AI hardware and startups. It's clear that the AI industry is moving rapidly from pure technology showcases to brutal commercial cost realities. Thank you so much for listening. See you next time, bye!