🎙️ AI 日報 Podcast：巨頭挖角戰、ChatGPT 跌破半壁江山、主權 AI 的焦慮

Introduction

Hello everyone, and welcome to "Mark's Tech Insights"! Today we have some absolutely mind-blowing topics to discuss. First off, Google DeepMind has lost two legendary, god-tier figures in a single week—a devastating, nuclear-level blow to Google. We'll also cover ChatGPT's market share dipping below 50% for the first time, and the political maneuvering as world leaders begin to worry that the US might "cut off" access to AI technology. We've got an action-packed lineup, so let's dive right in!

Today's Top Stories

1. Transformer Co-Author Noam Shazeer Leaves Google Again to Join OpenAI as Head of Architecture Research

Source: Build Fast with AI (https://www.buildfastwithai.com/blogs/ai-news-today-june-22-2026）
Summary: Noam Shazeer, co-author of the groundbreaking 2017 paper "Attention Is All You Need" that changed the world of AI, has decided to leave Google DeepMind to join rival OpenAI as Head of Architecture Research. Ironically, Google spent a staggering $2.7 billion in 2024 just to bring him back from Character.AI, only for him to jump ship again less than two years later.
Taiwan Perspective: While Taiwanese tech hardware giants like TSMC and Quanta have long supported global AI compute, this incident reminds us that the battle for top-tier talent is the ultimate deciding factor in the direction of AI development. Taiwan needs more unconventional thinking when it comes to retention strategies for software R&D talent.
Discussion Points:
1. Google spent $2.7 billion and still couldn't retain him. Is big-tech bureaucracy truly holding back top researchers?
2. Despite experiencing multiple leadership shakeups, OpenAI can still attract legendary figures like Shazeer. What is their secret sauce?
Script Suggestion: This move has completely stunned the tech community. Google played the fool just two years ago, spending nearly NT$90 billion to acquire Character.AI just to bring Noam back. Now, he's packed his bags and headed straight to their biggest rival, OpenAI. It's basically training troops for your competitor, which is a massive blow to Google. This also reflects that in the AI era, what truly decides the winner isn't how many servers you have, but who the smartest minds are working for. It seems Google's internal research environment might have some real issues, making top talent willing to walk away even if it means leaving massive paydays on the table.

2. Nobel Laureate and AlphaFold Pioneer John Jumper Leaves Google for Anthropic

Source: CNBC (https://www.cnbc.com/2026/06/19/john-jumper-to-leave-google-deepmind-for-anthropic.html）
Summary: John Jumper, the world-class scientist who won a Nobel Prize for his work on AlphaFold's protein structure prediction, has announced his move to competitor Anthropic after nearly nine years at Google DeepMind. Coming right on the heels of Noam Shazeer's departure, this marks a historic talent drain for Google DeepMind.
Taiwan Perspective: Taiwan is currently actively promoting the integration of biomedicine and AI, with biotech and healthcare being key industries of strength. John Jumper's move suggests that Anthropic might make major moves in the AI biomedicine space in the future. Taiwanese startups and research institutions in this field must closely monitor this shift in technology.
Discussion Points:
1. Does John Jumper's departure mean the golden age of Google DeepMind is over?
2. How will Anthropic leverage this Nobel laureate to reshape its strategy in scientific research-oriented AI?
Script Suggestion: Oh my goodness, Google is bleeding top talent. Just days after losing a Transformer co-author, they couldn't even retain Nobel laureate John Jumper, who is jumping ship to Anthropic. You have to realize, AlphaFold has been Google's poster child for AI's scientific applications over the past few years, and now the poster child himself has switched teams. This makes you wonder if DeepMind, after being fully integrated into Google and becoming heavily commercialized, has lost its original pure research atmosphere, prompting these world-changing scientists to leave in droves.

Source: TechCrunch (https://techcrunch.com/2026/06/16/chatgpts-market-share-slips-below-50-for-first-time/）
Summary: According to the latest data, ChatGPT's global market share has slipped to 46.4%, marking the first time it has lost its majority hold since its launch. Meanwhile, Google Gemini's market share has surged to 27.7%, and Anthropic's Claude has grown to 10.3%, indicating that the AI assistant market is entering a highly competitive, multi-polar era.
Taiwan Perspective: For Taiwanese developers and enterprises, this means the era of "betting everything on ChatGPT" is over. When developing Traditional Chinese applications, adopting a multi-model architecture and flexibly switching between Gemini or Claude based on specific scenarios is essential to achieve the best balance of cost and performance.
Discussion Points:
1. Is ChatGPT's declining market share due to competitors advancing too quickly, or are users experiencing "ChatGPT fatigue"?
2. What benefits does this three-way rivalry bring to average consumers and enterprise users?
Script Suggestion: Folks, this is truly a historic moment. While ChatGPT is still the king with over 1.1 billion monthly active users, its market share has dropped below 50% for the very first time. What does this mean? It means Gemini, with its deep integration into the Android ecosystem, and Claude, with its superior writing and coding capabilities, have successfully chipped away at the market. This is fantastic news for us as users because competition breeds progress. We no longer have to be locked into a single platform, and our options for AI assistants will only grow from here.

4. OpenAI Partners with Top Investment Banks to Secretly File for IPO, Valuation Soaring to $730 Billion

Source: QverLabs / AI Tools Recap (https://qverlabs.com/blog/openai-vs-google-deepmind-vs-anthropic-2026）
Summary: OpenAI is currently working with top investment banks, including Goldman Sachs and Morgan Stanley, to secretly file for an IPO, with a public listing expected as early as September 2026. The company's private market valuation has reached a staggering $730 billion, making it one of the most highly anticipated mega-IPOs in tech history.
Taiwan Perspective: Behind such a massive valuation lies an insatiable hunger for hardware compute. Taiwan's semiconductor supply chain and server ODMs, including TSMC and Foxconn, are bound to continue playing a critical role in this capital rush, ushering in a new wave of peak orders.
Discussion Points:
1. Is a $730 billion valuation backed by solid revenue, or is it another tech bubble?
2. Once public, OpenAI must answer to shareholders. Will this push them toward hyper-commercialization, causing them to abandon their original non-profit mission of "benefiting humanity"?
Script Suggestion: This is absolutely the wildest financial news of the year, if not the decade. OpenAI is actually planning to go public, and at a mind-boggling valuation of $730 billion. To put that in perspective, it eclipses many long-standing tech giants. This also explains why they've been rushing to roll out various paid features lately—to go public, you need to show investors a beautiful balance sheet. However, once they become a public company, OpenAI will face quarterly earnings pressure. Whether they can still stick to developing safe Artificial General Intelligence (AGI) under those conditions is a massive question mark.

5. Even Allies Are Worried: G7 Leaders Fear the US Might Unilaterally Restrict Access to Advanced AI Models in the Future

Source: TechCrunch (https://techcrunch.com/2026/06/17/world-leaders-want-american-ai-they-just-dont-want-america-to-be-able-to-turn-it-off/）
Summary: At the G7 summit, world leaders, including French President Emmanuel Macron and Indian Prime Minister Narendra Modi, expressed concerns that the US might unilaterally cut off other countries' access to advanced US AI models in the future for political reasons. This underlying mistrust is accelerating nations' resolve to develop "Sovereign AI" alternatives.
Taiwan Perspective: Situated at the hub of geopolitical tensions, Taiwan feels this risk of "supply disruption" most acutely. This once again proves that the National Science and Technology Council's decision to promote local large language models like TAIDE (Taiwan AI Dialogue Engine) was highly visionary. We must possess our own AI software capabilities.
Discussion Points:
1. If the US truly weaponizes AI models for geopolitics, how will it fracture the global tech ecosystem?
2. If every country builds its own "Sovereign AI," can they really compete with US tech giants given their limitations in compute and data?
Script Suggestion: This G7 summit revealed a very subtle message: even America's allies are keeping their guard up against the US. The sentiment from Macron and Modi is simple: we want advanced US AI technology, but we're even more terrified that one day the US government might get upset and pull the plug, leaving us stranded. This is exactly why "Sovereign AI" has become a major trend this year. Everyone wants models trained by their own country. This isn't just a technology issue; it's a matter of national security, and Taiwan must keep this firmly in mind.

6. A Powerful Soul for a Steel Body: Boston Dynamics Successfully Integrates Google Gemini Robotics Model into Spot Robot Dog

Source: Build Fast with AI (https://www.buildfastwithai.com/blogs/ai-news-today-june-22-2026）
Summary: Boston Dynamics has partnered deeply with Google Cloud and DeepMind to integrate the latest Gemini Robotics-ER 1.6 model into its Spot robot dog and Orbit vision inspection platform. This upgrade enables the robot dog to understand natural language commands, execute complex multi-step industrial inspection tasks, and autonomously learn from its surroundings.
Taiwan Perspective: As a global hub for industrial manufacturing and automation hardware, the fusion of robotics and AI represents a golden opportunity for Taiwan's industrial upgrading. Taiwanese manufacturers should actively pursue OEM/ODM and co-development opportunities for this type of Edge AI robotics hardware.
Discussion Points:
1. Now that robot dogs can understand language and learn from their environment, what impact will this have on future factories and the labor market?
2. While Google is losing software talent, it continues to demonstrate powerful technology in robotics. Could this be their future moat?
Script Suggestion: This final story feels straight out of science fiction. Boston Dynamics' famous Spot robot dog has finally been equipped with a real "brain." Previously, the robot dog could only follow pre-programmed paths. Now, with Google's Gemini robotics model, you can just talk to it naturally: "Go check if there's anything unusual with that dashboard over there," and it can understand, plan its route, and complete the task on its own. This is the smart factory of the future we've always imagined—a perfect marriage of software and hardware. I'm incredibly excited to see its future applications.

Conclusion

All right, that wraps up today's edition of "Mark's Tech Insights." From Google's talent crisis and ChatGPT's battle to defend its market share, to global anxieties over Sovereign AI, we can see that the AI industry is shifting from a pure technology race to a much more complex game of business and geopolitics. If you enjoyed our content, don't forget to subscribe and share it with your friends. See you next time, bye!