🎙️ AI 日報 Podcast，頂尖人才爭奪戰、開源大模型新局、地緣政治基礎建設

Opening Remarks

Hello everyone, and welcome to today's episode of "Mark's Tech Insights." Today, we're going to talk about the earth-shattering talent shakeup in the AI world over the past few days, as well as NVIDIA flexing its muscles once again with a massive new open-source model. Plus, the latest geopolitical sparring between the US and China over AI infrastructure and policy is also on the main stage today. It's an action-packed lineup, so don't miss out!

Today's Top Stories

1. Nobel Laureate John Jumper Leaves Google DeepMind to Join Anthropic in Blockbuster Move

Source: ExplainX AI (https://explainx.ai/blog/john-jumper-leaves-google-deepmind-anthropic-alphafold-2026）)
Summary: AI giant John Jumper, the creator of AlphaFold and winner of the 2024 Nobel Prize in Chemistry, has officially announced his departure from Google DeepMind after nearly nine years to join competitor Anthropic. This is undoubtedly a major, critical talent win for Anthropic as it gears up for an IPO.
Taiwan Perspective: Taiwan's biomedical AI sector is also booming. The migration of such top-tier scientists could accelerate Anthropic's footprint in biomedical computing and drug discovery, making this a trend that Taiwanese biotech startups and academia should monitor closely.
Key Discussion Points:
1. Why would even a Nobel laureate choose to leave Google, with all its vast resources?
2. How much of a boost will this bring to Anthropic's upcoming IPO valuation?
Script Suggestion: This news is absolutely mind-blowing. John Jumper is the mastermind behind AlphaFold, and for him to be poached by Anthropic shows that the "talent arms race" in AI has reached a fever pitch. It's no longer just about compute power; it's about securing the world's sharpest minds. Taiwan's tech sector should take note: as global giants turn biomedical AI into the next major battleground, we need to move even faster in cross-disciplinary collaboration between biomedicine and semiconductors.

2. OpenAI Aggressively Recruits AI Legend Noam Shazeer and Former White House Official to Push for Late 2026 IPO

Source: TechCrunch (https://techcrunch.com/2026/06/18/openai-is-bringing-on-some-big-guns-in-the-lead-up-to-its-ipo/）)
Summary: In preparation for an IPO that could happen as early as September 2026, OpenAI has been on a massive hiring spree. They have not only poached Noam Shazeer, former co-lead of Google Gemini and founder of Character.ai, but have also brought on Dean Ball, a former AI policy official from the Trump administration, showcasing their ambition to balance both technical prowess and government relations.
Taiwan Perspective: OpenAI's IPO will be a bellwether for global tech stocks. Taiwan's investment community and supply chain are watching with bated breath, as this will directly determine whether the funding frenzy in the AI market can be sustained over the next few years.
Key Discussion Points:
1. What kind of chemical reaction will Noam Shazeer's arrival spark in OpenAI's product roadmap?
2. Does recruiting a former Trump administration official suggest OpenAI is paving the way for the future regulatory landscape?
Script Suggestion: Have you guys noticed how strategic OpenAI's hiring has been lately? On one hand, they brought in tech legend Noam Shazeer to solidify their R&D muscle. On the other hand, they recruited Dean Ball, who understands the Trump administration's policies. This is a clear move to shore up both "technology" and "political risk" ahead of their IPO. This is also a great lesson for tech startups in Taiwan: once a company reaches a certain scale, public relations and policy navigation become just as critical as technical R&D.

3. NVIDIA Releases 550-Billion-Parameter Nemotron 3 Ultra, Aiming for the Largest Open-Source AI Model in the US

Source: LLM Stats (https://llm-stats.com/llm-updates）)
Summary: NVIDIA has introduced its brand-new Nemotron 3 Ultra model, boasting a massive 550 billion parameters and utilizing a Mixture of Experts (MoE) architecture with 55 billion active parameters. This makes it the most powerful and largest open-weights model released by a US tech company to date.
Taiwan Perspective: This is a major boon for Taiwan's vast community of AI developers and its hardware supply chain. Larger, more powerful open-source models mean enterprises can fine-tune models locally at a lower cost, driving demand for Taiwan's on-premises AI servers.
Key Discussion Points:
1. As NVIDIA continues to push open-source models, what kind of competitive pressure will this place on Meta's Llama series?
2. How high is the hardware barrier to entry for small and medium enterprises looking to deploy a 550-billion-parameter open-source model?
Script Suggestion: NVIDIA really pulled out the big guns this time. Making the 550-billion-parameter Nemotron 3 Ultra open-source is a massive shot in the arm for the open-source community. In the past, people thought open-source models couldn't compete with closed-source ones like GPT. But NVIDIA has just proven that open-source performance can go toe-to-toe with the absolute best. This also means Taiwanese server manufacturers are going to be incredibly busy, because running such a massive on-premise model means hardware specs will inevitably need a major upgrade.

4. China Splurges $295 Billion to Build AI Compute Network, Mandating 80% Domestic Chip Ratio

Source: LLM Stats (https://llm-stats.com/llm-updates）)
Summary: The Chinese government has announced a massive 5-year plan worth 2 trillion RMB (approximately $295 b i l l i o n U S D) a i m e d a t b u i l d i n g a n a t i o n w i d e i n t e r c o n n e c t e d n e t w o r k o f A I d a t a c e n t e r s . T h e m o s t s t r i k i n g a s p e c t o f t h i s p l a n i s t h e e x p l i c i t m a n d a t e t h a t a t l e a s t 80 295 billion USD) aimed at building a nationwide interconnected network of AI data centers. The most striking aspect of this plan is the explicit mandate that at least 80% of the infrastructure must use domestic chips, such as those from Huawei. When factoring in power grid upgrades, the total investment is expected to reach a staggering$ 740 billion USD.
Taiwan Perspective: China's forced "de-Americanization" and push to increase the share of domestic chips will accelerate self-reliance within supply chains like Huawei's. This presents both a challenge and a potential transition opportunity for Taiwan's semiconductor foundry and packaging/testing industries.
Key Discussion Points:
1. Can the 80% domestic chip mandate truly bridge the gap in advanced manufacturing processes caused by US sanctions?
2. How will an overall investment of up to $740 billion reshape the global AI infrastructure landscape?
Script Suggestion: This 2-trillion-RMB initiative from China is arguably their strongest counterattack yet against US chip sanctions. Instead of hoping to buy high-end NVIDIA chips, they are leveraging state power to mandate that over 80% of chips be domestic, like Huawei's. While this "internal circulation" strategy might face some technical growing pains, once this massive market scale takes shape, it will undoubtedly trigger long-term and profound shifts for Taiwan's mature semiconductor nodes and packaging/testing supply chains.

5. Trump Signs Latest AI Executive Order, Requesting Developers Voluntarily Submit Frontier Models and Establishing Cybersecurity Exchange

Source: The White House (https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/）)
Summary: US President Donald Trump has signed an executive order titled "Promoting Advanced Artificial Intelligence Innovation and Security," requesting AI developers to voluntarily share model information with the federal government up to 30 days before releasing new frontier models. Additionally, the order directs national security agencies to establish an AI Cybersecurity Security Exchange within 30 days.
Taiwan Perspective: This shift in US policy indicates a ramp-up in efforts to mitigate national security and cybersecurity risks posed by AI. As a core partner in the US tech supply chain, Taiwan will see its cybersecurity compliance standards raised accordingly.
Key Discussion Points:
1. Will a "voluntary sharing" mechanism be effective in regulating tech giants like OpenAI or Anthropic?
2. How will the newly established AI Cybersecurity Exchange impact the tech cold war between the US and China?
Script Suggestion: Although the Trump administration's executive order uses the word "voluntary" on the surface, anyone can see that this is sending a strong regulatory signal. It demands that tech giants prioritize national security while chasing technical breakthroughs. For hardware and software developers in Taiwan, this means that AI products shipped to the US will face increasingly stringent cybersecurity inspections. Security safeguards must be baked in right from the initial design phase.

6. Capacity Reaching 1 GW: Meta Splurges $10 Billion to Expand El Paso, Texas AI Data Center

Source: AI Funding Tracker (https://aifundingtracker.com/top-50-ai-startups/）)
Summary: Meta has dramatically scaled up its investment in the El Paso, Texas AI data center, skyrocketing from an initial $1.5 b i l l i o n t o o v e r$ 10 billion. Slated to open in 2028, the data center will boast a staggering power capacity of 1 gigawatt (GW), dedicated to handling massive AI training demands.
Taiwan Perspective: A scale of $10 billion and 1 GW of power translates to massive demand for high-performance servers and liquid cooling systems. This is a huge win for Taiwanese supply chain players like Wistron, Quanta, and Auras Technology.
Key Discussion Points:
1. A power consumption of 1 GW is equivalent to that of a medium-sized city. How much of a strain will this put on the local power grid?
2. Does Meta's aggressive capital expenditure hint at even grander ambitions for Llama 5 or subsequent models?
Script Suggestion: Meta is really throwing around serious cash here, scaling up from $1.5 b i l l i o n s t r a i g h t t o$ 10 billion. The sheer scale of this appetite is mind-boggling. A gigawatt of power is basically funneling the output of an entire power plant directly into a data center. This just goes to show that the future of AI competition is a battle of energy and hardware. As the global leader in cooling systems and server manufacturing, Taiwan's tech sector is sitting front row, ready to reap massive rewards from this mega-construction wave.

Wrap-up

We've covered a lot today—from the migration of top-tier talent and new benchmarks in US open-source models, to the massive infrastructure spending sprees by both the US and China. You can really feel that the competition in the AI space has entered a whole new dimension. Thanks for listening! If you enjoyed today's episode, don't forget to subscribe and share it with your friends. See you next time, bye!