🎙️ AI 日報 Podcast，2026未來沙盤推演、巨頭專利大戰、腦機介面商用化

Opening

Hello everyone, welcome to "Mark's Tech Insights," I'm Mark! Today's episode is very special. Instead of talking about today's news, we are going to step into a time machine and conduct a "July 2026 Tech Scenario Simulation"! Combining current technology trends and industry directions, we are boldly predicting and simulating the tech bombshells that could go off two years from now. This is a future tech rhapsody based on realistic deductions. Let's see what kind of crazy shape the tech world will take when we reach 2026!

Today's Key News

1. Moonshot AI Launches Kimi K3: The Strongest 2.8-Trillion-Parameter Open-Source Model Dominates the Coding Arena

Source: Tech Startups (https://techstartups.com/2026/07/17/top-tech-news-today-july-17-2026-anthropic-apple-google-meta-moonshot-ai-nvidia-more/）
Summary: Beijing-based AI startup Moonshot AI has launched the Kimi K3 model. This is a Mixture of Experts (MoE) model boasting up to 2.8 trillion parameters and a 1-million-token context window, setting a record for the scale of open-source models. It claimed first place in Arena.ai's front-end coding arena with a 76% win rate, defeating Claude Fable 5 and GPT-5.6 Sol.
Taiwan Perspective: Taiwan's software development and startup communities rely heavily on highly cost-effective open-source models. Kimi K3's powerful coding capabilities and open-source nature will significantly reduce fine-tuning costs for Taiwanese enterprises developing high-end Traditional Chinese applications, especially in the IT services sector which requires complex system integration.
Discussion Points:
1. Can the open-source community's hardware actually run a 2.8-trillion-parameter MoE model? Will it be a case of "look but can't touch"?
2. With Chinese AI models surpassing US giants in the coding domain, what impact will this have on the global software outsourcing and development landscape?
Script Suggestion: "Imagine a massive 2.8-trillion-parameter beast of a model going the open-source route! It's like someone suddenly publishing Ferrari's blueprints online for free and telling you: 'Go ahead and modify it!' This time, Kimi K3 isn't just huge in parameters; in terms of coding performance, it directly wiped the floor with Claude 5 and GPT-5.6. For us engineers in Taiwan, this is absolutely a godsend. In the future, we won't have to be held hostage by the subscription fees of major US vendors. We can directly fine-tune open-source models to create super-strong localized services. The cost-performance ratio is simply off the charts!"

2. After a Six-Week Delay, Google Gemini 3.5 Pro Officially Goes Live: Featuring a Massive 2-Million Context Window and a $250 Ultra Premium Subscription

Source: Build Fast With AI (https://www.buildfastwithai.com/blogs/ai-news-today-july-17-2026）
Summary: After resolving architectural flaws related to "recursive tool-calling," the six-weeks-late Google Gemini 3.5 Pro has finally launched. The model features a massive 2-million-token context window and offers a "Deep Think" reasoning mode in its $250/month Ultra subscription plan, alongside the release of its API pricing.
Taiwan Perspective: A subscription fee of $250 U S D (a p p r o x i m a t e l y$ 8,000 TWD) per month is a huge sum for average consumers in Taiwan. This indicates that AI services have officially transitioned from "mass-market toys" to "high-yield professional tools." Taiwan's business-to-business (B2B) applications must think about how to create vertical value that justifies this price tag.
Discussion Points:
1. Did the delay caused by "recursive tool-calling" errors expose fatal security risks in AI Agent automated decision-making?
2. Will the $250/month subscription pricing shut out small and medium-sized enterprises and individual creators?
Script Suggestion: "Google was six weeks late this time, and the reason turned out to be that 'recursive tool-calling' got stuck. Simply put, when the AI called its own tools, it fell into an infinite loop and crashed. But the Gemini 3.5 Pro released this time is truly powerful. With a 2-million-token context, it means you can feed it an entire year's worth of financial statements and contracts all at once. However, the most jaw-dropping part is that the Ultra version, which features the 'Deep Think' mode, costs $250 a month! In Taiwan, that price can buy you several high-end Wagyu BBQ meals. Google is clearly telling everyone: 'My AI is not for casual chatting; it's a professional special assistant designed to help you make big money!'"

3. Partnership Shattered! Apple Officially Sues OpenAI for Alleged Theft of Trade Secrets

Source: Tech Startups (https://techstartups.com/2026/07/17/top-tech-news-today-july-17-2026-anthropic-apple-google-meta-moonshot-ai-nvidia-more/）
Summary: Following the breakdown of their publicly announced partnership, Apple has filed a trade secret lawsuit against OpenAI. This legal battle could severely impact OpenAI's planned development of proprietary consumer hardware devices and its 2026 product roadmap.
Taiwan Perspective: The legal battle between Apple and OpenAI directly tests Taiwan's tech hardware supply chain. If OpenAI's hardware plans are blocked, Taiwan's related AI device (AI PC / AI Phone) supply chain may need to adjust production capacity and partners.
Discussion Points:
1. Moving from "joining forces" to "suing in court," why was the honeymoon period between Apple and OpenAI so short?
2. How will this lawsuit change the rules of cooperation between AI software vendors and hardware manufacturers in the future?
Script Suggestion: "This is absolutely the biggest soap opera in the tech world this year! Apple and OpenAI were just happily announcing their partnership not long ago, and now they are tearing each other apart in court. Apple is suing OpenAI for stealing their trade secrets. The reason behind this isn't hard to guess: OpenAI has always wanted to make its own physical AI hardware, and the one thing Apple tolerates least is others encroaching on its hardware turf. As these giants clash, our tech supply chain in Taiwan needs to keep its eyes wide open. After all, whether it's Apple's chips or OpenAI's future hardware, there is a high probability they will end up being manufactured in Taiwan. With these two fighting, the order winds for Taiwanese factories might shift again!"

4. Moving Toward Commercialization! China Completes the First Commercial Invasive Brain-Computer Interface Surgery

Source: Build Fast With AI (https://www.buildfastwithai.com/blogs/ai-news-today-july-17-2026）
Summary: China has announced the completion of the world's first commercial invasive brain-computer interface (BCI) implant surgery, marking the official transition of BCIs from clinical trials to the commercial market. This move intensifies global competition in the BCI field while Western rivals like Neuralink remain in testing phases.
Taiwan Perspective: The commercialization of brain-computer interfaces requires extremely high-precision manufacturing and chip packaging. Taiwan possesses strong advantages in microelectronics, precision medical devices, and semiconductor packaging (such as advanced packaging technologies), and is poised to become a core supply chain partner for global BCI equipment in the future.
Discussion Points:
1. What risks to ethics and safety does bypassing long-term clinical trials to directly commercialize invasive brain-computer interfaces bring?
2. How will the "race for speed" between China and the US in BCI reshape the future biomedical and defense industries?
Script Suggestion: "Do you all remember the scene in the movie The Matrix where plugging a cable into the back of your head lets you download a helicopter pilot manual? This might not be too far off. China has bypassed lengthy trials to complete the world's first 'commercial' invasive brain-computer interface surgery. While Elon Musk's Neuralink is still carefully conducting tests, its competitor has already started selling tickets to get on board. But honestly, to install a chip in the brain, the quality and packaging requirements for that chip are unimaginably high. This is where Taiwan's semiconductor and precision medical industries have a massive opportunity. After all, when it comes to making absolutely safe, non-rejectable chips, there are very few places in the world that can pull it off!"

5. Thinking Machines, Founded by Former OpenAI CTO Mira Murati, Secures $2 Billion Seed Round and Launches 975-Billion-Parameter "Inkling" Model

Source: Distill Intelligence (https://www.distillintelligence.com/briefings/ai-leaders-2026-07-17）
Summary: Thinking Machines, a startup founded by former OpenAI CTO Mira Murati, has completed a massive $2 billion seed funding round. Simultaneously, it launched the 975-billion-parameter "Inkling" model, directly declaring war on its founder's former employer and competitors like Anthropic.
Taiwan Perspective: A $2 billion "seed round" is astronomical, demonstrating global capital's frantic pursuit of top-tier AI talent. While it is difficult for Taiwan's startup ecosystem to secure such massive early-stage funding, they can focus on applying these powerful foundation models to specific vertical domains (such as smart manufacturing and healthcare).
Discussion Points:
1. With $2 billion raised just in a seed round, is the risk of an AI bubble rapidly accumulating?
2. Does Thinking Machines, led by Mira Murati, have a chance to break the duopoly of OpenAI and Anthropic?
Script Suggestion: "After former OpenAI CTO Mira Murati left, everyone was guessing what she would do next. It turns out her first move was securing a $2 billion 'seed' round! Yes, you heard that right, a seed round! In Taiwan, that amount of money could fund several unicorns. And they didn't just take the money; they immediately rolled out a 975-billion-parameter model called 'Inkling.' This is a clear message to her former employer: 'I'm going solo, and I can do it faster and better than you!' This kind of tech schism brought by top talent shows us that the future AI landscape definitely won't be dominated by just one or two players. The era of a hundred schools of thought contending is just beginning!"

6. Marching Toward a Trillion-Dollar Empire? Anthropic Reportedly Preparing for Late 2026 IPO, Valuation Could Surpass $1 Trillion

Source: Distill Intelligence (https://www.distillintelligence.com/briefings/ai-leaders-2026-07-17）
Summary: Backed by billions of dollars in credit lines and massive venture capital interest, Anthropic is in deep preparations for an IPO in late 2026. The valuation for this listing could surpass $1 trillion, potentially making it one of the largest IPOs in tech history.
Taiwan Perspective: The birth of a trillion-dollar AI company means that AI is no longer just a "trend," but an infrastructure giant on par with Apple and Microsoft. Taiwan's capital markets and large tech conglomerates should actively evaluate how to participate in this historic wealth restructuring through overseas investments or business alliances.
Discussion Points:
1. Is it reasonable for an AI company that has not yet achieved explosive profitability to challenge a "trillion-dollar valuation" solely on technology and vision?
2. How will Anthropic's IPO reshape the strategic alignment of its major backers, Amazon and Google?
Script Suggestion: "What does one trillion dollars mean? That is equivalent to several years of Taiwan's annual GDP! Anthropic, which focuses on AI safety, is now rumored to be going public at the end of 2026, with a valuation directly challenging one trillion dollars. We used to think only Apple selling iPhones or Microsoft making Windows could be worth that kind of money, but now a pure AI company is standing on equal footing. This actually serves as a warning to us in Taiwan: the commercial value of AI is expanding at the speed of light. We cannot just be the foot soldiers manufacturing chips. How to find our own software positioning in this trillion-dollar market is the most critical task for the next two years!"

Outro

Alright, that's the special "July 2026 Tech Scenario Simulation" episode we prepared for you today! After looking at these exciting future simulations, do you also feel that the tech world two years from now is both thrilling and a bit nerve-wracking? No matter how the future changes, stay tuned to "Mark's Tech Insights," and I will keep you at the absolute forefront of technology. See you next time, bye-bye!