Please support this podcast by checking out our sponsors: - Invest Like the Pros with StockMVP - https://www.stock-mvp.com/?via=ron - SurveyMonkey, Using AI to surface insights faster and reduce manual analysis time - https://get.surveymonkey.com/tad - Consensus: AI for Research. Get a free month - https://get.consensus.app/automated_daily Support The Automated Daily directly: Buy me a coffee: https://buymeacoffee.com/theautomateddaily Today's topics: Brain-to-text AI breakthrough - Meta’s Brain2Qwerty v2 turns non-invasive MEG brain recordings into sentences in real time, hitting much higher word accuracy than earlier approaches. Keywords: brain-to-text, MEG, accessibility, open research. South Korea’s trillion-dollar AI push - South Korea outlined a roughly $1 trillion public-private plan to expand memory-chip fabs, build AI data centers, and commercialize humanoid robots by 2028. Keywords: Samsung, SK Hynix, DRAM, data centers, humanoid robots. BIS warns of AI bubble - The Bank for International Settlements cautioned that hyperscaler AI capital spending may be outrunning earnings and could unwind sharply if returns disappoint or financing tightens. Keywords: BIS, hyperscalers, overinvestment, private credit, market repricing. Amazon, Anthropic, OpenAI reshuffle - Amazon is reportedly looking at cheaper model options after shifting its Anthropic deal toward per-token pricing, while ties with OpenAI and Anthropic’s multi-cloud moves complicate alliances. Keywords: AWS, Claude, OpenAI, token pricing, cloud competition. Open-weight cyber AI spreads - China’s Z.ai released an MIT-licensed open-weight coding and security model, raising concern that advanced vulnerability-finding capability is now available without provider-side controls. Keywords: open-weight, vulnerabilities, offensive AI, patch cycles, security. China tech controls and chips - China tightened export controls aimed at Japanese defense-linked entities as Nvidia’s China position keeps eroding under export restrictions and Huawei gains ground, alongside new supercomputing claims. Keywords: export controls, rare earths, Nvidia, Huawei Ascend, TOP500. Apple tooling deal and iPhone leak - Apple quietly acquired assets behind the award-winning SwiftUI prototyping tool Play, while a leak suggests major iPhone 18 Pro chip packaging changes for sustained performance and on-device AI. Keywords: SwiftUI, Xcode, A20 Pro, thermal design, Neural Engine. Myanmar scam factories using AI - Investigations describe industrial scam compounds in Myanmar using U.S.-made AI and infrastructure to run romance and investment scams at scale, with Starlink frequently implicated. Keywords: scam compounds, trafficking, AI translation, cloud routing, Starlink. Clean energy growth vs demand - New energy data shows wind and solar added more global supply than any other source in 2025, yet total fossil use still hit records as demand grows and data centers draw more power. Keywords: renewables, electricity demand, nuclear, data centers, primary energy. Rethinking web caching and attention - Commentary this week questioned when service workers are truly necessary versus simpler HTTP caching, and why adding friction can help people reclaim attention from always-on devices. Keywords: service workers, HTTP caching, reliability, notifications, digital boundaries. Episode Transcript Brain-to-text AI breakthrough We’ll start with the biggest ambition-on-paper story of the day: South Korea says it’s lining up about a trillion dollars in combined government and corporate commitments to expand memory-chip production, build new AI data centers, and push humanoid robots toward mass commercialization by 2028. The headline driver here is memory. With AI systems hungry for high-end DRAM, shortages have been rippling into electronics costs, and Korea wants to protect its lead. Samsung and SK Hynix are tied to the largest chunk of investment, with a national goal of sharply increasing DRAM output over the next several years. Alongside that, major Korean groups plan data centers outside Seoul to broaden compute capacity beyond the capital. What makes this more than an industrial-policy headline is the social and political tension it carries: these projects demand huge amounts of power and water, and they land at a moment when workers are pushing back—especially in manufacturing—over job protections as robots arrive on the factory floor. South Korea’s trillion-dollar AI push That theme of “AI money, and what happens if it goes wrong,” showed up in a warning from the Bank for International Settlements. In its 2026 annual report, the BIS argues that the global surge in AI investment—especially by the biggest tech players—could end in a painful reversal. The BIS isn’t saying AI is a fad. It’s saying spending is starting to run ahead of the cash coming in, and that the competitive dynamic can push everyone to build too much, too fast, even when returns are uncertain. It also flags a web of financing relationships—hyperscalers, model labs, chipmakers, data-center builders, and private credit—where stress in one corner can spread quickly. And here’s the kicker: if AI-linked stocks reprice sharply, it can leak into the real economy through household wealth effects, particularly given how central U.S. equities are to global portfolios. BIS warns of AI bubble Speaking of the economics of AI, one of the most closely watched partnerships in the model-and-cloud world looks like it’s getting complicated. Reports say Amazon is exploring cheaper AI model options, including OpenAI, after renegotiating how it pays for Anthropic’s models. The key shift is billing: moving toward per-token pricing next year. If you’re running lots of customer-facing features and internal tools on a model, token-based pricing can turn “pretty expensive” into “hard to predict,” fast. Amazon even removed an internal leaderboard that had nudged employees to use more tokens—an early sign it’s trying to clamp down on runaway usage. What makes this strategically interesting is the triangle it creates. Amazon has been a major backer of Anthropic, but it’s also been edging closer to OpenAI through infrastructure and access deals. Meanwhile, Anthropic is spending more with Google Cloud, signaling it wants less dependence on AWS. The alliances in AI aren’t just about model quality anymore—they’re about cost, leverage, and who controls the relationship when the bill comes due. Amazon, Anthropic, OpenAI reshuffle Now to the most surprising research update of the day: Meta announced Brain2Qwerty version 2, an AI system that decodes sentences from non-invasive brain recordings in real time. Until recently, the most impressive brain-to-text results usually came from implanted sensors, which limits who can use them and when. Meta’s approach uses magnetoencephalography, or MEG, paired with language-model fine-tuning so the system can use context to clean up noisy signals. Meta reports an average word accuracy a bit above sixty percent, with one participant reaching the high seventies. This is still early-stage, lab-based work—not a consumer product. But it’s a meaningful step toward communication tools for people who can’t speak due to injury or disease. Meta is also releasing code, and a partner lab is releasing a dataset, which could speed up independent research and clinical exploration. Open-weight cyber AI spreads Staying with AI capability—but from a security angle—China’s Z.ai, previously known as Zhipu, released an open-weight model aimed at long-horizon coding and vulnerability discovery, published under a permissive license. The story here isn’t just performance. It’s distribution. With open-weight models, there’s no provider “switch” to flip off, no centralized monitoring, and no practical way to stop misuse once it’s running locally. Reporting suggests offensive workflows and jailbreaks spread quickly after release. For defenders, the implication is blunt: patch faster, audit more continuously, and assume capable attackers can bring sophisticated AI to bear without asking anyone’s permission. China tech controls and chips That leads neatly into the broader geopolitics of compute. Nvidia’s attempt to keep selling advanced AI chips into China continues to hit walls as U.S. export controls tighten and China pushes buyers toward domestic alternatives, especially Huawei. Estimates now suggest Nvidia’s share of China’s AI-chip market has fallen sharply from its earlier dominance, with Huawei climbing fast on both chips and large computing clusters. Even if Nvidia retains an edge at the frontier, the center of gravity is shifting: developers adapt their software to whatever hardware they can reliably get, and that rewrites ecosystems over time. Meanwhile, China is also escalating pressure points elsewhere. Beijing expanded export controls aimed at Japan, adding multiple defense-linked research institutes to a list that blocks transfers of Chinese-origin dual-use items, and tightening scrutiny for other firms. It’s another reminder that supply chains for critical materials and components are now policy tools, not just commerce. Apple tooling deal and iPhone leak And on the supercomputing bragging-rights front, China is claiming it’s back on top of the TOP500 rankings with a new system called LineShine in Shenzhen. The claim is notable for two reasons: it’s framed as achieving extreme performance without GPUs, and it serves as a signal that China can still assemble world-leading systems under restrictions. But there’s also a tradeoff highlighted in the reporting—power draw. Even when raw performance climbs, energy efficiency is becoming the defining constraint, especially as AI and HPC expand. Myanmar scam factories using AI Now to Apple, which had two very different kinds of headlines. First, a quieter one that matters to developers: Appl
