In partnership with
In Today’s Issue:
🧠 Baidu claims its new ERNIE 5.1 model delivers frontier performance
🍎 Apple may turn iOS into a neutral AI marketplace
💊 Novo Nordisk partners with OpenAI
💳 AI agents are getting wallets as Visa and Mastercard build the infrastructure
✨ And more AI goodness…
⚡ The Signal
The AI race is moving from models to rails.
The most consequential moves today are not about who has the smartest model — they are about who controls the infrastructure AI runs through. Apple is turning iOS into a neutral distribution layer where rival models compete for attention. Visa and Mastercard are building the trust architecture that lets AI agents spend real money. Washington and Beijing are negotiating the diplomatic rails to keep the superpower AI rivalry from spiraling. Even Baidu's cost-efficiency claim is really an infrastructure argument: that the moat is shifting from raw compute spending to training economics and distribution. The pattern is clear — the model war is maturing into a platform war, and the winners will be whoever owns the layers AI has to pass through to reach people, money, and markets.
All the best,
Kim Isenberg
🧠 Baidu Claims ERNIE 5.1 Trained for 6% of Rival Cost
Baidu says ERNIE 5.1 delivers first-tier model performance while needing only 6% of the training cost usually associated with comparable frontier models. The headline is not just that China has another strong model, but that the economics of frontier AI may be changing: if Baidu can get competitive results with dramatically lower training cost, the moat shifts from raw capex toward data quality, training efficiency, and distribution.
👉 tl;dr: Baidu is positioning ERNIE 5.1 as a cost-efficiency shock to frontier AI, claiming near top-tier performance at a fraction of the usual training bill.
🍎 Apple Opens Siri to Rival AI Models
Apple is reportedly preparing AI Extensions for iOS 27, a system that would let users choose rival models like ChatGPT, Claude, Gemini, and Perplexity inside Siri, Writing Tools, and other Apple Intelligence features. That would mark a major strategic turn: instead of forcing Apple to win the model race outright, the iPhone could become the neutral distribution layer where the best assistants compete for default behavior.
👉 tl;dr: Apple may turn iOS into an AI model marketplace, making Siri less of a single assistant and more of a front door to whichever model users trust most.
💊 Novo Nordisk Turns OpenAI Into Pharma Infrastructure
Novo Nordisk and OpenAI are expanding AI from isolated drug-discovery experiments into a company-wide operating layer spanning research, manufacturing, supply chain, commercial operations, and internal workflows. The partnership includes workforce upskilling, strict governance and human oversight, and pilots across R&D, manufacturing, and commercial operations, with full integration targeted by the end of 2026. The partnership lands as OpenAI itself targets a fully automated AI researcher by 2028, Novo's bet suggests Big Pharma does not want to wait for that milestone but is already wiring the infrastructure to absorb it.
👉 tl;dr: Novo is not just using OpenAI to speed up lab work; it is trying to wire AI into the full pharmaceutical value chain from molecule to market.
When a big tech headline breaks, ask AI to explain what it actually changes for your job.
Why it helps: Most AI news sounds important but feels abstract. A quick, personalized translation turns noise into something you can act on — or safely ignore.
Try this:"Here's a headline: [paste headline]. I work as a [your role] in [your industry]. In two sentences, tell me: does this change anything for me, and if so, what should I do or watch?"
Cuts through the hype cycle and gives you a filter that works whether the news is about AI safety talks, new models, or agentic payments.
🎬 Watch This
Robotics' End Game: Nvidia's Jim Fan
Why it’s worth your time:
Jim Fan maps the next phase of AI as the move from chatbots into physical AI: robots trained through world models, simulation, synthetic data, and first-person human video rather than brittle hand-coded behavior. The useful frame is that robotics may follow the LLM scaling playbook, but with the hard bottleneck shifted from internet text to embodied data and real-world feedback loops.
Best bit:
The strongest part is Fan’s “Great Parallel” idea: instead of treating robotics as a bespoke hardware category, Nvidia is building the equivalent of a foundation-model stack for the physical world, from world simulators to robot policies to data flywheels like egocentric video. It makes humanoids feel less like a gadget race and more like the next compute platform.
Watch if you care about:
robotics / physical AI / world models / Nvidia / embodied agents / automation / the next AI platform
Dario Amodei, The Adolescence of Technology (January 2026). It captures the current AI mood perfectly: the debate is shifting from “what can models do?” to “are our systems mature enough to handle what we are giving them?”
The AI hardware race is getting weirdly real.
Analyst Ming-Chi Kuo says OpenAI is fast-tracking an AI-agent phone for mass production as early as the first half of 2027, with MediaTek reportedly picked as the silicon supplier and the device’s “headline spec” focused on camera sensing: an enhanced HDR pipeline so the AI can better understand the real world around you. Meanwhile, Apple is reportedly still working on an AirTag-sized AI pendant with cameras, microphones, Siri, and Visual Intelligence, possibly launching in 2027. The interesting part is not whether either gadget becomes the next iPhone; it is that the AI platform war is moving from apps into physical devices that can see, hear, remember, and act.
U.S. and China Talk AI Safety
The Takeaway
👉 Washington and Beijing are negotiating a formal, recurring AI safety dialogue to be announced at the May 14-15 Trump-Xi summit in Beijing, with Treasury Secretary Bessent leading the U.S. side.
👉 The Biden-era AI dialogue produced one key result, keeping humans in control of nuclear launches, but fell short on substance because China staffed its delegation with diplomats rather than technical experts.
👉 Private-sector channels led by former Microsoft CRO Craig Mundie and Chinese counterparts from Tsinghua University are already working on frontier-model safety and guardrail design, creating a foundation the governments can build on.
👉 The Trump administration is simultaneously pivoting toward AI safety, with the NEC floating an FDA-style approval process for frontier models, a sharp shift from earlier rhetoric dismissing safety concerns.
The world's two AI superpowers might be about to sit down and set some ground rules. Ahead of the Trump-Xi summit on May 14-15 in Beijing, Washington and Beijing are considering launching formal AI safety discussions. Think of it as the nuclear arms control playbook, but for artificial intelligence.
Treasury Secretary Scott Bessent is leading the U.S. side, while China has signaled openness through Vice Finance Minister Liao Min. The idea: a recurring dialogue covering risks like AI models behaving unpredictably, autonomous military systems, and misuse of open-source tools by bad actors. This isn't entirely new. Biden and Xi agreed in 2024 that humans, not AI, should control nuclear launch decisions. But experts say that earlier dialogue lacked depth because Beijing sent diplomats instead of technical experts.
Now, private-sector channels are already working on frontier-model safety, with former Microsoft executive Craig Mundie leading talks alongside Chinese counterparts from Tsinghua University and major AI labs. As one participant put it: the goal is stability, not alignment.
Why it matters: The two nations building the world's most powerful AI systems are acknowledging they need shared rules before something goes wrong. Whether this dialogue produces real safeguards or just diplomatic theater will shape the future of global AI governance.
Crash Expert: “This Looks Like 1929” → 71,105 Diversifying Here
Mark Spitznagel, who made $1B in a single day during the 2015 flash crash, warned markets are mimicking 1929. Seems extreme but we did just see the worst quarter for the S&P since 2022.
So it’s not so surprising that Vanguard and Goldman Sachs forecasted 5% and 3% annual S&P returns respectively for 2024-2034.
Late last year, Apollo’s chief economist Torsten Slok put it this way: "expect zero in return in the S&P 500 over the coming decade."
Almost no one knows this, but postwar and contemporary art appreciated 10.2% annually with near-zero correlation to equities from 1995–2025 overall.*
And sure… billionaires like Bezos can make headlines at auction, but what about the rest of us?
Masterworks makes it possible to invest in legendary artworks by Banksy, Basquiat, Picasso, and more – without spending millions.
29 exits. Net annualized returns like 16.5%, 17.6%, and 17.8% on works held over 1 year+. $1.3 billion invested. 500+ offerings.*
Shares in new offerings can sell quickly but…
*According to Masterworks data. Past performance is not indicative of future returns. Investing involves risk. Important Reg A disclosures: masterworks.com/cd.
The chart: Microsoft’s Q1 2026 AI diffusion leaderboard shows the UAE still far ahead at 70.1% of the working-age population using generative AI, while Singapore sits at 63.4% and the U.S. only reaches rank 21 at 31.3%. Germany is just behind at 31.1%, while South Korea makes the biggest visible jump in the top 25, rising from 30.7% to 37.1%.
The lesson: AI adoption is no longer just a Silicon Valley story. The most interesting AI race may now be diffusion: which countries make AI normal fastest across language, work, education, and everyday software.
The caveat: Microsoft’s measure is based on aggregated, anonymized Microsoft telemetry adjusted for device share, internet penetration, and population, so it is a powerful directional signal rather than a perfect census of all AI usage.
AI Agents Are Getting Wallets
⚡ Bottom line: Payment networks are racing to make AI agents financially useful: not just recommending what to buy, but discovering products, checking out, and triggering payments with identity, limits, and fraud controls built in.
💡 Why it matters: Finance is about to move from human-clicked transactions to delegated transactions, where banks and card networks become the trust layer between people, merchants, and autonomous software.
🔎 What it means: The next big AI finance battle may not be “which chatbot gives better advice,” but who owns the rails, permissions, and liability when an AI agent spends money on your behalf.
AI agents are crossing a line that matters for finance: from answering questions to moving money. Mastercard and Rabobank completed the first AI-agent payment in the Netherlands, booking a coffee tasting via Priceless.com after a user prompt, with the transaction handled through Mastercard Agent Pay and a Rabobank Mastercard credit card.
The shift is that “agentic commerce” changes who the payment system is designed around. Today, payments mostly authenticate a human clicking a button. The next version has to authenticate an AI agent, verify the user’s intent, enforce spend limits, handle refunds and disputes, and decide who is liable when the agent buys the wrong thing.
That makes banks and card networks unexpectedly central to the AI stack. Visa is rolling out Agentic Ready so issuers can test agent-initiated payments with live cards, tokenization, authentication, and controlled real-world flows. Mastercard is pushing a similar trust layer through Agent Pay, where the agent does not see raw card details and acts only within the consumer’s permission. The money question is simple: if AI agents become the interface for shopping, travel, procurement, banking, and treasury, the real platform may not be the chatbot. It may be the payment rail that can say, “yes, this agent is allowed to spend.”
Real-World Ads, Simple to Run
With AdQuick, executing Out Of Home campaigns is as easy as running digital ads. Plan, deploy, and measure your real-world advertising effortlessly — so your team can scale campaigns and maximize impact without the headaches.