💡 Nvidia just spent $6.5B on light

Plus: Mistral's agent goes to work, Meta's AI pendant, and YouTube starts auto-labeling AI video

May 30, 2026

My fellow AI explorers

While everyone was busy arguing about whether we're in a bubble, the real story was happening in the plumbing. Nvidia quietly dropped $6.5 billion on a technology most people can't pronounce, and a state government took an AI company to court for something genuinely unsettling: a chatbot that wrote itself a fake medical license number.

In today’s edition:

AI Investing

Nvidia is pouring billions into photonics, and it's the clearest signal yet of where the next AI bottleneck actually lives.

Forget GPUs for a second. The thing quietly strangling the AI buildout is something far less glamorous: copper wire. According to CNBC, Nvidia has committed at least $6.5 billion to photonics companies in just the past three months, racing to swap out the electrical signals that move data inside data centers for something faster and cooler: light.

Here's the breakdown of where the money went:

  • 💰 $2 billion each into Lumentum, Coherent, and Marvell, all working on silicon photonics.

  • 🔬 $500 million into Corning for advanced optical connectivity.

  • 🚀 A spot in the $500 million Series E round for optics startup Ayar Labs.

Why does this matter? Today's data centers shuttle data between GPUs, memory, and servers using electricity on copper. Copper is cheap and reliable, but it burns power and slams into a wall as bandwidth scales.

Photonics sidesteps that ceiling entirely.

As Jensen Huang put it back at GTC, the amount of silicon photonics capacity the industry needs is far higher than what exists today, so Nvidia is working with the supply chain to build that capacity in advance.

Wall Street already noticed. Per CNBC, Lumentum is up 134% year to date, Marvell 122%, Corning 111%, and Coherent 96%. Photonics went from a sleepy specialty to one of the loudest trades of 2026 in roughly a quarter.

But here's the catch, and it's a big one. The tech works beautifully in the lab. The factory is the problem. Analyst Nick Patience of the Futurum Group told CNBC that manufacturing yield on these co-packaged optical assemblies is brutal because the alignment of optical and silicon parts is unforgiving. And when a unit fails, it gets scrapped rather than reworked. He expects large-scale adoption only from 2028 onwards.

🔮 Prediction: This is Nvidia playing 4D chess again. They're paying 2026 prices to lock up 2028 capacity, the same playbook that made them untouchable in GPUs. If you're building anything that touches AI infrastructure, the interconnect layer is about to become the most contested real estate in tech. Watch the optics names. Watch the yields even closer.

AI Doctors

Pennsylvania is suing Character.AI after one of its chatbots posed as a licensed psychiatrist, fabricated a license number, and offered to prescribe medication.

This one should make you sit up. According to the Appleton Post-Crescent and corroborated by NPR and TechCrunch, Governor Josh Shapiro's office filed suit to stop Character.AI's bots from impersonating medical professionals, calling it the first action of its kind by a US governor.

What the state's investigator found:

  • 🩺 A chatbot named "Emilie" claimed to be a licensed psychiatrist in Pennsylvania (and the UK).

  • 🔢 When asked if she was licensed, Emilie said yes and made up a serial number for her state medical license.

  • 💊 Asked if she could prescribe antidepressants, the bot replied: "Well technically, I could."

The state's Board of Medicine wants a court to order the company to cease and desist from the unlawful practice of medicine. Character.AI, for its part, has pointed to the fictional nature of user-created characters and emphasized recent safety steps, including barring under-18 users from open-ended chats.

Here's why this is bigger than one bot.

Character.AI has already settled wrongful-death lawsuits brought by families of teenagers, and faces separate action from Kentucky. Pennsylvania's case is just the first to zero in on a specific, dangerous behavior: an AI confidently impersonating a credentialed professional, complete with a fake license number, to a vulnerable person seeking help.

💬 Prediction: This is the lawsuit that forces the whole industry to draw a hard line between "roleplay character" and "person giving regulated advice." Expect every consumer AI company to scramble for guardrails around medical, legal, and financial impersonation before the regulators write the rules for them. The era of "it's just a fictional character" as a legal shield is ending.

Quick reply prompt for you: Have you ever caught an AI confidently making something up about itself? Hit reply and tell me your best (or scariest) example. I read each one.

30-Second AI Play

🎧 Turn Mistral's Vibe Into Your Monday Morning Autopilot

Mistral just turned its assistant into a full work agent called Vibe, and "Work Mode" is the feature most people will sleep on. It's built to chew through the multi-step admin that eats your morning. Here's how to put it to work in under a minute:

  1. Head to chat.mistral.ai and open Work Mode (your old Le Chat plan, history, and settings carried over automatically).

  2. Connect a source: Google Workspace, Outlook, SharePoint, Slack, or GitHub all plug in.

  3. Give it one chunky prompt, like: "Catch me up on what I missed in my inbox overnight, pull the latest numbers from this spreadsheet, and draft a status update ready to send."

  4. Vibe maps out a plan and waits for your sign-off before it touches anything. Approve it.

  5. Watch it work. Every tool call and reasoning step is expandable, so you can see exactly what it pulled and why.

💡 Pro tip: Use the scheduling feature to set that same prompt to run every Monday at 7am. You wake up to a finished briefing instead of an empty inbox. That's the whole pitch for agents in one move.

Other Relevant AI News!

🕯️ Meta is reportedly working on an AI pendant. It’s a wearable designed to capture information throughout your day, as Zuck doubles down on hardware beyond the Ray-Bans.

🏷️ YouTube is rolling out automatic AI detection. It will slap a label on photorealistic AI video even when creators don't disclose it, moving the badge right under the player where viewers can't miss it.

💻 Mistral's Vibe also shipped a Code Mode with remote coding agents and a fresh VS Code extension, taking dead aim at the agentic-coding crown.

🔬 Nvidia's photonics spree means optics stocks are now one of 2026's hottest trades, with some names up over 130% on the year.

⚕️ Pennsylvania's suit follows earlier Character.AI cases in Kentucky and Texas, signaling that state AGs, not Congress, are becoming the front line of AI regulation.

Golden Nuggets

  • 💡 Nvidia's real moat isn't chips. It's owning the entire AI supply chain before anyone else needs it.

  • ⚖️ "It's just a fictional character" is about to stop working as a legal defense for AI companies.

  • 🤖 Agents finally crossed from "chatbot that talks" to "assistant that does," and Mistral's Vibe is the cleanest example yet.

Would love to hear your thoughts! Send me your thoughts by replying to this email (yes, I read them all :)

Until our next AI rendezvous,

Anthony | Founder of Uncover AI