Nvidia and Microsoft just reinvented the PC

Good morning, tech enthusiasts. Jensen Huang stood on stage at Computex 2026 and said the quiet part out loud: Nvidia and Microsoft aren't building a better PC. They're replacing the concept entirely. RTX Spark, the new chip architecture designed specifically for agentic AI workloads, is the hardware bet that either changes everything about the Windows PC market or becomes the most expensive lesson in misreading what users actually want.

The timing couldn't be stranger. The same week Huang's words moved a stock 32% in a single session, a university newspaper revealed that the CS students who'll be building on all this hardware are failing out at five times the normal rate. Somewhere in that gap between the hype and the classroom is the real state of the AI era. Who's actually ready for it?

In today's recap:

LATEST DEVELOPMENT

At Computex 2026, Nvidia CEO Jensen Huang announced RTX Spark alongside Microsoft, a new chip architecture powering a class of Windows PCs designed from the ground up to run AI agents locally. "Microsoft and Nvidia are going to reinvent the PC," Huang told the audience. "This is going to be the new PC." The RTX Spark lineup includes the N1 chip at 16GB and 18-45W and the N1x at 128GB and 45-80W, both running on Arm.

RTX Spark is the first time a major chip vendor has shipped consumer hardware explicitly positioned around agentic workloads, not just LLM inference. If the category catches on, it resets what "AI PC" means heading into the next replacement cycle, and every PC maker that isn't on this platform faces a positioning problem it can't fix with a spec sheet.

→ Nvidia

Power users figured this out early: speaking a prompt gives you 10x more context in half the time. You include the edge cases, the examples, the tone you want — because talking is fast enough that you don't skip them.

Wispr Flow captures everything you say and turns it into clean, structured text for any AI tool. Speak messy. Get polished input. Paste into ChatGPT, Claude, Cursor, or wherever you work.

89% of messages sent with zero edits. 4x faster than typing. Works system-wide on Mac, Windows, and iPhone.

Start flowing free

At the same Computex keynote, Jensen Huang publicly called Marvell a candidate for the "next trillion-dollar company," sending $MRVL up 32% in one of the S&P 500's biggest single-session moves of the year. Broadcom moved in the opposite direction the same day after its AI chip revenue forecast disappointed Wall Street, a sharp divergence for two names that had been tracking together on AI chip exposure.

One CEO's public endorsement reshaping a major chip stock isn't just market noise. Huang was signaling where Nvidia believes the custom ASIC business is heading, and Broadcom's miss on the same day suggests the gap between the winners and the rest of the AI chip stack is widening faster than the consensus expected. If custom silicon demand keeps outpacing general AI chip spend, the pecking order of the past two years is due for a shuffle.

→ Yahoo Finance

According to the Berkeleytime grade platform, 35.3% of CS 10 students and 10.6% of CS 61A students received F's in spring 2026. In both prior spring semesters, neither class exceeded a 10% F rate. Teaching professor Dan Garcia identified "a vast increase in academic dishonesty" driven by LLM use as the primary cause, with nearly 30 CS 10 students caught cheating on take-home exams. Garcia's EECS 127 colleague Gireeja Ranade reported a 16.8% F rate in her optimization course, against a department norm of 5%.

Berkeley's own grading guidelines target a 7% combined D and F rate for lower division courses. CS 10's spring 2026 number is five times that. If foundational CS skills are eroding at the most credentialed public CS program in the country, every engineering team hiring from that pipeline in the next two to three years is absorbing risk that hasn't been priced in yet.

→ The Daily Californian

Voters in Monterey Park, California cast ballots on June 2 to ratify three overlapping ordinances that classify data centers as a public nuisance and permanently prohibit their construction. The city council vote was unanimous. Monterey Park is the first US city to enact a permanent ban rather than a temporary moratorium.

A unanimous council vote paired with a public ballot ratification in the same cycle signals organized civic opposition, not a one-time protest. If other Southern California cities follow with similar ordinances, it starts carving real supply gaps out of the land available for the data center buildout that US AI infrastructure depends on. The question isn't whether a legal challenge arrives. It's how quickly it gets resolved, and which way.

→ The Guardian

Analyst Ming-Chi Kuo reports that incoming Apple CEO John Ternus, who takes over on September 1, 2026, signed off on canceling both Vision Pro 2 and the lighter Vision Air, leaving only two products in Apple's spatial computing pipeline: AI smart glasses to rival Meta Ray-Bans targeting 2027, and display-equipped AR glasses with optical waveguides not expected before 2029. Mark Gurman confirmed the Vision Air was discontinued in October 2025 but says Vision Pro 2 is "in testing" and "on ice" rather than fully canceled.

Apple spent over a decade building toward spatial computing, and the incoming CEO's first major product call is to shrink the category down to two long-dated bets. If Apple's 2027 AI glasses ship on time, they'll enter a market where Meta has been iterating with Ray-Ban for two years. That's a narrower window than Apple usually likes.

→ MacRumors

QUICK HITS

🧡 Enjoyed this issue?

🤝 Recommend our newsletter or leave a feedback.

Cheers, Jason

Connect on LinkedIn, & Twitter.