In partnership with
Four stories today, and three of them are really the same story wearing different hats. The AI industry's cost structure is cracking open faster than anyone priced in. A Chinese lab just gave away, for free, a model that beats the best paid options at real coding work. Meanwhile IBM had its worst trading day on record, and the reason traces straight back to AI spending. Let's get into it.
Signed World Cup Legend Jersey. Every Day. Free.
For 10 days during the World Cup knockouts, Kalshi is giving away a signed, authenticated jersey from a different soccer legend every 24 hours. One winner per day. The roster: Di María, Ochoa, Dempsey, Modrić, Falcao, Varane, Ferdinand, Villa, and the grand prize: Messi's actual game-worn jersey from the match he scored his 800th career goal. It's free to enter in the Kalshi app. Every daily check-in earns you an entry toward Messi. The more days you show up, the better your odds. Download the app and start entering now.
Free to enter. App required. No purchase necessary. 18+, U.S. only (excl. NY & FL).
A Free Chinese AI Model Just Beat Claude at Coding
Moonshot AI, a Chinese lab, released a new model called Kimi K3 this week. It's a 2.8 trillion parameter model, and the full weights go public by July 27, meaning anyone will be able to download and run it themselves. On an independent front end coding benchmark, it ranked first, ahead of Anthropic's Claude Fable 5. On a broader test of real world work across dozens of industries it placed third overall, just behind Claude Fable 5 Max and OpenAI's GPT-5.6 Sol Max. And it's priced far below what the big US labs charge, about 30 cents per million tokens on a routine request versus several dollars for comparable premium models.
Why it matters to you: You don't need to know what a "mixture of experts" is. What you need to know is that the AI tools you're paying for got harder to justify overnight. If a nearly free open model can match or beat premium tools on real work, every AI vendor selling you a subscription now has to prove they're worth the premium instead of assuming it. Expect price pressure and expect this to keep happening every few weeks this year.
What to do about it: Before you renew any AI subscription this quarter, ask what you're actually paying for beyond the model itself. The model alone is getting commoditized fast.
The Ex-OpenAI CTO's New Company Just Shipped Its First Model
Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, released its first public model this week, called Inkling, according to TechCrunch. It's a 975 billion parameter model that only uses about 41 billion of those at once to stay fast and cheap to run, and it's open weight, meaning any company can download and customize it themselves. The company isn't claiming it's the best model out there. Its own materials call it "not the strongest model available today, closed or open." Instead, the bet is that businesses who fine-tune their own model will beat the one-size-fits-all AI that OpenAI, Anthropic, and Google are selling everyone.
Why it matters to you: This is the second story today about an AI company openly admitting that biggest and most expensive isn't the same as best for you. Microsoft's own CEO said something similar last week: if you're renting someone else's AI model, you're paying twice, once for the subscription and again by handing over your own business knowledge in every prompt you type. That's worth sitting with if you've been feeding a chatbot your client details, pricing strategy, or internal processes over the past year.
What to do about it: If you're using AI for anything sensitive, ask your vendor directly where your data goes and whether it trains future models.
OpenAI's First Product You Can Hold In Your Hand Costs $230
OpenAI launched its first piece of hardware this week, a $230 keypad called Codex Micro, built with keyboard maker Work Louder, per Engadget. It has 13 mechanical keys, a joystick, a dial, and six lights that change color to show what OpenAI's coding assistant, Codex, is doing at any given moment. It's meant to sit next to your keyboard so a developer can approve code, switch tasks, or dial up how hard the AI "thinks" without touching a mouse.
Why it matters to you: You probably don't need this specific gadget. What it tells you is where OpenAI thinks this is going. AI agents that run in the background all day, need a physical way to check on and control, and are common enough that a company will build custom hardware just to manage them. If your team isn't using AI agents yet, the gap between "AI as a chat window" and "AI as a coworker you check in on" is closing faster than most business owners have noticed.
IBM Just Had Its Worst Trading Day Ever, and AI Is Why
IBM's stock dropped 25% in a single day this week, its steepest one day fall on record, wiping out roughly $67 billion in value, according to Invezz. CEO Arvind Krishna told investors the company "faltered," and the real story underneath is that IBM's own clients pulled spending away from software and consulting so they could buy AI servers, storage, and memory before prices go up. Software revenue actually grew 5%. Infrastructure spending, the traditional stuff, fell 7% as that money got redirected.
Why it matters to you: This is the clearest sign yet that AI isn't just changing what software does, it's changing where company budgets go entirely, even at a company as established as IBM. If your own vendors or suppliers are quietly shifting spending toward AI infrastructure, expect ripple effects: price increases on the hardware and cloud services everyone depends on, and possibly some belt tightening from vendors who bet wrong. Keep an eye on your own tech stack's pricing over the next two quarters.
The Bottom Line
Here's the pattern underneath all four stories today: the AI industry is quietly admitting that bigger and more expensive isn't the same as better, and the money is already moving because of it. A free model is beating premium ones at real work. The company founded by an ex-OpenAI executive is betting against OpenAI's own playbook. And IBM just took a $67 billion hit because its own customers decided AI hardware mattered more than the software IBM has been selling for decades. None of this means you need to rip out your tools tomorrow. It means the ground is shifting under every AI subscription and every AI vendor pitch you'll hear this year, and the businesses that ask hard questions about what they're actually paying for will end up ahead of the ones who don't. Stay skeptical. Stay curious. That's the whole point of this newsletter.
Enjoying the Ride?
If this saved you from digging through five different AI newsletters just to figure out what actually matters, forward it to one person who needs to stop guessing about AI too. Someone forward this to you? Subscribe so the next issue lands straight in your inbox.
Talk tomorrow,
Mark Shilensky
Follow me personally on Social Media:
Facebook: https://www.facebook.com/MarkShilenskyPage
Instagram: https://www.instagram.com/markashilensky/
Threads: https://www.threads.com/@markashilensky
LinkedIn: https://www.linkedin.com/in/shilensky/
X: https://x.com/markshilensky
TikTok: https://www.tiktok.com/@mark.shilensky
BlueSky: https://bsky.app/profile/markshilensky.bsky.social