The 12-month window for AI startups is closing, and Vercel got hacked
Foundation model companies are expanding into categories that AI startups currently occupy, and many founders privately admit their window is about 12 months. Meanwhile, Vercel confirmed a hack by ShinyHunters that exposed employee data. The global RAM shortage driven by AI demand may not ease until 2030, which means hardware costs stay high. And Simon Willison built a handy Claude token counter that compares costs across models.
AI startups face a 12-month window before models eat their lunch
TechCrunch reported that many AI startups exist only because foundation model providers haven't expanded into their specific category yet. Founders privately acknowledge this gap won't last forever. As models from OpenAI, Anthropic, and Google get broader capabilities, the thin-wrapper startups built on top of them face an existential timeline of roughly 12 months.
Sources: TechCrunch
Why this matters to you
If you're paying for an AI tool that's essentially a wrapper around ChatGPT/Claude/etc with a nice UI on top, that tool might not exist in a year. The foundation models keep shipping features that make middleware unnecessary. This is actually good news for zero-human operators: fewer subscriptions, more capability built into the tools you already use.
I've been saying this for a while. If your favorite AI tool is just a prompt template on top of someone else's model, it's on borrowed time. Build your workflows on the foundation models directly whenever you can.
Jason
Vercel hacked by ShinyHunters, employee data exposed
Vercel confirmed it was compromised by a person claiming to be a member of ShinyHunters, the group behind the Rockstar Games hack. Stolen data includes employee names, email addresses, and activity timestamps. The hackers are attempting to sell the data online.
Sources: TheVerge
Why this matters to you
A huge number of solopreneurs and vibe coders deploy on Vercel. If you have a Vercel account, your data could be part of this breach. Even if customer project data wasn't compromised, employee-level breaches often lead to phishing attacks and downstream supply chain issues. This is a reminder that your deployment platform is a single point of failure.
This is the downside of everyone converging on the same platforms. Vercel is fantastic, but when one platform hosts a massive chunk of the indie web, a single breach has outsized blast radius. Diversify your critical infrastructure if you can.
Jason
RAM shortage could last until 2030, keeping hardware costs high
The Verge reported that even as Samsung, SK Hynix, and Micron ramp up DRAM production, manufacturers are only expected to meet 60 percent of demand by the end of 2027. SK Group's chairman said shortages could persist until 2030. AI workloads are the primary driver of the demand spike.
Sources: TheVerge
Why this matters to you
Memory manufacturers will only meet 60% of global DRAM demand by end of 2027. That means laptop prices, server costs, and cloud compute pricing all stay elevated. For anyone running a zero-human business, this reinforces why API-based AI tools are smarter than trying to run models locally. Let the big providers absorb the hardware cost pressure.
This is one of those stories that sounds abstract until your next laptop costs $400 more than expected. For most solopreneurs, the move is clear: stay on cloud APIs and let OpenAI and Anthropic fight over GPU and memory allocation. That's their problem, not yours.
Jason
Simon Willison ships a Claude token counter with model cost comparisons
Simon Willison published a Claude token counter tool that lets you paste in text and see token counts across different Claude models, with cost comparisons. He also published a detailed breakdown of the system prompt differences between Claude Opus 4.6 and 4.7, revealing how Anthropic tweaks model behavior between versions.
Sources: HackerNews · HackerNews
Why this matters to you
If you're using the Claude API for automations or apps, knowing your token costs across models is the difference between a $50/month bill and a $500/month bill. The system prompt comparison also reveals that Anthropic meaningfully changed how Opus behaves between 4.6 and 4.7. If your prompts feel different lately, this is why.
Simon Willison is the single most useful person to follow if you use Claude for anything. These tools save real money. If you're spending more than $20/month on Claude API calls, spend 10 minutes with the token counter. You'll probably find savings.
Jason
Quiet Sunday, but the takeaways are practical. Audit your AI tool stack before the foundation models eat your wrappers, lock down your Vercel account, and bookmark Simon Willison's token counter. Small moves, real money saved.
Frequently asked
Was Vercel hacked in April 2026?
Yes. Vercel confirmed a breach by a person claiming to be part of ShinyHunters. Stolen data includes employee names, email addresses, and activity timestamps. The hackers are attempting to sell the data.
How long will the RAM shortage last?
According to SK Group's chairman, the RAM shortage could persist until 2030. Manufacturers are only expected to meet 60% of DRAM demand by the end of 2027, driven primarily by AI workload growth.
What changed between Claude Opus 4.6 and 4.7?
Simon Willison documented the system prompt differences between Claude Opus 4.6 and 4.7. Anthropic made meaningful changes to how the model behaves between versions, which explains why some users noticed different outputs after the update.
Teach your AI agent to run your business.
5 premium skills for OpenClaw, Claude Code, and other SKILL.md-compatible agents. Automation audits, tool stacks, content engines, email systems, and operations. $49 one-time.
Get the Skills Bundle