In partnership with
Sam Altman is in the news again, and not for good reasons.
This time it's about RAM.
On a podcast last week, he called it "quite terrible" that so much of the world's limited RAM sits in people's own computers, where it gets used about 5% of the time, while the cloud can keep it busy.
So if you’ve got 32GB sitting in a gaming PC at home, someone in a data centre has a few questions for you 👀
The internet did not take it well but scroll through enough of those replies and a bigger question shows up.
If more and more of what your laptop does runs on someone else's server, do you even own your computer anymore??
I'll get into that, but first, here's what happened this week.
Altman flew to Korea to meet the people who make RAM.
He met around 50 Samsung executives in Yongin on Saturday and then had lunch with SK Group's chairman. According to The Korea Herald, he's chasing two things: enough memory chips, and the money to build the data centres they go into.
ChatGPT now limits how many paid accounts can run on one device.
Go, Plus and Pro users are hitting a "Device limit reached" error when too many paid accounts sit on one device or card, and OpenAI hasn't said what the limit is.
DeepSeek open-sourced a "proactive" coding agent.
Its Super-agent watches your team's GitHub repo (and, optionally, its Feishu chats) and picks up coding work when someone asks or when rules your team wrote allow it. It's out on GitHub as an MIT-licensed research prototype.
S&P just put Meta's AI spending on notice.
It cut Meta's outlook to negative, expecting about $71B of negative free cash flow in 2027 as data centre spending climbs, which means Meta would be burning way more cash than its business brings in. S&P says it's unclear when the AI bets will pay off.
OpenAI reportedly cut its funding target from $30B to ~$20B, with roughly $11B committed so far. → Read more
iQIYI says it opened China's first AI film studio in Shanghai, and its first feature is already shooting. → Read more
In Louisville, people opposing a proposed hyperscale data centre marched to the mayor's house on Sunday. → Read more
A $109 crowdfunded box called RAMDeck wants to pool the RAM from your old laptops and phones to run bigger AI models. TechRadar points out that the software underneath is free, so make of that what you will. → Read more
Okay, back to those replies.
One guy runs Hello World on 64GB and wants Altman to know it was him 😂
Barely anyone in there was arguing about efficiency. The general mood was "this is my machine, keep your hands off it."
A lot of that anger is aimed at OpenAI's RAM deals, so did OpenAI go and sign for all that RAM?
Sort of.
On October 1, 2025, Samsung and SK hynix signed letters of intent with OpenAI for its Stargate data centres, for up to 900,000 DRAM wafers a month.
Tom's Hardware puts that at roughly 40% of the world's supply, and these two companies make about 70% of the world's DRAM between them.
Now, a letter of intent is not a purchase order, it's closer to "we'll probably take it".
But contract DRAM prices shot up soon after anyway, and analysts valued those wafers at more than ₩100 trillion (about $70B).
OpenAI didn't have to buy a single chip to change what you pay for RAM, and that's what has people so worked up.
When you own your laptop, it doesn't need anyone's permission.
You can switch off the wifi or let a subscription lapse and it'll still open your files and run whatever you installed, and no usage cap or price hike can take that away.
A machine that leans on the cloud only works for as long as someone else's servers, pricing and rules allow it to.
The catch is that owning is getting expensive.
A 32GB DDR5 kit was going for around ₹60,000 in August, and laptop makers are bringing back 8GB machines.
An 8GB laptop can't run AI on its own, so everything ends up going through someone's cloud.
Nobody has to ban your RAM. If prices stay this high, "cloud-first" slowly turns into "cloud-only" for most of us without anyone announcing it.
To be fair, Altman isn't completely wrong.
He brought it up while answering a question about why sky-high memory prices haven't pulled in new chipmakers, and gaming PCs never even came up.
On the maths, he has a point too.
A server shared by thousands of people keeps its memory far busier than your laptop ever will, and the best models are way too big to run on a home machine anyway.
If you use AI for 20 minutes a day, a subscription is honestly cheaper than buying a 64GB laptop at today's prices.
The replies were making a different argument though, about who gets to control your machine, and on that, the guy selling the cloud isn’t the most neutral person to ask.
A free, open-source Mac app from the team behind Draw Things. You download a model like Qwen or DeepSeek and it works on your files without sending them to anyone's cloud, as long as you stick to local models (very on-theme for today).
You can edit a recording by deleting sentences from its transcript, and it'll offer to strip out filler words for you. The new camera cutout (still in beta) removes your background without a green screen, right on your Mac.
AI agents that prep, edit, clip and schedule your videos, and nothing goes live until you approve it.
Keep it local: client documents, anything about your money or health, and drafts you'd rather nobody trained on.
Use the cloud: deep research and heavy video or image work.
This week: install one local tool, switch off your wifi and run one task, then see what still works.
Buying a laptop? Treat 16GB as the bare minimum.
Call me old school, but I'd still like to own the computer sitting on my desk.
So tell me, if your laptop could only do AI through someone else's server, how long would it take you to notice? Hit reply.
Until next time,
Vaibhav 🤝🏻
If you read till here, you might find this interesting
#AD 1
The AI Work Handbook That Cuts Your Workday in Half
The 8-hour workday is becoming a 4-hour workday for people who know how to use AI.
Everyone else is still catching up.
This AI work playbook shows you exactly how to cut your work hours in half using AI.
Sign up for Superhuman AI and get:
50+ step-by-step AI tutorials to cut your workload in half — covering every part of your workday, from emails to strategy, used by 1M+ professionals at Google, Microsoft, and NASA
Superhuman AI newsletter (4 min daily) so you keep discovering new AI tools and skills to stay ahead in your career — the playbook is just the start
#AD 2
When Postgres analytics slow down, most teams add a second database. Then comes sync lag and drift. TimescaleDB extends Postgres instead: analytics on live data, no pipeline. Get $1000 to start.
