Flavor of the week? More like flAIvor of the week… never mind. Ahem, it’s your boy, Eric. This week ran on memory chips and electricity. Sam Altman spent November 3rd on X saying he does want you to have RAM, a few days after a clip of him calling home-computer RAM a poor use of the world’s supply made the rounds. Memory makers say the shortage behind all of this gets worse before it gets better. On Tuesday, voters in dozens of towns weighed in on the data centers driving that demand, and nearly all of them basically said “pause”.
The short of it all: Altman says his comments were taken out of context and that the world should build more memory. Team Group’s CEO says supply won’t balance until the second half of 2028 or later, and South Korea’s chip exports rose 259% in October. Voters approved limits on data centers in 37 of the 38 ballot measures called so far. OpenAI’s latest funding round is reportedly coming in short of its original $30 billion target. And Google’s Project Suncatcher team got its first AI response from a TPU in orbit, in the form of a haiku. Let’s get into it.
Here’s what’s on the menu today:
Sam Altman says his RAM comments were taken out of context and that he wants you “to have so much RAM”
Team Group’s CEO expects the memory shortage to worsen through 2027, as South Korea’s chip exports hit $56.4 billion in October
Voters approved limits on data centers in 37 of the 38 local ballot measures called so far
OpenAI reportedly cut its funding target from $30 billion to $20 billion, at a $1.25 trillion valuation
Google’s Project Suncatcher satellite ran Gemma 4 on a TPU in orbit, while a Google executive put space data centers at up to 100 times the cost
Altman says he wants you to have “so much RAM”
The story: In an interview on Cheeky Pint, the show hosted by Stripe co-founder John Collison, OpenAI CEO Sam Altman (in full “foot in mouth” mode) said it’s “quite terrible” for so much of the world’s limited supply of DRAM (the main working memory inside computers, phones and AI servers) to sit in home computers that are “being used 5% of the time.” Naturally, almost everyone who uses RAM on a daily basis universally threw him under the bus, severely flaming him online. On November 3, Altman posted on X that his comments had been “taken out of context.”
What happened:
Altman’s original point was that “you want maximum utilization in the cloud” when memory supply is this tight
In his November 3 posts, Altman wrote, “I actually do want the world to have tons of memory and people to put as much as they want in their computers”
He added that he has “been arguing for the world to ramp up memory production for a long time”
When a user asked why he wanted their RAM, he replied, “I want you to have so much RAM. Let’s build more.”
He also changed his X display name to “Ram Altman” as a joke
Why it matters:
Altman kept his efficiency argument and paired it with a call for more memory production, and TechRadar noted that new memory factories take time to come online
OpenAI has its own stake in that supply, since Samsung and SK hynix signed letters of intent in October 2025 for Stargate, OpenAI’s data center project, which OpenAI projected could need up to 900,000 DRAM wafers a month
Tom’s Hardware pointed out that local AI could eventually cut into demand for paid subscriptions, and that OpenAI’s main business is models hosted in data centers (so, of course, we all know where Sammy Boy’s DRAM priorities lie)
Memory makers say the shortage gets worse before it gets better
The story: SPEAKING OF DRAM, on its November 2 earnings call, Team Group (sounds like a generic-sounding superhero squad), a Taiwanese maker of memory modules and SSDs, said the memory shortage will worsen from the end of 2026 through the end of 2027. CEO Chen Ching-wen said supply and demand won’t balance until the second half of 2028 or later. The same week, South Korea, home to Samsung and SK hynix, reported record exports led by chips.
What happened:
Chen said some buyers could reach a point where “even money can’t buy goods,” with spot prices (chips bought on the open market right away) running 30% to 50% above contract prices (prices locked in under supply agreements)
“We have contracts but can’t get enough volume,” he said, so Team Group has to buy on the more expensive spot market
Contract prices for 16GB of DDR5 rose from $8.70 in October 2025 to $48 in September 2026, an increase of about 450%
South Korea’s October exports hit a record $115.3 billion, up 93.7%, and semiconductor exports rose 259% to $56.4 billion
TrendForce, a market research firm, expects DRAM contract prices to rise about 20% in the fourth quarter, after gains of about 40% in the third and about 80% in the second
Why it matters:
Acer CEO Jason Chen sees it differently. He expects PC prices to rise 5% to 20% through the end of 2026, level off in the first half of 2027 and then decline, and he says major suppliers are hyping the shortage to protect margins
TrendForce attributes the slower price gains to a higher starting point, weaker consumer buying power and price caps in long-term supply agreements
Chips (and no, not those once-viral Turtle Chips) made up nearly half of South Korea’s October exports, and the trade ministry says full-year exports could reach $1 trillion as early as November
Voters said no to data centers in nearly every ballot measure
The story: On November 3, voters in 10 states decided 41 local ballot measures about data centers, the large buildings full of servers that run AI models and cloud services. As of midday November 4, 38 had been called. Voters limited data center development in 37 of them and backed it in one, according to a tracker kept by State Capital Lobbyist. No statewide measure made the ballot, nor was Mayor McCheese re-elected in McDonaldland (beat, once again, by Grimace who maintains his hardline milkshake policy).
What happened:
Ohio held 21 of the measures in 18 communities, with bans passing in places like Granville (82%) and Trenton (80%)
In Illinois, all five questions asking whether to permit or approve data centers lost, including Joliet’s, where 21% voted yes
Outside Ohio and Illinois, bans or moratoriums passed in California, Florida, Kansas and Maine
The one pro-development result came in Harwood, North Dakota, where only 25% voted for an extra review process for industrial projects including data centers
Ohio’s proposed statewide ban on data centers above 25 megawatts missed its July signature deadline, and backers are now aiming for November 2027
Why it matters:
Local bans have slowed only a small part of the buildout so far: Tom’s Hardware reports that of more than 300 local bans enacted across the U.S., three projects totaling 2.3 gigawatts have been delayed, and about 38 gigawatts is still expected online by 2027
On November 2, OpenAI launched Grid Protectors (again with the generic-sounding superhero squad names), a program where its data centers agree to cut power use when local grids are strained, starting with up to 1,000 megawatts at a planned 3.2-gigawatt campus in Effingham County, Georgia
More votes are on the calendar, including two pro-project bond referendums in Clarksville, Arkansas on December 1
OpenAI’s latest funding round is coming in short
The story: On October 30, The Wall Street Journal reported that OpenAI had about $11 billion committed toward a funding round it originally aimed to make $30 billion. The next day, CNBC reported that OpenAI cut the target to up to $20 billion. The Financial Times put the round’s post-money valuation (what the company is worth once the new money is counted) at $1.25 trillion, below the roughly $1.4 trillion OpenAI had hoped for. Aw beans, I guess Sam won’t be able to afford all the RAM after all.
What happened:
OpenAI first sought at least $30 billion at about a $1.4 trillion valuation, according to reports in late September
CNBC reported that some investors chose to reduce their allocations, and that the round hasn’t closed
CNBC’s earlier reporting named investors from the United Arab Emirates and Coatue, an investment firm, as participants
CNBC tied the smaller target to a broad tech selloff and investor doubts about returns on AI spending, including Alphabet’s worst trading day ever on October 29
OpenAI didn’t immediately respond to CNBC’s request for comment and hasn’t confirmed the figures
Why it matters:
Investors are asking the same question of everyone this fall, which is whether hundreds of billions in AI spending will pay off
The same week, Anthropic’s amended S-1 showed about $17 billion in third-quarter revenue and a $567 million operating loss, with growth slowing to about 47% from the prior quarter
The money funds the same buildout behind the first three stories, from the Stargate (like the movie, Stargate??) memory deals to the Georgia campus piloting Grid Protectors
Google’s test satellite wrote its first haiku in orbit
The story: On October 30, Google said its Project Suncatcher prototype satellite sent back its first AI response from orbit. The satellite carries Google’s TPUs (tensor processing units, the chips Google designs for AI work) and ran Gemma 4, Google’s open model. Project Suncatcher is a research effort Google announced in November 2025 to test whether solar-powered satellites could someday host AI computing (or, spAIce computing?… okay, that was worse than flAIvor).
What happened:
The satellite launched October 1 on SpaceX’s Transporter-18 rideshare mission and was built with Planet, a satellite company
After powering up the satellite’s systems, the team uplinked a prompt asking for a haiku about being in space, and the reply came down as the satellite passed over Svalbard, Norway
The haiku: “Blue marble below / Data streams in silent grace / Starlight feeds my core”
Travis Beals, senior director of Paradigms of Intelligence at Google, wrote that “Our chips have officially earned their space wings”
Next up are tests of how the TPUs handle heat and radiation in orbit, followed by a launch of two interconnected satellites in 2027
Why it matters:
Google’s head of data center energy told The Information that space data centers currently cost up to 100 times more than ones on Earth
Google’s 2025 research assumes launch prices fall to about $200 per kilogram by the mid-2030s, the point where space-based energy costs become roughly comparable to Earth’s
The same executive said U.S. grid supply is adequate and permitting is the bottleneck for building on the ground (but, hey man, having a space datacenter would be sci-fi levels of bonkers so, quite literally, shoot for the stars!)
What ties it all together
Every story this week runs into the same two limits: memory and electricity. Team Group sees the memory shortage lasting into 2028, and Sam Altman’s answer is building more of it. Voters approved limits in 37 local ballot measures on the data centers that use that memory and power. OpenAI is still raising money for its buildout at a lower target than it started with, and Google is testing whether some of that computing can leave the planet entirely. And here I am, just trying to figure out the best mac & cheese recipe in existence. Priorities.
Building more is the answer everyone keeps reaching for, whether that means memory factories, data centers or satellites (or mac & cheese). Each one takes time, and each one needs a place to go (for mac & cheese, it’s my tum tum), whether that’s a factory site, a town that votes yes or a spot in orbit. Ambitious or out of reach? We’ll see, as the craziness that is AI continues to unfold.
What I’m watching
Whether Altman or OpenAI follows “Let’s build more” with an actual memory investment or new supply deal
Where DRAM contract prices land in the fourth quarter against TrendForce’s forecast of about 20%
Whether OpenAI closes its round at $20 billion, and whether the rumored GPT-6.2 models ship as early as November 6
What the Suncatcher team learns from heat and radiation testing before the twin-satellite launch in 2027
Thanks so much for reading this edition of The Weekly Think.
See you next time, fellow thinkers!