@jukan05 SK Hynix: “HBM4 next year.”
Nvidia: “We needed it yesterday.”
DRAM investors: “Thanks kings.”
The AI memory shortage just got a little more spicy
@SharesGem@aleabitoreddit Yes. $CCXI is the pre-IPO ticket to $AGLT — same company, same Digit robots already clocking real hours, just before the ticker flip
@jvance80 Most people panic-sell the dilution dip. You’re adding with a 7–10 year horizon while CIFR pivots hard into AI data centers. That’s the trade
$NVDA IS PUSHING GLASS SUBSTRATES TO EXTREME THICKNESSES
According to Korean semiconductor publication KIPOST, NVIDIA is requesting glass substrates with a thickness of 2,000 µm (2 mm), potentially driving the adoption of new metallization technologies, including atomic layer deposition (ALD).
As AI packages grow larger to accommodate more GPUs and HBM, substrate rigidity becomes increasingly important. Thicker glass improves mechanical stability and reduces warpage, but makes through-glass vias (TGVs) harder to manufacture.
For perspective, a 2 mm substrate with 50 µm vias would have an aspect ratio of 40:1.
Conventional sputtering struggles to deposit uniform conductive seed layers inside such deep, narrow holes. ALD's superior conformality could provide a solution.
However, ALD isn't the only competing technology.
In September, TRUMPF announced its HiPIMS technology for coating deep TGV structures, using highly ionized plasma to improve coverage. Meanwhile, MKS Atotech introduced a wet metallization process combining electroless copper seeding with advanced electroplating.
Progress is also being made in drilling. Korean equipment manufacturer Philoptics demonstrated TGV processing on 2 mm-thick, 510 × 515 mm glass panels, claiming zero defective holes in its demonstration.
@babyfolio Most traders add indicators to feel smart. The ones who actually make money remove them until only the moving average is left. Simplicity isn’t basic — it’s the only thing that survives a real drawdown
When the guy building the hungriest AI clusters on the planet publicly ranks memory suppliers as “reasonable” vs “extortionate,” he’s not venting. He’s handing out long-term loyalty scores in a market where HBM demand is set to keep outrunning supply through 2028. Micron just got the preferred-customer badge. The others just got a reminder that today’s customers have very long memories
@jukan05 Teflon for your GPUs.
Same stuff that keeps eggs off a pan, now carrying NVLink at densities that would cook regular cables. Non-stick signals. Absolute uniqe
@Kalshi They used their own models to reverse-engineer the device and identify the employee who planted it. Peak OpenAI — the AI snitched on the guy trying to spy on the AI
381% YTD, 88% win rate, 10x on $NBIS, and now front page of Liquid.
The AI buildout doesn’t sleep and apparently neither does your trading account.
If you actually pass the 500% portfolio I’m going to need the exact setup before the algorithm copies you.
Short, acknowledges the flex without pure glazing, adds a bit of tension around leverage + the “pass the 500%” line, and ends with an ask that invites engagement. Works well under a performance screenshot
@StormDirac This is the moment optical I/O stops being a slide and starts being real silicon in a live system.
CW-WDM isn’t dead — it just found its first real job. Whose lasers? 👀
@pequityresearch US companies are lighting tokens on fire with no ROI. China is shipping frontier models at a fraction of the cost and actually deploying them. The efficiency gap is the story
@MilkRoadAI Fed just confirmed AI demand is big enough to move U.S. inflation for years.
Token prices crashed. GPU prices didn’t.
Jevons Paradox is undefeated