EXPLORINEER
en

DE — heise online

OpenAI Chief Sam Altman Calls Private DRAM Ownership "Terrible"

Chips1 min read

OpenAI boss Sam Altman says it is "terrible" that memory chips (DRAM) sit unused inside people's own computers. He also says the chip shortage that is pushing up prices is not the fault of chipmakers.

What happened

In a podcast called "Cheeky Pint", Sam Altman talked about the high price of memory chips. He said that local AI models, which run on a user's own computer instead of in the cloud, make the shortage worse. He also said the market's "invisible hand" is failing right now, and he defended OpenAI's plan to build more data centers.

Key numbers and facts

  • Memory chip (DRAM) prices have multiplied for more than a year.
  • Altman said DRAM in a home computer is used only about 5 percent of the time.
  • Altman wanted about 7 trillion US dollars to build 36 new data centers in 2023. The plan did not work out.
  • The PC market fell sharply in the third quarter of 2026.

Why memory prices are so high

AI data centers need huge amounts of memory. Chipmakers have moved their factories toward special memory for these centers, such as HBM and LPDDR. This leaves less memory for regular PCs and phones, so prices for those devices go up too. Altman also named generators, batteries, cement and steel as materials whose prices are rising.

Why it matters

The memory shortage makes computers, phones and graphics cards more expensive for everyday buyers. Altman's comments show that AI companies are now part of a fight over scarce parts. His idea of robots that build more data centers shows how far AI companies want to grow.

What to watch

  • Whether memory chip prices keep rising, as the industry expects for at least next year.
  • Whether chipmakers build new memory factories.
  • Whether OpenAI's robots that build data centers become real.

Glossary

OpenAI
A US company that makes AI tools such as ChatGPT.
DRAM
Working memory in a computer. It holds data the computer is using right now.
HBM
A fast, special type of memory used in AI data centers.
LPDDR
A low-power memory type used in phones and data centers.

Same story, different eyes

Ad

Add 150 words. Learn 10 a day.

Dicter word books send only the words you learn to your review notes.

ChipsZDNet Korea

Vitrontech Named a Materials, Parts and Equipment Excellence Company, to Develop ₩12 Billion Semiconductor Equipment

Vitrontech, known for display heat-treatment equipment, has been selected as a "Materials, Parts and Equipment Excellence Company" by the Ministry of Trade, Industry and Energy. The company will develop a rapid thermal processing (RTP) system worth over ₩12 billion with domestic semiconductor firms, aimed at next-generation 3D semiconductor processes such as 3D DRAM, using a laser-based temperature control approach and an all-in-one platform integrating deposition and etching chambers.

ChipsZDNet Korea

Lee Jae-yong Reviews AI Chip Battleground in Japan, Strengthens Advanced Packaging Cooperation

Samsung Electronics Chairman Lee Jae-yong visited the Advanced Package Lab (APL) in Yokohama, Japan, to review material and process technology development for AI chip competitiveness and cooperation with local companies. The visit, Samsung said, was the first on-site inspection since the lab opened on September 8, about two months earlier. Samsung plans to expand APL's role into next-generation processes such as wafer bonding and glass substrates, and to grow it into a global R&D hub for packaging technology.

Chipsheise online

Funding for 300mm Wafer Production in Itzehoe Handed Over to Vishay

Semiconductor maker Vishay plans to invest a "high three-digit million" amount in its Itzehoe site in Schleswig-Holstein, reportedly around one billion euros. Schleswig-Holstein's economy minister Claus Ruhe Madsen handed over a funding notice of about 215 million euros, and the site will add 300mm wafer production for power MOSFETs, with up to 250 new jobs expected. The EU Commission has already approved the aid.

ChipsThe Register

Backbone of a beast – charting Nvidia's rise to networking titan

Nvidia has become a major networking vendor with larger datacenter Ethernet sales than any other company, reporting over $17.5 billion in networking revenue in Q2, compared with Cisco's $17.3 billion across all products in the same period. The article traces this shift to the 2019 $6.9 billion acquisition of Mellanox, which gave Nvidia InfiniBand and ConnectX technology and helped it position itself as a system builder. Analysts and customers credit its tightly integrated compute and networking architecture for simplifying deployments and optimization.