跳到正文
Hacker News (official RSS)(respawnfirst.com)·· 9 天前精選品質評分77

報導稱 Jensen Huang 面臨晶片與記憶體供應過剩,NVIDIA 縮減 OpenAI 保證額度

Jensen Huang Admitted He Was Facing an "Avalanche" of Excess Wafers and Memory

重點導讀

AI 資料中心需求曾被視為推高 GPU 和記憶體價格的主因,但這篇報導說,雲端業者正轉向自研晶片,NVIDIA 也面臨已承諾購買的晶圓與記憶體供應過多。

  • 報導引述《華爾街日報》指出,財務長 Colette Kress 擔心對雲端業者和 AI 公司提供財務保證會增加風險,於是把 NVIDIA 對 OpenAI 算力的保證上限,從約 8 GW、最高 $250 billion,降至約 4.25 GW、最高 $125 billion。報導也稱 NVIDIA 已聯絡 TSMC 討論減少晶片訂單。
  • 文中其他庫存與需求數字則是分析人士的估計,並非已確認的公司資料。

AI 伴隨問答

自備 Key 深度分析

針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:

推薦理由

報導引述《華爾街日報》稱雲端業者轉向自研晶片,並詳述 NVIDIA 如何調降對 OpenAI 算力承諾的保證上限,呈現晶片供應與財務承諾的連動。

正文 · 原文

For the past two years, the narrative around inflated GPU and RAM prices has been the insatiable AI demand eating up wafer at TSMC and every module on Earth must go to AI data centers. However, the narrative is cracking ever since reports of channel stuffing and over commitments came out.

Now with Nvidia’s former CEO Jensen Huang gone, more details are out that corroborate our earlier reports on inflated demand on AI. According to Wall Street Journal, Huang admitted in August that major cloud providers are moving away from Nvidia in favor of in-house chips, leaving Nvidia staring down a massive oversupply of prepaid TSMC silicon and memory that was “coming at him like an ‘avalanche'”.

The WSJ report details a heated clash in August 2026 between Huang and Nvidia’s longtime Chief Financial Officer, Colette Kress, the executive who has now taken over as Nvidia’s interim CEO.

Kress was concerned about Nvidia’s growing financial exposure and wanted to rein in the company’s “circular financing” commitments to cloud providers, and AI companies like OpenAI and Anthropic.

While early talks had floated backstopping roughly 8 gigawatts of OpenAI compute (worth up to $250 billion), Kress successfully scaled Nvidia’s signed guarantee cap down to roughly 4.25 gigawatts, or up to $125 billion.

In an August meeting, Huang belittled Kress for questioning the deals, calling her “a deer in the headlights”. To justify why Nvidia needed to financially backstop OpenAI and smaller cloud providers to keep buying its GPUs, Huang admitted that Nvidia was caught in a massive squeeze.

Then he went on to describe the situation the company was in. Major cloud providers were running away from Nvidia, making plans to use alternative chips and lessen their reliance on the firm. Even worse, Huang said, the vast supply of chip wafers and memory Nvidia had committed to buying were coming at him like an ‘avalanche.’

Per the WSJ

Nvidia has reportedly contacted TSMC to wind down their chip orders.

For PC builders and hardware enthusiast the reported admission is validating, and quite glaring. Jensen Huang, instead of allowing excess TSMC wafer capacity and memory flow back into reasonable priced GeForce products and broader DRAM production, allegedly scrambled to use its own balance sheet to finance data center buildouts so AI startups and absorb Nvidia’s over-ordered chips.

The $500bn deal? Fake. The $100bn OpenAI deal? Fake. Jensen and NVIDIA have built their legacies on deception. They tried to rig 5060 interviews, and threatened @GamersNexus. This company sucks and has sucked for a while and everybody wants to pretend it's a sudden event. https://t.co/cpaIXwfp5d

— Ed Zitron (@edzitron) November 12, 2026

Reports of glut of unsold hardware have been around for a while, but only now mainstream media is catching up to what Nvidia was reportedly doing. Many industry analysts including Ed Zitron reported months ago how Nvidia and other companies have exposed themselves to devastating financial risk, and also made sure to channel stuff META, Google, Amazon, and others by leveraging their FOMO.

Zitron estimated that at least $200 billion worth of Nvidia AI chips are sitting idle in warehouses, purchased by cloud giants that lacked the electrical grid capacity or immediate end-user demand to plug them in.

The bottom line is, AI demand is no way near as high that it would consume this much silicon.

Sikandar Mahmood

Sikandar Mahmood

I’m Sikandar Mahmood, a gaming journalist and content strategist with over a decade of experience covering video game news and crafting in-depth guides. I began my career at SegmentNext as a news reporter and editor from 2014 to 2019, before founding RespawnFirst in 2019. I currently serve as Head of Content Production on RF, where I manage writers, shape editorial direction, and focus on accurate, high-quality gaming news and guides.

來源: Hacker News (official RSS) (respawnfirst.com)
開啟原文

讀者討論

理性交流 · 即時討論