Name: 100,000 GPUs Swept Up in Humanoid Contract… Don't Be Afraid Now, Samsung Electronics and SK Hynix / The Real Reason 90% of Bad News is 'Noise' | CEO Noh Geun-chang
Uploaded: 2026-09-07T07:00:12+00:00
Duration: 1731 s

Speakers

Noh Geun-chang — Center Head, Hyundai Motor Securities Research Center

Summary

CEO Noh Geun-chang argues that the new GPT-6 and Claude model cycle is another ChatGPT-like inflection, supporting AI and memory semiconductor demand. He highlights the Nscale-Figure AI contract for up to 100,000 Nvidia Vera Rubin platforms, expects SoftBank to rally as OpenAI worries fade, and sees Samsung Electronics as undervalued with shareholder-return support. He also says tariff-driven drops in Korean memory names are buying opportunities and Oracle earnings are a key AI-infrastructure check.

New AI model launches are viewed as an AGI-like upgrade and positive for memory demand.
Nscale's humanoid contract for up to 100,000 Nvidia Vera Rubin platforms is a major GPU demand signal.
SoftBank is framed as an OpenAI proxy through its 11% stake and Stargate US financing.
Anthropic and OpenAI IPOs could fund several more years of AI data-center investment.
Nvidia's data-center mix shows non-hyperscale demand doubling; chip competition benefits memory and TSMC.
Samsung Electronics is flagged as cheap with dividend and buyback upside.
US tariff rules include exemptions; Korean semiconductor tariff fears are seen as a dip-buying opportunity.
Oracle earnings are a potential weak-link test for AI infrastructure.

Ideas

Noh Geun-chang Center Head, Hyundai Motor Securities Research Center 5:00

New AI models drive sharp memory demand

OpenAI's GPT-6 and Anthropic's Claude Opus 5.1 / Astra upgrade cycle feels like the 2023 ChatGPT moment and shows models approaching AGI; AI diffusion means more AI fixed-cost usage, and because memory is the store-and-transmit layer, memory demand should expand sharply.

Noh Geun-chang Center Head, Hyundai Motor Securities Research Center 8:00

Nscale's 100,000 Vera Rubin deal boosts Nvidia

Nscale, a UK neocloud backed by Nvidia, announced a strategic relationship with Figure AI and a contract to use up to 100,000 Nvidia Vera Rubin platforms for humanoid robot training; although humanoid robots are not yet mainstream, this is a large AI GPU demand signal because each rack is extremely expensive and 60-70% above Blackwell, and it illustrates Nvidia directing neocloud customers toward its chips via equity stakes.

Noh Geun-chang Center Head, Hyundai Motor Securities Research Center 10:48

Oracle earnings test AI infrastructure weak link

Oracle is the weak link in the AI infrastructure trade; its stock has already fallen sharply and its earnings on the 10th are very important because financing costs and credit spreads have risen for AI infrastructure names, so Oracle's results are a key sentiment check on the AI capex trade.

Noh Geun-chang Center Head, Hyundai Motor Securities Research Center 14:54

SoftBank rallies as OpenAI worries fade

SoftBank Group holds an 11% OpenAI stake and is the key financing partner for Stargate US; the stock has pulled back from its June peak, but GPT-6 removes much of the OpenAI worry, so SoftBank should rally as an OpenAI proxy with near-one correlation and as the AI data-center funding vehicle.

Noh Geun-chang Center Head, Hyundai Motor Securities Research Center 16:53

AI IPOs extend data-center investment cycle

A September 3 White House document imposes tariffs on imported Nvidia and AMD chips but has exemptions for US supply-chain competitiveness and waives tariffs once a company announces US factory construction; Korean memory names may look disadvantaged without announced US memory fabs, but if they announce fabs they become tariff-free, so tariff-driven drops in Samsung Electronics and SK Hynix should be viewed as opportunities.

Noh Geun-chang Center Head, Hyundai Motor Securities Research Center 20:11

Chip arms race benefits memory and TSMC

Nvidia's newly disclosed data-center mix shows hyperscale revenue of about $48.7B and 'other' AI cloud/industrial/enterprise/neocloud revenue of about $40B, which doubled sequentially and uses only Nvidia solutions; hyperscalers are becoming custom-chip competitors, but the resulting two-way AI chip competition is structurally very good for memory semiconductors and TSMC no matter which camp wins.

100,000 GPUs Swept Up in Humanoid Contract… Don't Be Afraid Now, Samsung Electronics and SK Hynix / The Real Reason 90% of Bad News is 'Noise' | CEO Noh Geun-chang

Summary

Ideas

Up Next