April 18, 2025

Round‑up

Highlights

1. Intel’s new CEO Lip‑Bu Tan flattened management layers in his first 24 hours in the job¹
2. Jensen Huang’s Beijing charm offensive. Nvidia’s CEO met Vice‑Premier He Lifeng and telecom start‑ups, calling China a “key market” even as a new licence is required for AI‑chip exports.²
3. Micron spins out a pure‑play HBM business. Boise’s memory maker will report high‑bandwidth‑memory sales as a stand‑alone line from 30 May, underscoring how AI has turned memory into margin gold.³

Other developments:

• South Korea drafted an ₩12.2 trn support budget for chipmakers hit by tariffs⁴
• Samsung hit a 40 %+ test yield on 4 nm logic dies for 12‑high HBM4 stacks⁵
• TSMC is mulling a 30 % price premium for 4 nm wafers at its Arizona fab⁶.

Did You Know? Micron’s Cloud Memory Business Unit will be the first in the industry to disclose HBM revenue separately, starting with its fiscal Q4 close on 30 May 2025.³

In‑depth

1  Government & Corporate Policy

• Seoul fires a fiscal shield

  • A ₩12.2 trn ($8.6 B) extra budget earmarks ₩2.1 trn for tariff‑risk relief and ₩1.8 trn for AI programmes.⁴
  • Bank of Korea hinted at a May rate cut to soften the blow.

• Nvidia’s quiet Beijing diplomacy

  • Huang reassured officials that Nvidia will tailor compliant chips for China after the H20 export clamp‑down, calling the mainland a “key market”.²
  • Meetings included start‑up DeepSeek and electric‑vehicle OEMs.

• China brands tariffs ‘economic bullying’

  • Foreign Minister Wang Yi told ASEAN envoys that unilateral duties “threaten the rules‑based order,” previewing a UN Security Council session on 23 April.⁷

2  Economics, Finance & Business Outlook

• Intel goes flat and fast

  • New CEO Lip‑Bu Tan scrapped two management layers; data‑centre, PC and AI groups now report directly to him.¹
  • Networking boss Sachin Katti becomes CTO & AI chief, tasked with unifying Gaudi, Habana and GPU road‑maps.

• Micron’s four‑square overhaul

  • Cloud, Core‑DC, Mobile/Client and Auto/Embedded units replace the old product matrix, each with its own P&L.⁸
  • The cloud unit is expressly chartered to expand HBM capacity for Nvidia and AMD GPUs.

• TSMC weighs an Arizona price premium

  • Market sources say the Phoenix fab could charge ~30 % more than Taiwan for 4 nm wafer starts, citing higher labour and utility costs.⁶
  • Clients including AMD EPYC 5 and Nvidia “Blackwell” are said to be on board.

3  Technology & R&D

• Dedicated HBM BU at Micron

  • Financial reporting for HBM will break out separately from 30 May, signalling a long‑term bet on AI‑centric memory.³
  • Custom “Cloud‑HBM” SKUs aimed at hyperscalers are in development.

• Samsung clears 40 % test yield for 4 nm logic die

  • The milestone unlocks 12‑layer HBM4 stacks; next hurdle is stabilising 1c‑DRAM and TC‑NCF packaging.⁵
  • Yield is considered “mass‑production‑ready” by internal KPIs.

• Intel’s Gaudi AI chips caught in fresh export net

  • New U.S. controls mean Gaudi accelerators exceeding 1,400 GB/s DRAM bandwidth now need licences to ship to China.⁹
  • Intel has warned customers to expect lead‑time extensions.

• TSMC price signals could ripple into design wins

  • A U.S. premium on 4 nm may accelerate the rush to 3 nm and chiplet partitioning for cost control.⁶

Footnotes