404K Tech Evening News: Storage Super Cycle, Cloud Computing Power Price Increase, AI Hardware Bottleneck Spreads

404K Technology Evening Post 2026-06-26 — Supercycle in Storage, Cloud Computing Power Price Increases, Diffusion of AI Hardware Bottlenecks

Storage prices, cloud computing power prices, and AI hardware supply have simultaneously become the main line of today’s technology chain: Micron’s performance pushes DRAM/NAND from cyclical assets back to bottleneck assets, AWS’s GPU instance price increase shows that computing power supply and demand remains tight, and Apple’s price hike indicates that cost pressures are spreading to the end-user.

Pre-market highlights

The core of the technology chain before the market tonight is not a single AI application, but the migration of profit pools in AI infrastructure: Micron’s financial report, SK Hynix’s target price revision, Samsung’s potential major investment, and Applied Materials’ target price hike all point to the repricing of storage, equipment, advanced packaging, and upstream materials.

Downstream pressure is also becoming apparent.Apple has raised the price of hardware other than iPhone, AWS is said to have increased GPU instance prices by 20%, while the Microsoft Xbox and PC market chains are also being pushed up by memory costs; this means AI demand remains strong, but cost pressures are moving beyond the capital expenditure of cloud providers and gradually entering end-device pricing and software valuations.

Investment banking activity is focused on Micron, SK Hynix, Applied Materials, onsemi, Qualcomm, and the MLCC supply chain.Corporate-level facts mainly fall among AI/Semiconductors, cloud, software, internet platforms, consumer electronics, and smart devices.

Full Industry Chain of AI/Semiconductors

AI Models/Applications and Capital Expenditure

• OpenAI
  
  1) According to reports, OpenAI tends to delay its IPO to 2027, and the internal valuation target still revolves around $1 trillion; this will impact the valuation window for unlisted AI labs, employee incentives, and the pace of computing power financing.
  2) According to the TSMC supply chain, the OpenAI Jalapeno inference chip will use 3nm, and Jalapeno2 may use A16 with CoWoS. AI model companies continue to extend towards self-developed ASICs and advanced packaging.

"OpenAI intends to postpone its IPO to 2027"

• Cerebras
  
  1) UBS has raised the target price of Cerebras from $300 to $320, maintaining a buy. The company reported quarterly revenue of $191 million, beating UBS and market expectations of $181 million, and earnings per share (EPS) of -$0.04, better than expected.
  2) Core cloud and services revenue was $80 million, hardware revenue was $111.6 million; UBS says the formal Amazon agreement, OpenAI scaling up, and new client expansion are all being validated at the same time, but the true bottleneck has shifted to data center capacity.

"current industry tightness in HBM, CoWoS, and TSMC N3 are not constraints"

• Zhiyu
  
  1) Zhiyu GLM-5.2 report raised its target price from HKD 920 to HKD 1,900, maintaining a hold rating; revenue forecasts are CNY 4.328 billion/12.76 billion/26.612 billion for 2026/2027/2028, up 497%/195%/109% year-on-year.
  2) The report focuses not on short-term profitability, but on improvements in long-term coding, inference efficiency, and enterprise deployment capabilities, indicating that competition among model companies is shifting from narrative around parameters to task completion rates and computing cost.

• AI Deployment Bottleneck
  
  The Nomura Technology Monthly puts FDE capability at the center of AI implementation, believing that model progress is faster than enterprise system deployment—the bottleneck has shifted from “is there a model” to “can it be integrated into business”. This explains why CPUs, analog semiconductors, MLCCs, and power supplies have also entered the AI trades.