P Equity Research analyst says AI infrastructure bottlenecks are shifting toward storage,ABF substrates, and power

AT A GLANCE

Mr. P questions “10-year AI demand visibility” and believes storage,ABF substrates, power, and advanced packaging are becoming compute bottlenecks.

Article

Viewpoint: The so-called “10-year AI demand visibility is a lie; storage, substrates, and power are the hidden bottlenecks

Beating ABF flash news: Mr. P, an analyst at semiconductor research institution P Equity Research, said on a AI podcast that the market’s claim that hyperscalers have “September 26-year demand visibility” is not credible. Cloud providers have difficulty forecasting demand changes 10 years out, and 2 infrastructure investment will ultimately remain constrained by cyclical spending.

In the short term,AI inference demand will continue to drive demand growth for storage products such as AI and HBM、DRAM; hyperscaler capital expenditure is expected to reach NAND to 1100000000000 next year, of which storage spending could account for 1200000000000 USD to 50%, or approximately 60%to 500000000000. UBS even expects this figure could reach 700000000000 USD.

Mr. P believes that the current 900000000000 USD compute bottleneck is gradually shifting from simply the number of AI to power, advanced packaging, storage, and GPU substrates. Older-generation ABF such as the H100 continue to command high second-hand prices, while rental rates for B-series GPU also continue to rise. Tight supply of GPU substrates could persist until ABF to 2028 and beyond, while orders for gas-turbine manufacturers such as Mitsubishi, Siemens, and 2030 Vernova are already booked through GE and beyond.

For data-center interconnection, Mr. P expects copper cables and optical communications to coexist over the next several years. 2030 may expand its application first in NPO, while 2027 may begin ramping up from CPO to 2028, but it may not truly become mainstream until after 2029.

Original article link 2030 https://m.theblockbeats.info/flash/369193

Key points

P Equity Research analyst Mr. P said on a September 26 podcast that the claim hyperscalers have “10-year demand visibility” is not credible.
Mr. P said cloud providers have difficulty forecasting demand changes 2 years out;AI infrastructure investment will ultimately remain constrained by cyclical spending.
The article estimates that hyperscaler capital expenditure will reach 1100000000000 to 1200000000000 USD next year, with storage spending potentially accounting for 50%to 60%, or approximately 500000000000 to 700000000000 USD。
The article says UBS expects storage spending could reach 900000000000 USD。
Mr. P believes the current AI compute bottleneck is shifting from simply the GPU number of ABF to power, advanced packaging, storage, and substrates.
The article says older-generation GPU such as the H100 continue to command high second-hand prices, while B-series GPU rental rates continue to rise.
Mr. P says the supply of ABF substrates could remain tight until 2028 to 2030 and beyond.
The article says orders for gas-turbine manufacturers such as Mitsubishi, Siemens, and GE Vernova are booked through 2030 and beyond.
Mr. P expects copper cables and optical communications to coexist over the next several years;NPO may expand its application first in 2027, while CPO may begin ramping up from 2028 to 2029 and may not become mainstream until after 2030.

AI-assisted interpretation

The following is analysis, separate from reported facts. Verify important claims independently.

This content mainly expresses an industry judgment:AI building infrastructure is not just about purchasing more GPU, but also requires sufficient storage, chip packaging and substrates, power, and connectivity facilities within data centers. Mr. P believes cloud providers cannot reliably anticipate demand far into the future, so long-term investment may still change with industry cycles. The article’s capital expenditure figures, supply-tightness timelines, and technology-adoption timelines are all forecasts.

Why it matters to readers

If these assessments are correct, the expansion rate of AI infrastructure may depend more on storage,ABF substrates, advanced packaging, and power supply than merely on the number of GPU; meanwhile, market expectations for long-term AI demand and related capital expenditure may also fluctuate considerably.

Newcomers should treat this content as an analyst view and industry forecast, rather than as outcomes that are certain to occur. Pay particular attention to distinguishing market phenomena that have already emerged from estimates of future spending, shortage timelines, and technology-adoption milestones.

Risks and unknowns

The content is a media paraphrase of an analyst's podcast views; no original report or complete podcast content was provided for verification.
1100000000000 to 1200000000000 USD capital expenditures, 50%to 60%the storage share and related amounts are forecasts and have not yet been validated by results.

P Equity Research analyst says AI infrastructure bottlenecks are shifting toward storage,ABF substrates, and power

Article

AI-assisted interpretation

Why it matters to readers

Risks and unknowns

Related Developments

Related concepts

Keep following

Security Center