What are the key takeaways from this Odd Lots episode?

Key insights include: **HBM Production Trade-offs:** High Bandwidth Memory requires three times more wafer capacity than commodity DRAM to produce the same number of bits. When manufacturers shift production to HBM for AI accelerators, they crowd out supply for PCs, smartphones, and gaming consoles. This ratio worsens with HBM4 and HBM5 generations, intensifying the shortage for consumer electronics despite HBM representing over half the DRAM market.; **Memory Demand Across AI Pipeline:** AI applications consume massive memory at every stage—training requires HBM and CPU DRAM, inference needs HBM for memory-bound decode operations, and agentic AI demands extensive CPU-based servers with commodity DRAM. Long context windows amplify this demand as models process increasingly complex prompts and generate lengthy outputs, with ChatGPT usage growing from novelty queries to 800 million users requesting detailed multi-page reports.; **Supply Constraints Through 2026:** Memory manufacturers face clean room space limitations preventing rapid capacity expansion. The primary solution involves node migration to 1b and 1c processes using EUV machines, producing more bits per wafer. However, this migration takes time and competes with HBM production needs. Major producers increased capital expenditure significantly in 2025, but meaningful new fab capacity won't come online until 2028.

What did Ray Wang discuss on Odd Lots?

Ray Wang from Semia Analysis explains how AI demand is creating a severe DRAM shortage, causing spot prices to surge dramatically since late 2024. Memory chip makers face clean room constraints while balancing production between high-margin HBM chips for AI accelerators and commodity DRAM for consumer electronics, creating supply shortages expected to persist through 2027. Key topics include: **HBM Production Trade-offs:** High Bandwidth Memory requires three times more wafer capacity than commodity DRAM to produce the same number of bits. When manufacturers shift production to HBM for AI accelerators, they crowd out supply for PCs, smartphones, and gaming consoles. This ratio worsens with HBM4 and HBM5 generations, intensifying the shortage for consumer electronics despite HBM representing over half the DRAM market.; **Memory Demand Across AI Pipeline:** AI applications consume massive memory at every stage—training requires HBM and CPU DRAM, inference needs HBM for memory-bound decode operations, and agentic AI demands extensive CPU-based servers with commodity DRAM. Long context windows amplify this demand as models process increasingly complex prompts and generate lengthy outputs, with ChatGPT usage growing from novelty queries to 800 million users requesting detailed multi-page reports..

How long is this episode of Odd Lots?

This episode is 45 minutes long. SignalCast provides an AI-generated summary so you can get the key insights in about 3 minutes.

Odd Lots

Ray Wang on How AI Is Causing DRAM Prices to Surge

February 16, 2026

45 min episode · 2 min read

Ray Wang

Episode

45 min

Read time

2 min

Topics

Startups, Fundraising & VC, Artificial Intelligence

AI-Generated Summary

Published Feb 16, 2026

Key Takeaways

✓HBM Production Trade-offs: High Bandwidth Memory requires three times more wafer capacity than commodity DRAM to produce the same number of bits. When manufacturers shift production to HBM for AI accelerators, they crowd out supply for PCs, smartphones, and gaming consoles. This ratio worsens with HBM4 and HBM5 generations, intensifying the shortage for consumer electronics despite HBM representing over half the DRAM market.
✓Memory Demand Across AI Pipeline: AI applications consume massive memory at every stage—training requires HBM and CPU DRAM, inference needs HBM for memory-bound decode operations, and agentic AI demands extensive CPU-based servers with commodity DRAM. Long context windows amplify this demand as models process increasingly complex prompts and generate lengthy outputs, with ChatGPT usage growing from novelty queries to 800 million users requesting detailed multi-page reports.
✓Supply Constraints Through 2026: Memory manufacturers face clean room space limitations preventing rapid capacity expansion. The primary solution involves node migration to 1b and 1c processes using EUV machines, producing more bits per wafer. However, this migration takes time and competes with HBM production needs. Major producers increased capital expenditure significantly in 2025, but meaningful new fab capacity won't come online until 2028.
✓Demand Destruction Patterns: PC makers including Dell, Lenovo, and Asus implemented price hikes across product lines. MediaTek cut 2026 mobile chip outlook by ten to fifteen percent, representing 10 million fewer units. Companies face choices between downgrading product specifications, delaying launches of higher-margin new products, or accepting margin compression. Apple reports managing impacts better than competitors but expects meaningful effects in second half 2026.
✓Pricing Dynamics Reversal: Commodity DRAM spot prices remained stable during early AI hype from 2023-2024, then surged dramatically starting late 2024. Current commodity DRAM margins actually exceed HBM margins despite HBM's reputation for higher profitability. Memory suppliers still prioritize HBM production for strategic positioning in the growing AI market, viewing it as a critical growth driver for the next several years despite lower current margins.

What It Covers

Ray Wang from Semia Analysis explains how AI demand is creating a severe DRAM shortage, causing spot prices to surge dramatically since late 2024. Memory chip makers face clean room constraints while balancing production between high-margin HBM chips for AI accelerators and commodity DRAM for consumer electronics, creating supply shortages expected to persist through 2027.

Key Questions Answered

•HBM Production Trade-offs: High Bandwidth Memory requires three times more wafer capacity than commodity DRAM to produce the same number of bits. When manufacturers shift production to HBM for AI accelerators, they crowd out supply for PCs, smartphones, and gaming consoles. This ratio worsens with HBM4 and HBM5 generations, intensifying the shortage for consumer electronics despite HBM representing over half the DRAM market.
•Memory Demand Across AI Pipeline: AI applications consume massive memory at every stage—training requires HBM and CPU DRAM, inference needs HBM for memory-bound decode operations, and agentic AI demands extensive CPU-based servers with commodity DRAM. Long context windows amplify this demand as models process increasingly complex prompts and generate lengthy outputs, with ChatGPT usage growing from novelty queries to 800 million users requesting detailed multi-page reports.
•Supply Constraints Through 2026: Memory manufacturers face clean room space limitations preventing rapid capacity expansion. The primary solution involves node migration to 1b and 1c processes using EUV machines, producing more bits per wafer. However, this migration takes time and competes with HBM production needs. Major producers increased capital expenditure significantly in 2025, but meaningful new fab capacity won't come online until 2028.
•Demand Destruction Patterns: PC makers including Dell, Lenovo, and Asus implemented price hikes across product lines. MediaTek cut 2026 mobile chip outlook by ten to fifteen percent, representing 10 million fewer units. Companies face choices between downgrading product specifications, delaying launches of higher-margin new products, or accepting margin compression. Apple reports managing impacts better than competitors but expects meaningful effects in second half 2026.
•Pricing Dynamics Reversal: Commodity DRAM spot prices remained stable during early AI hype from 2023-2024, then surged dramatically starting late 2024. Current commodity DRAM margins actually exceed HBM margins despite HBM's reputation for higher profitability. Memory suppliers still prioritize HBM production for strategic positioning in the growing AI market, viewing it as a critical growth driver for the next several years despite lower current margins.

Notable Moment

The analyst revealed that commodity DRAM margins currently exceed HBM margins despite conventional wisdom, creating a strategic dilemma for manufacturers. Memory suppliers continue prioritizing HBM production anyway because it represents a new growth driver in a market historically dependent on flat-growth sectors like PCs and mobile phones, making technological leadership in HBM critical for long-term competitive positioning.

Know someone who'd find this useful?

Episode Transcript

UKG. Their HR, pay, and workforce management tools help business leaders empower their people. Because when work works, everything works. Learn more at ukg.com/work. Support for the show comes from Public. Public is an investing platform that offers access to stocks, options, bonds, and crypto, and they've also integrated AI with tools that can assist investors in building customized portfolios. One of these tools is called generated assets. It allows you to turn your ideas into investable indexes. So let's say you're interested in something specific like biotech companies with high r and d spend, small cap stocks with improving operating margins, or the S and P 500 minus high debt companies. Chances are there isn't an ETF that fits your exact criteria. But on public, you just type in a prompt and their AI screens thousands of stocks and build a one of a kind index. You can even back test it against the S and P 500, then you can invest in a few clicks. Go to public.com/market and earn an uncapped 1% bonus when you transfer your portfolio. That's public.com/market. And paid for by Public Holdings, brokered services by Public Investing member FINRA SIPC, advisory services by Public Advisors, SEC registered adviser, crypto services by Xero Hash. Sample prompts are for illustrative purposes only, not investment advice. All investing involves risk of loss. See complete disclosures at public.com/disclosures. Small businesses are the pulse of every community. They bring people together, create opportunities, and drive growth. Chase for Business helps business owners like you with personalized guidance and convenient digital tools all in one place. With that guidance and your determination, you can take your business farther and help build a brighter future for your community. Learn more at chase.com/business. Chase for business. Make more of what's yours. The Chase mobile app is available for select mobile devices. Message and data rates may apply. JPMorgan Chase Bank, NA. Member, FDIC. Copyright 2026. JPMorgan Chase and Company. Bloomberg Audio Studios. Podcasts, radio, news. Hello, and welcome to another episode of the Odd Lots podcast. I'm Joe Weisenthal. And I'm Tracy Alloway. Tracy, you know what I'm worried about? I don't know if I'm worried about it, but it it kind of feels like the direction things are going, which is that, like, all energy, industrial commodities, anything that we use for any purpose is just going to, like, feed the AI beast. And us humans, we're just gonna get left out in the cold. They'll be like, nothing for you. We gotta, like, feed we gotta feed it all to the AI. And and and maybe in ten or twenty or fifty years, the AI is so powerful to just decide why why do humans get any of this? We just, we should keep it all to ourselves. A legitimate concern, I would say. I mean, to some extent, we are already seeing this crowding out effect. Yeah. Right? So energy prices in certain areas have been going up, …

Get the full transcript (10,310 words) + summary by email — free

One-time email with the complete transcript and AI summary of this episode. No account needed.

One email, no spam. We’ll also show you what SignalCast does.

Browse all Odd Lots transcripts →