Newsletter · · Ashutosh Agarwal

Memory Chip Shortage Drives Micron and SK Hynix Past 1 Trillion Dollars - Semis & AI Infrastructure Podcast Recap - Week of May 25-31, 2026

The Semis & AI Infrastructure podcast recap for May 25-31, 2026. Memory is the story of the week as Micron and SK Hynix cross $1T and Dell's blowout becomes the proxy for AI-infrastructure breadth, while the debate shifts from whether demand is real to whether the tape is paying too much for it.

Semis & AI Infrastructure Podcast Recap

Week of May 25-31, 2026: Memory Chip Shortage Drives Micron and SK Hynix Past 1 Trillion Dollars

🧠 Top of Mind This Week

Memory is the story of the week. Micron and SK Hynix both crossed $1T market cap, UBS triple-banked MU's price target from $535 to $1,625, and every operator and strategist on the tape called HBM/DRAM/NAND the binding constraint of the AI build-out for "at least two years." Dell's blowout (AI server guide raised from $50B to $60B, FY revenue guide to $167B, AI-optimized server revenue +757% YoY) became the proxy for how broad and how supply-constrained AI infrastructure has become. The bull-bear debate has shifted from "is the demand real" (consensus says yes) to "are we paying too much for it" (SOX 69% above its 200DMA, unprecedented in 25 years, vs. earnings growth that has actually kept pace).

🔑 Dominant Themes

  • Memory supercycle, with a 1999 echo. MU and SK Hynix entered the trillion-dollar club; SanDisk is up roughly 600% YTD; UBS price target on MU went to $1,625. Bulls (Chisholm, Cramer, Mui) point to long-term agreements, sold-out capacity through 2027, and HBM as the AI choke point. Bears (Boockvar) note MU at about 9x P/S vs. its 1999 peak of about 8x, and the pull-forward signal from CDW and others suggesting double or triple ordering.
  • Dell as the canary for AI server demand breadth. AI server revenue $16.1B (+757% YoY), backlog growing, demand spread across neoclouds, sovereigns, and enterprise. CFO David Kennedy on Bloomberg Tech: "as we move from training models into inferencing, those inferencing workloads are creating a net new environment, a net new TAM." Truist's Nicknum is the dissenter, citing pull-forward and a valuation now at an S&P multiple vs. a historical 9x.
  • Memory shortages cascading into broader hardware inflation. Dell is warning of lead times up to a year on memory, CPUs, and HDDs. BI's Woojin Ho: traditional server costs up 200-300% YoY; PCs up about 20% YoY in H2 2026. Beneficiaries include SMCI, HPE, ARM, and the WFE complex (AMAT, LRCX, KLAC all on allocation).
  • Anthropic and the private-credit-funded AI capex stack. Anthropic ARR went from $9B (Jan) to $47-50B (May), raising $65B at roughly $965B valuation, with Apollo and Blackstone leading what could be the largest private credit deal ever (around $36B) to finance chip and server purchases for neoclouds without balance sheets. SpaceX and xAI are projected to take in around $15B per year renting compute to Anthropic.
  • NVIDIA: dominant fundamentals, stock disconnect, China zero. Vera Rubin is driving the Dell pipeline; Goldman calls semis the only profitable layer of the AI stack today. But Cramer: "just not performing... mind-numbing." Boockvar: "essentially zero market share in China," permanently. Multiple compression to 10-11x on 2027 estimates per Ben Rogers cited on CNBC.
  • Huawei's "Tao Scaling Law," a non-EUV path to 1.4nm-equivalent. He Tingbo proposes "time scaling" vs. geometric scaling, targeting 1.4nm-equivalent performance by around 2031 (vs. TSMC's commercial 1.4nm in 2028). Read-through: a long-term threat to ASML's EUV monopoly and to NVIDIA's China-locked-out position. Ian (Telecoms.com): "leans toward taking it seriously"; one industry contact: "usual Huawei propaganda."

⚖️ Key Debates

Are semis in a bubble?

Michael Batnick (Compound and Friends): SOX 69% above 200DMA, unlike anything in 25 years; MU's $100B to $1T move evokes Cisco and Lucent in 2000. Denise Chisholm: aggregate CapEx/FCF still under 1x vs. 3.5-4x at the 2000 peak; price has actually lagged earnings growth over the past two years; she puts a 70% probability on semi outperformance over the next 12 months from a bottom-quartile valuation.

MU valuation: cyclical peak or structural re-rating?

Boockvar (bear): MU at about 9x P/S matches the 1999 cyclical peak; double and triple ordering inflates near-term signals; extra supply does not arrive until 2027-2028. Jake Silverman (BI, bull): LTAs and the AI tailwind make this 3-5 years of structural earnings, not a typical cycle.

Semis vs. software as a structural relative trade

Adam Parker (Excess Returns): "My North Star is still semis over software." CTO pricing pressure plus the need to invest in AI compresses software margins; he sees software revenue misses in 5-6 quarters. The counterview is that software is defensive given multi-year contracts (Parker rejects this).

Semi-cap equipment vs. memory

Gavin Baker (referenced by Parker): semi-cap is very expensive vs. DRAM, and they "can't both be right." Parker concedes directionally (MU around 7x vs. CAT around 30x is incongruous) but notes momentum keeps semis acting well.

AI productivity and job displacement pace

Carson Block (Muddy Waters): 15% knowledge-worker displacement in 3-5 years. Matthew Weir (Goldman): a 10-year gradual process; 25% of US tasks (not jobs) automatable; trend GDP rises to 2.4%.

NVIDIA stock vs. fundamentals

Cramer (bull on the business, bear on the action): the "biggest fundamental winner" but underperforming for two years; a "pitiful, helpless giant" on the tape. Faber and Quintanilla note 10-11x on 2027 numbers as the bull's defense.

Tao Scaling Law: innovation or propaganda?

Ian (Telecoms.com) takes it seriously and flags the geopolitical bargaining-chip motive. An anonymous industry contact dismisses it as Huawei propaganda. The implication for ASML's roughly €9.5B China DUV revenue is real either way.

AI capex sustainability

Bears (channeled by Brown): half the AI capex backlog comes from equity-funded OpenAI and Anthropic, a dot-com pattern. Chisholm: hyperscalers are FCF-positive above capex, so it is not systemic. Bradley Tusk: OpenAI at $13B revenue with $9B losses and a $1.4T debt stack is "irrational for retail."

📊 Stocks Discussed With Bull/Bear Angle

Ticker Direction Source / Speaker Argument
DELL BULL Woojin Ho, Mandeep Singh (Bloomberg Intelligence) AI server guide raised from $50B to $60B; neoclouds and sovereigns driving multi-billion deals; gross margins expanding
DELL BULL Faber, Cramer (Squawk on the Street) AI-optimized server revenue $16.1B, +757% YoY; revenue beat by about $10B; meaningful backlog
DELL BULL David Kennedy, Dell CFO (Bloomberg Tech) $167B FY guide (+$27B); 88% revenue growth, 214% EPS growth; broad demand across NeoClouds, Sovereign, Enterprise
DELL BEAR Matthew Nicknum, Truist Hold (Closing Bell) Valuation now at S&P multiple vs. historical 9x P/E; pull-forward; PC growth decelerating to low single digits
MU BULL Jake Silverman (Bloomberg Intelligence) LTAs reduce cyclicality; AI structural tailwind; multiple expansion thesis from about 10x to 15x+
MU BULL Cramer (Squawk on the Street) Susquehanna $175 PT; HBM "choking point"; "Micron's going to earn $100 per share"; +78% in May
MU BULL Denise Chisholm (The Compound and Friends) UBS PT $1,625 (from $535); sold-out memory about 1 year; #2 S&P 500 YTD contributor; went from #127 to #11 mcap in about a year
MU BULL Adam Parker (Excess Returns) About 7x next-year earnings vs. CAT around 30x; many institutions underweight; momentum; crossed $1T +16% on recording day
MU BEAR Peter Boockvar (RiskReversal Pod) About 9x P/S matches 1999 cyclical peak (about 8x); double and triple ordering inflating signals; extra supply not until 2027-2028
NVDA BULL Janet Mui, Matthew Weir (Bloomberg Tech) Forward PE in the 20s, not a bubble; Goldman: semis the only profitable layer of the AI stack today
NVDA BULL Adam Parker (Excess Returns) CUDA moat; market-weight-plus; hyperscaler spend not waning
NVDA BEAR Cramer (Squawk on the Street) "Not performing" despite being core enabler; "pitiful, helpless giant" on share price
NVDA BEAR Jake Silverman (Bloomberg Intelligence) Competition from AVGO (TPU), MRVL (Trainium), AMD, and hyperscaler custom ASICs
NVDA BEAR Peter Boockvar (RiskReversal Pod) "Essentially zero market share in China"; permanently lost the world's second-largest economy
AVGO BULL Jake Silverman (Bloomberg Intelligence) Best-positioned NVIDIA competitor via the Google TPU relationship
MRVL BULL Cramer (Squawk on the Street) "Most explosive model of any semis"; guide raised $4B to $10B; 45% to 50% to about 55% growth trajectory
AMD BULL Jake Silverman (Bloomberg Intelligence) Expected to gain AI accelerator share
TSM BULL Jake Silverman (Bloomberg Intelligence) Capacity constrained; long-term agreements signed
TSM BULL Janet Mui (Bloomberg Tech) Forward PE in 20s; Taiwan +40% export growth driven by semis
TSM BEAR Ian (Telecoms.com Podcast) Node cadence slowing (3nm to 2nm took 3 years vs. prior 2-year pace); Huawei could close the gap by 2031
AMAT BULL Cramer (Squawk on the Street) On allocation; "shortage starts with AMAT, Lam Research, and KLA"
AMAT BULL Nicholas Rossolillo (Chip Stock Investor Podcast) Aggressive M&A (ASMPT panel-level deposition); dominant in deposition
LRCX BULL Nicholas Rossolillo (Chip Stock Investor Podcast) 2026-2027 expected record revenue years; benefits from AI build-out
KLAC BULL Nicholas Rossolillo (Chip Stock Investor Podcast) Largest metrology; recurring service and software revenue; durable segment
ASML BULL Nicholas Rossolillo (Chip Stock Investor Podcast) Controls the EUV chokepoint; about 70% Fab Five share
ASML BEAR Ian (Telecoms.com Podcast) Time-scaling reduces long-term EUV dependence; tighter DUV restrictions to China possible (about €9.5B China revenue at risk)
ONTO BULL Nicholas Rossolillo (Chip Stock Investor Podcast) Creative M&A (27% of Rigaku); durable metrology segment
ARM BULL Faber (Squawk on the Street) +218% YTD
QCOM BULL Cramer (Squawk on the Street) PC refresh beneficiary
HPE BULL Woojin Ho (Bloomberg Intelligence) Corporate AI adoption to reinvigorate traditional server demand with higher ASPs
HPE BULL Matthew Nicknum, Truist Buy (Closing Bell) Discount to Dell; Juniper networking stickiness
ANET BULL Matthew Nicknum (Closing Bell) Higher GM, more durable than hardware OEMs
CSCO BULL Matthew Nicknum (Closing Bell) Networking has more durable margins than server OEMs
SMCI BULL CNBC desk (Closing Bell) Up about 10% on Dell shortage spillover
NBIS BULL Cramer (Squawk on the Street) Leopold Aschenbrenner 5.6% stake disclosed; neocloud beneficiary
SNPS NEUTRAL Faber (Squawk on the Street) Elliott (Jesse Cohn) added to the board
INTC BULL Denise Chisholm (The Compound and Friends) Among top-10 S&P contributors YTD
INTC NEUTRAL Ian (Telecoms.com Podcast) Dual designer and foundry role; question whether Intel has explored time-scaling
Software (CRM/NOW) BEAR Adam Parker (Excess Returns) CTO pushback on pricing; OPEX/CAPEX pressure; revenue miss risk in about 5-6 quarters

🎯 What Changed This Week

The framing across the tape moved from demand validation to valuation discipline. With memory names in the trillion-dollar club and the SOX a historic distance above trend, even the bulls (Chisholm, Parker, Silverman) are now arguing from earnings durability and underweight positioning rather than from cheapness. The bears (Boockvar, Nicknum) have a cleaner setup: pull-forward signals, 1999 P/S analogs, and a China door that stays shut for NVIDIA.

The other genuinely new read-through is structural: Huawei's "time scaling" pitch and Anthropic's private-credit-funded compute stack both point at the same question, which is who ultimately finances and supplies the next leg of the build-out if EUV and balance-sheet constraints bind.

Sources