# Memory Becomes the Bottleneck the AI Buildout Pays For - Foundry & Chip Equipment - Week of June 28, 2026 > Foundry and chip-equipment newsletter for the week of June 28, 2026. Micron's blowout print reframes memory as a scarcity asset and ASML's CEO confirms a supply-limited equipment market, while the bear case sharpens into falsifiable, inflation-adjusted capex math. ## Foundry & Chip Equipment ### Week of June 28, 2026: Memory Becomes the Bottleneck the AI Buildout Pays For --- Funny kind of week. The names we actually cover, the litho, deposition, etch and metrology toolmakers, plus the leading-edge foundries, barely got a word in. Yet the most important thing the AI buildout learned this week is a foundry-and-equipment story dressed up as a memory story: the supply chain is now genuinely capacity-limited, customers are signing multi-year take-or-pay paper to lock in scarce supply, and pricing power is migrating to whoever owns the bottleneck. Micron just proved the playbook on memory. ASML's CEO quietly confirmed the same dynamic is running through litho. And the bears showed up with a real argument, not a reflex. ## TL;DR - **Micron's blowout (May quarter: USD 41.46B revenue, 84.9% gross margin, EPS USD 25.11) reframed memory from commodity cyclical to scarcity asset**, roughly 16 long-term supply agreements covering ~USD 100B of committed revenue with price floors out to 2030. The same "lock in scarce capacity" behavior is the bull case for WFE. - **ASML's CEO says he sees "a pull" on both DUV and EUV orders and a "supply-limited market for AI, for semiconductor, for quite a few years,"** the cleanest operator read on equipment demand this week. - **The bear case got sharper, not quieter:** real hyperscaler capex may be flat-to-down once you strip component inflation, the semi-dollar growth curve is set to decelerate hard into 2027, and double/triple-ordering may be inflating the order book. ## What's New **Micron turned a print into a thesis.** On [Schwab Network's earnings alert](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOgVt84YS7gtXPHxkqQAXVTjUl-2BrJ2xyZkZtku9VJyF-2F1TadPDErpy3hYjiNr6Ek04Ny3FD0NV-2B7PMSnZksqcVLAdsMcGtVUZdXqjoR0HbmfcA-3D-3DU82S_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBREriCM9NEfn0j0PgETsaswuyGbUxX9nLFwdRtWPB8vWo-2FeNb5kt1kvV-2FxcEcKIzhTgl4GWVmpqewyhrWf6eiGcmt-2FbyUTiEeJ0Zlc0VlXToYcBa18nlqSAEYD0x3csgc-2FA-3D-3D) (Jun 24), the numbers landed: May-quarter revenue of USD 41.46B against a ~USD 36.5B estimate, 84.9% gross margin, EPS of USD 25.11, and an August-quarter guide of USD 49–51B revenue with EPS of USD 30–32. CEO Sanjay Mehrotra framed it as "the strategic value of memory in the AI era." On [Closing Bell Overtime](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOiz-2FwLrZ2bH6AgzSxRaFy6npfEiTPpMudb6A91kYODSQ0lWwZ3rK-2BiAJUMA-2FynBux846Bzl6he-2FDR0AxaYPwJpCissCA4sTP9ImDeolTIvj8A-3D-3DZ7n8_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRFJBSYJwmZlK02-2F8L-2B6dKi3-2BdwQXUT3ElLKLoRXJIc9z3JVEjFiepEzntZaao3dRc9pCSyFsVUjQx-2FQ6eAUxVB-2Fiq-2Fqo0-2F9wxQHwB2wq1cTHCTytfzUbF0OM-2BOVFBwyS8g-3D-3D) (Jun 24), Susquehanna's Mehdi Hosseini, a sell-side analyst who covers the name, put hard structure on it: ~16 multi-year strategic agreements, DRAM demand growth in the low-to-mid 20s% for 2026, supply staying tight through at least 2027. Why it matters: this is the demand pull that ultimately fills fabs and books tools. **ASML's Christophe Fouquet gave the operator read on equipment.** On [Bloomberg Talks](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOgbV-2BQXDXJ0aN4-2FY90f6OpM0Zy9h68ZuH-2BNu9w9MWc8eB7V-2FIREjxcZRO34UF7yqaHsBnE7C3RAE-2BkV0Om7GgpOCwkc5gV0KRazqm5upxJ-2B5w-3D-3DUYKd_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRBSfrVYLYFRkhiLElKzFjaNcFP4kofkDBhJi-2BoebMv4awvVDiVZdmhnjUZGk6nXrF94TESnMAKI9zKlmCkkZwCwbL7ejlfwXNSDwS3UejqdQZq2-2BN0GXzigLCzAagdZXwA-3D-3D) (Jun 17), the ASML CEO, the one insider whose order book *is* the WFE cycle, said plainly: "We see a pull. We see customer continuing to create visibility, longer-term visibility, telling us, well, this is a bit here to stay." He expects "a supply-limited market for AI, for semiconductor, for quite a few years," and on the recurring China export rumor (a US House member's April letter alleging DUV shipments), he was emphatic: "we have been following every single rule... that's true for DUV, that's true for EUV." No High-NA or backlog numbers in the interview, but the directional signal is the strongest live data point on tool demand this week. **Lip-Bu Tan made Intel Foundry's case in his own words.** On [No Priors](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOiOEZu1cE-2BdhA84m0Q-2F0rbng9ZQ-2FJ542e-2Ftrrywn4aCzHjkppM2QiF3hKV4MKXgeuvoSFC7J1fbF7ZpqSwhXxy3Ey-2BGPpq7jHebR1EkNI-2FHuQ-3D-3D0XU0_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRGn8OPaCqDJKBA8tdiM1dOqen-2BtN5C7tVnGnKG1g6h-2FOLVp97-2F9DrEu6msk012hhCUDTPSQp-2BtqM8Ft6Rr7HT6UHpUOWo7Rtpk5V-2F6juVHWBhTvdWzj0VBY-2BUaICt9bwTg-3D-3D) (Jun 18), Intel's CEO described the foundry honestly as "a service business... a trust business. If people want to give you orders... if the yield is not good, they will be toast," so the focus is "yield, defect density, cycle time." He confirmed the strengthened balance sheet (US government stake, Jensen Huang's USD 5B that he says is "USD 25B now," SoftBank), the TerraFab collaboration with Musk, and named the real bottlenecks for the cycle: power, memory, and, less appreciated, helium. This is the strategy; execution is still the question. **OpenAI and Broadcom put a number on custom silicon.** On [Tech Brew Ride Home](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOiFLn1Qls7LxM4UXQeotJFoIOx2tCMaYV-2BemvPKww0AWYJGrsvi1zLD4evcwtpbwjFZkf5QVBlh5-2F27ZMmgCU-2FR4df25rJt1uD9cA3kmb6liw-3D-3DmI2R_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRMDgR7E3rHulX6j0FlQWyEMEJuB-2F3RhMT1LpEO-2BW231LaDLla4MkhVrwotkcMM5GTjRLWSTh36btdc-2BdRJ2k9sbx8LT0JbeLwVByUCyi-2FxawYWqcDMeOp-2BkR7uCVpi9hXg-3D-3D) (Jun 24), Broadcom CEO Hock Tan said the OpenAI "Jalapeno" inference chip shows ~50% cost savings versus typical AI GPUs, with finalized parts going into Microsoft data centers later this year, and that he expects to "do better" than his prior 1.3-gigawatt deployment forecast for next year "because there is a lot of demand." OpenAI hardware head Richard Ho said it was purpose-built for LLM inference with "substantially better performance-per-watt." Read-through: more leading-edge wafers and more advanced packaging, just routed through a custom-ASIC channel rather than merchant GPUs. ## The Debate The bull side was voiced loudly and by the right people. The structural case is that AI infrastructure is supply-limited for years (Fouquet), that customers are now signing price-floored, multi-year supply paper that didn't exist in prior cycles (Micron's agreements), and that the bottleneck resource captures the economics. On [Squawk on the Street](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOgIjJqIjPoPkGsWTXJa3-2FhUGd0ZO6k0FozC5jLk0Xk-2BM56R0FmAOERqWwvg2LIw9XaSmqIpeoaUeZF6or5XOqvsEcTrqD81rvwa5HjXc-2FIlQQ-3D-3DKBBY_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBROtjxeTtiDUofQD-2BzZLHwy4ddMhNs-2Fl25rdvGKBFvdAk6C-2BMgLgaMx-2BLcneswZIaCG-2B6f7jWWdw3g7pvgMz0coVOTNQqOkA7F-2BijobxRsRXrCGMtaMHqhvZt-2BaQXteaSXA-3D-3D) (Jun 25), Trivariate's Adam Parker and Roundhill's Dave Mazza argued memory deserves to be re-rated as a structural, not cyclical, opportunity. But, and this is the part that earns its keep, the bear case this week was not a reflex; it was a model. On [Monetary Matters](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOgXa9gT1kOJP4z9zFbPneLN1sxQJ6QpqxdNS3h-2F-2BTSP-2FYzSxpdG1SlhnLX6fkZVMePfraeZ-2FBeHNy0FY6Qjf-2BrsvQDfbUuGqiPR7OTpbHMx9Q-3D-3DK8y3_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRCZR1JewhrwP9DGsPTfMe6F7PSsdrGsl8W5cxaIqWEWGJSAki7RZapozDijI-2FvADncfPv-2BfbwiMo4w4ywmtwVBP4ZXpvEOLC9-2BUr-2BDeV4Pf77n9qIj-2B3Beo0GdgACpF8-2FQ-3D-3D) (Jun 15), economist David Woo argued the five hyperscalers' combined capex actually *declined* quarter-on-quarter in Q1 2026 and that headline guidance raises mostly reflect component inflation, "an income transfer to memory suppliers" dressed up as growth. On [The Pomp Podcast](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOiDMBFOXRD5xjIxL9X-2BtprLoa5riGqo4rtexaAlZbsisISbql-2Br77dWOg7p-2BNtEbsQ7Odp0G0BfIWsJDAemy7-2Fuem8MgzYuJq9zTGwzZ1WitQ-3D-3D1euK_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRKo5bEjPUqEM0x-2FmLBhDx8T6IehDBy7OxtQi6QBPOOMgV9gK66GtEar5jCxPo7zD5YH7BQo6IVY74NCwcmbOcQ-2FCvsELuM3t3DjnOWgV-2FM8TUENKIvrdsJTHEB5f4SWr9g-3D-3D) (Jun 20), Jordi Visser cited Gartner's curve, semi-dollar growth near 100% in 2025 collapsing toward 30% by 2027, and flagged a possible "capex air pocket." And on [RiskReversal](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOgxzH1dESpuj7aji3gj-2FWPU-2F-2FBojB-2BR10Kx5tsW3Y3nxvJcIBp5sqXlYgITZC4tf4ra8FlqBIGeoHA3qXr1x7H-2BZNwQq8D5TczTs-2Bt1JebiaA-3D-3Dmj6h_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRF2I-2BMfweYfSIxAJXQTRaGrBCUsaOsHuhSeElJA95in8eFiGle5ikWfGEGexsJnnSQ5UEvCKOgv-2B2NqU7yNQpvsmQypQWVibbH6SsxuSqcK1kmb12GkuUIjzGvlDiH2Z1Q-3D-3D) (Jun 22), Danny Moses pointed at double- and triple-ordering inflating the apparent demand. The crispest single bear data point came from [Cabot Street Check](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOi68DXPsq8I1kcafh8fIXeB-2BCHnU6KgzgK04s3bT7vo61f-2Fhco3FLw4KlXsQQmq0X3lE9p8xaVC-2FtnFhe1-2FQXb6UeoqJZvYdj9Xtx5ebweLmw-3D-3DhVxb_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRKcoTlP2pAGDhfqAsZg4368uvdMXl4CIFwz9cBd0lD6dVpP8Am946yL7BEvtNGxARGtXwwn8kT0w0aF9bXgyt1NCjqkFFlgSTW27EY1kOCu0Q80jNKT0kgycUV234y-2Bhbg-3D-3D) (Jun 26): Micron's ~USD 100B of committed contracts is, against a USD 50B/quarter run-rate, *only about six months of backlog*, visibility, but not the multi-year fortress the headline implies. The Intel/Apple skeptic case was sharper than the partnership hype: [Deepwater's Gene Munster on Fast Money](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOhIPVwR8Xv9q9VvJzigGgivq1lsuu5hKeSuKWNhCmsNDegQlyTH4TtHpai-2F3PhjXtzdk-2BGBg21E2Sej1ZWmbi1B0Q8-2FmXdBb9WeAxMhjxvnXQ-3D-3DamGt_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRHlBSAFMB4pPUX41BNVROBuk1501Nn93aqqq6is71WrvYuVMZzpKZExhdUYtIubyOd6Betkr-2FLOZAQ2rP7yIomJX-2BT6J3BL3GGzR00FnIVir3oLth-2FCtAyxZY80DdFi8mA-3D-3D) (Jun 18) pegged best-case Intel capture at ~10% of Apple's TSMC business (~USD 21B) and called the move a "meme stock" phenomenon, against a more constructive 4x-book framing from tech research on [Squawk on the Street](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOig1LZKflpFPra3CZwWD236Pbau2ODjHn5VVPZtn3mCiW1w36-2Bv-2Bc1eCSQnSr-2FlPAcaPGlUd9L4DiNdADoQF-2BlH8Np-2FuysO92-2Fd6zb2GfgYLA-3D-3Dsp0O_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRM3YwCPIApyFiEWZJMl5eoSloJybm1piLr5zAc2m-2F-2BWNHZLb3ijA9Zq3Ru6mXnbpAc4ePtRFiacwxsRuXP6dVZPKkATfLYmk0KxysYB2X61BPgW0OXG0r3Yan-2FyLMHflaA-3D-3D) (Jun 18). ## Read-Throughs - **CoWoS / advanced packaging:** No standalone CoWoS or SoIC bottleneck discussion this week, but the HBM pull is the proxy, Micron said supply shortages will take "considerable time to improve" with "no line of sight" to balance before ~2028. That demand has to be packaged somewhere. - **EUV / litho:** Fouquet's "pull" comment is the read; treat it as confirmation of durable bookings, not a quantified guide. - **Fabless exposed to wafer-price:** Apple is already paying, it raised MacBook/iPad prices by USD 100–500 on memory costs ([Closing Bell Overtime](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOjNvN-2BNl-2BlvjMiPyXJ86OmiYJbNfsEcoHgfyUO40v8Fdpk2x5xT-2BpkROQIEDwIOjC3QiORJU1Z8Ilh4Brgh-2BOSkdri4Gf5zH6yDmB9MEwHWXw-3D-3Du6WX_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRE9eUMbBX4dF7VzSXet2ZrAJrHpCiIJNEIDrVpNs4AooES-2FKtuxF-2FZJfB2zzerNQLEb9ynXNWU-2FWPXhno0NkEMwYuhZw3QEyRbDCujGsfah86btTEeF70bKY8gxqA3jD7w-3D-3D), Jun 25). The same cost-pass-through logic applies as leading-edge wafer ASPs climb. - **Quantum hardware (long-dated frontier):** The frontier kept moving. On [Lead-Lag Live](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOjGRJJqAJTz87rMmqiUS78O7pzvxgTt36exFd1iyvKaFVVOEgzV6W3ErCSYMkPe23MIrXRJiRA1EOUpEKkyqE6KwOa3rA1i4Bx1qiZMuFQpxA-3D-3DRcqi_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRI1O2wJxUGEXT3KmXrRXQmOK2ASizrBD-2FpfK67deer-2FACP0PZrnW5K5z4RrpHh0zlP0uhTn-2FM9SpvNFIbZB9-2F-2FjuaYBH2cjbJ063tf-2FPKs8xw5ptgQ-2F-2F1dNFt-2Boyaf7FSA-3D-3D) (Jun 15), WisdomTree's Christopher Gannatti walked through IBM's superconducting stack, near-absolute-zero cooling, helium isotopes, gold piping, and IBM's public target of error-corrected machines by 2029, plus its USD 1B slice of the USD 2B government quantum award. On [The New Quantum Era](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOjqm8xN-2FD4N5A-2FOunvyvkN9wx0eWEQ0GY55QqYxM7QonCnp5APo3C5a2Ex-2BLl71R6UboB8ms0EfzRdw-2BWxPdouLyUJOzeelJZxCn5U1XA-2BXdw-3D-3DuH0T_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBROau5o2-2B3P-2BMdPFa6eOPG31EN3OXEVbA9JQlWv-2B49whh0zIOwWDVRyVp2TMrDbWNlugsZPeyI5J-2FpUN7MAu-2FXoZ-2B2tBOFkQY3vSfxUB4BAvE75D2HZn7ueAj8jiTolkjBQ-3D-3D) (Jun 22), ArrowQ CEO Nick Farina said they've controlled 2,432 electrons-on-helium on their "Wonder Lake" chip and are targeting a 10,000-qubit tape-out by late 2028. And on [Squawk on the Street](http://url7324.matterfact.com/ls/click?upn=u001.idHmPrr2Geh7KYLAsTy7NkrIVb-2FgA4pmf2rMXQwGcOhBu9adelyGYUDyJyWreeX-2BQP9-2Fr9d7bSG8Lap6R8P-2F93SN-2FaFHBu3-2B2mFNaGtwo-2BMnfyiPy-2FzRrcghslmdrKmvqBiiST5AblOM79pqtlUepw-3D-3DZ2RR_7mLGwmUci-2BLaXswv9WX1yTgqn3Wad-2FotHhzHgSNAZbVt9mTrySXo5HAhy8BnjCF-2FgMPL85dARsDOU9pAacPBRKN-2FkK8bFMNe26dKq5S8cGFGKuZKBo8H-2BwKGX5B-2FSfMfwLyjO0kLtuim-2Fja1rW0qs2JO0kBDxog5kxKYZHSZwZIM6TMd5zh-2FvoOiSJfJXfK2hr-2Bn2eLvh9Hi4lN6Vrj2AA-3D-3D) (Jun 23), SandboxAQ's Jack Hidary said PsiQuantum, Google, Microsoft, IonQ and Quantinuum have all hit "significant engineering milestones." For us this is a multi-year read on cryogenic and control-hardware demand, not a 2026 catalyst, but the cadence is accelerating. ## What Changed The notable shift versus prior weeks is the *quality* of the bear case. For months the skepticism was vibes; this week it was Woo's inflation-adjusted capex math and Visser's deceleration curve, falsifiable claims with dates attached. That, combined with SK Hynix's pre-Micron hint of an HBM slowdown rattling the tape, is the thing to watch into next earnings: not whether the cycle is "real," but how much of it is already in the numbers.