How Long Will the Memory Pullback Last? Phison CEO on the AI Super Cycle and the YMTC Black Swan

In episode 68 of The Storm Media's 'Off-Duty International Line,' host Lu Yi-chen sits with Phison Electronics CEO K.S. Pua (Pan Jiancheng in Chinese) and asks the question on every memory buyer's mind in 2026: how long can the pullback in soaring memory prices last — is it a durable top or a healthy breather inside a super cycle? With Yangtze Memory (YMTC) expanding aggressively in the background as a potential black swan, the answer matters for investors and hardware buyers alike.

Pullback or Trend Reversal?

Pua's frame breaks the classic cycle playbook. He argues memory demand has decoupled from PC and phone shipments and is now tied to the demand AI itself creates — like needing more fuel stations as traffic explodes. Inference workloads from chatbots, recommenders and enterprise agents generate trillions of tokens every second; those tokens are held in DRAM (working memory) and then persisted to NAND (non-volatile storage). A near-term price correction, in this view, does not erase a structural deficit; it is profit-taking inside a rally.

Phison's 2026 numbers back the thesis. The company posted record second-quarter revenue of NT$67.888 billion and EPS of NT$118.57, driven less by controller chip unit sales and more by a pivot to AI storage platforms. Its 'Phison 3.0' transformation is a shift from component vendor to systems partner: enterprise SSDs, aiDAPTIV memory expansion, an AI data platform and software layers. When Pua talks price, he is also describing his own business-model migration.

Super Cycle Math: $146 Billion Still Not Enough

Why does the super cycle (a multi-year stretch where demand outruns supply and prices stay elevated) keep extending? Deloitte projects combined 2027 CAPEX of Samsung, SK hynix and Micron at $146 billion — 3.4x the $43 billion of three years ago — yet new supply will not hit volume until 2029-2030, as fabs take 3-5 years to ramp. Meanwhile hyperscaler capex is headed for $1.14 trillion in 2027, with about $410 billion earmarked for memory purchases. Demand is outrunning the ability to build, and as SK hynix executive Kwak Noh-jung told Reuters, '2027 could be the worst supply year in industry history; demand will exceed capacity even after 2030.'

Price signals confirm it. Semianalysis expects DRAM prices to more than double in 2026 and keep rising double-digits in 2027, with structural tightness in high-bandwidth memory (HBM — the stacked ultra-fast memory next to AI accelerators) through 2027. TrendForce's September bulletin still sees NAND contract prices with upside. The spot-contract spread is widening — when the pump price runs ahead of the wholesale price, contract prices eventually catch up. The pullback is a delayed adjustment, not a trend reversal.

YMTC: From Third Place to a Shot at Number One

The second half of the show zooms in on the black swan: Yangtze Memory Technologies Corp. (YMTC). The Wuhan-based firm, China's only NAND maker, captured about 14% of global NAND shipments in Q2 2026 per Counterpoint, overtaking Kioxia for third (Samsung ~25%, SK hynix+Solidigm ~22%) despite a consumer-heavy mix that still leaves it fifth by revenue. Technologically it is mass-producing 267-layer 3D NAND on its proprietary Xtacking 4.0 architecture and developing 300-plus-layer generations.

Its capacity plan is aggressive. Omdia forecasts YMTC wafer output rising from 2.01 million in 2026 to 2.496 million in 2027 — a +486,000 wafer jump — while Samsung and SK hynix each add only about 75,000. Wuhan Fab 3 starts initial operation at year-end and ramps fully in 2027. In a window where the incumbents cannot open new NAND clean rooms at scale until 2028, YMTC is the only player adding meaningful incremental supply, which could soften spot prices after covering domestic demand.

Funding matches ambition. YMTC completed pre-IPO counseling in August for a Shanghai STAR listing targeting about $4.9 billion, with market chatter of a post-listing valuation above 1 trillion yuan. Its Q1 filing showed 47 billion yuan revenue and 33 billion yuan net profit — more than double last year's full-year total — with fab utilization near 98% and gross margin of 84.23% on storage chips versus 70.29% on SSDs. Proceeds are earmarked for capacity and R&D, and management has told investors it aims for the global No. 1 NAND spot by end-2027.

What the Market Expects: Loosening in Late 2027?

Will that black swan actually crash prices? Consensus is split. TrendForce argues that unlike DRAM, NAND supply will loosen from the second half of 2027 and price pressure will intensify — YMTC's new capacity plus soft consumer electronics demand. Chosun Biz and BusinessKorea flag YMTC's mix shift toward enterprise SSDs: in Q2 2026 eSSDs were 48% of total NAND bits shipped, up from 26% a year earlier. If YMTC grabs a bigger slice of that high-value pie, competition turns technological, not just about price. Morgan Stanley's super-cycle note takes the opposite side: AI data centers are the least price-sensitive buyers — Q4 server DRAM contracts up near 70%, NAND up 20-30% — they treat memory as a strategic must-have.

Pua's balanced takeaway is that the 2026 pullback does not mean the long cycle is over; it is a healthy digestion of a relentless run. The real test is late 2027. If YMTC's Wuhan ramp yields well, a short-lived NAND surplus could appear in late 2027 — but it would not solve the structural shortage in HBM and high-performance DRAM, and YMTC's access to the most advanced lithography tools remains constrained by U.S. Entity List restrictions. As Pua told Nikkei, it would be unhealthy if prices keep surging unchecked, and unsustainable premiums, like an overheating engine, could choke demand.

Three signals to watch are clear in practice: 1) YMTC's Wuhan Fab 3 yield curve and enterprise SSD mix — how fast it shifts from consumer to enterprise NAND, 2) Samsung/SK hynix Pyeongtaek P5 clean-room timing — whether new NAND supply indeed slips to 2028, 3) hyperscalers' multi-year supply contracts — Samsung's disclosure of 'requests for multi-year contracts for large-scale AI infrastructure beyond 2029' is telling. For near-term buyers the pullback is a buying window; for medium-term builders of AI infrastructure, memory remains a strategic bottleneck whose share of total spend keeps rising. The correction is short; the super cycle is long — YMTC may change not the price level but who gets to play the long game.