Nanya bets $10.7B on DRAM capacity - the memory capex race reaches Taiwan

Suivi de l'affaire : Capex mémoire : la course aux HBM/DRAM· Épisode 19/19

Infra & Compute il y a 1 h9Ajouter aux favoris

Nanya bets $10.7B on DRAM capacity - the memory capex race reaches Taiwan
Illustration : Léa Fontaine

Taiwan's Nanya Technology plans a $10.7 billion fab investment, adding another major player to a global DRAM expansion wave driven entirely by AI memory demand.

In plain terms. Nanya Technology, Taiwan's second-largest DRAM maker, announced plans for a $10.7 billion new fab targeting AI-driven memory demand and next-generation DRAM capacity.

Analysis. Nanya's commitment is the latest data point in a global memory capex race alongside SK Hynix, Micron, and Samsung. For Nanya - historically a laggard in advanced nodes, focused on conventional DRAM rather than HBM - a $10.7B bet signals that even second-tier players believe the AI memory supercycle has sufficient runway to justify a multi-year build. The risk is cycle timing: DRAM is notoriously boom-bust, and a fab coming online in 2028-29 faces demand uncertainty from two directions. First, AI efficiency improvements may reduce memory-per-parameter requirements as model architectures evolve. Second, if the hyperscaler capex cycle peaks before 2028, new capacity arrives into a cooling market. Nanya's financial position makes this bet existentially high-stakes - unlike SK Hynix, it has limited cushion if timing is off by even one cycle.

So what. The global memory capex expansion is now a multi-trillion-dollar industrial bet on AI compute growth persisting through 2029+. If model efficiency improves faster than supply comes online, consolidation follows - and smaller players like Nanya face the worst outcome.

Ressources, à tester

Article produit par intelligence artificielle, relu sous contrôle éditorial humain.

Notre rédaction
Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

SSHMonitoringAI Ops
Get early access
Cet article vous a-t-il été utile ?

10 personnes ont aimé cet article

J'aime
R
Ravi NairInfrastructure & compute
🇮🇳 Puces, datacenters, énergie, cloud.
Partager :
Commentaires (9)

Connectez-vous pour rejoindre la discussion.

ArtLover88 07 Aug 2026 · 08:43

A $10.7B gamble is brutal-but if AI keeps devouring DRAM, Nanya might just win big. Still, betting the house on a single trend feels risky.

Dr. L. 07 Aug 2026 · 07:01

This investment shows how AI is reshaping the entire tech supply chain. Wonder if smaller players will get squeezed out as costs spiral.

HistoryBuff 2 07 Aug 2026 · 06:55

Nanya’s bet mirrors Samsung’s recent moves, but I wonder if the market can really absorb all this new DRAM without a crash when the AI fad fades.

BookWorm88 07 Aug 2026 · 06:47

If the AI hype cools off, won’t Nanya be stuck with idle capacity? Maybe the real bet isn’t AI demand but who blinks first in this arms race.

TechGuru99 07 Aug 2026 · 06:46

That’s a staggering number. If AI demand stalls, won’t Nanya face the same oversupply nightmare as the 2019 DRAM crash?

TechSavvy 07 Aug 2026 · 06:43

The DRAM capex race is getting out of hand. $10.7B for a single fab? That’s a massive gamble on sustained AI growth. What if the market hits a wall before these assets even pay off?

Alex 07 Aug 2026 · 06:40

That capex is insane-$10.7B for a single fab? But if AI keeps gobbling up memory like it is, who’s really taking the risk here: Nanya or the rest of us waiting to see if the demand keeps up that hot?

FilmBuffNYC 07 Aug 2026 · 06:27

That’s a massive bet. Wonder if the ROI timeline aligns with AI’s actual memory needs-overbuilding seems risky.

J.P.R. 2 07 Aug 2026 · 06:25

If AI demand keeps growing this fast, will we see a price war before the new fabs even come online? Or is this just another bubble like the dot-com era.

Le fil de l'affaire

Capex mémoire : la course aux HBM/DRAM

  1. 1Nanya Tech quadruple son capex 2027 - Taïwan mise gros sur le retour du DRAM11/07/2026
  2. 2Samsung avance l'ouverture de la fab Yongin 1 à 2029 - le calendrier compute mondial se compresse12/07/2026
  3. 3SK Hynix -12 %, KOSPI casse le 7000 : le complexe HBM prend une correction13/07/2026
  4. 4Corée : les « big money » se dégagent de Samsung/SK Hynix - signal de pic HBM ?13/07/2026
  5. 5CXMT lève 8,5 Md$ : la Chine cote sa mémoire au sommet de la pénurie14/07/2026
  6. 6CXMT files $9.8B IPO: China's memory push seeks a public balance sheet at cycle peak17/07/2026
  7. 7Micron s'infiltre dans l'auto avec Qualcomm et Hyundai Mobis : la mémoire IA gagne le cockpit17/07/2026
  8. 8AI memory crunch reaches India's smartphones: the DRAM squeeze hits the consumer edge18/07/2026
  9. 9Oppo et Vivo refuseraient les tarifs mémoire Samsung Q3 : le pincement HBM/DRAM devient un bras de fer23/07/2026
  10. 10SK Hynix Q2 record: memory is where the AI capex pinch prints first26/07/2026
  11. 11CXMT surges 500% on debut, tops Intel's market cap - Chinese memory reprices at cycle peak27/07/2026
  12. 12Samsung pushes HBM5 base die to 2nm targeting +50% speed - the memory war moves to logic-class nodes27/07/2026
  13. 13SK Hynix eyes record Q2 profit on AI memory boom - the memory cycle enters its blow-off top27/07/2026
  14. 14Kioxia projette un bénéfice annuel multiplié par 31 : la NAND entre dans le supercycle IA31/07/2026
  15. 15CXMT closes in on LPDDR6 mass production at 12,800 Mbps - the Chinese memory step-up01/08/2026
  16. 16Montage starts CXL 3.2 chip production - China gets a second memory-fabric supplier01/08/2026
  17. 17Conventional DRAM and NAND hit record highs in July: the memory squeeze widens03/08/2026
  18. 18Apple's talks with China's CXMT memory chipmaker hit a price wall06/08/2026
  19. 19Nanya bets $10.7B on DRAM capacity - the memory capex race reaches Taiwan07/08/2026
Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

Get early access
Rubriques
Explorer
Informations