NVIDIA’s Huang Sees Low Risk of Global Memory Oversupply Before 2029
Редакция ЦОД:Главное
NVIDIA CEO Jensen Huang said at NVIDIA GTC in Berlin that a global memory oversupply is unlikely before 2029, citing continued investment in AI infrastructure. He also said NVIDIA will continue broad use of LPDDR in its data centers and identified operating capital availability, rather than demand, as the principal constraint on AI-sector growth.
- Huang said global memory oversupply is unlikely before 2029, in his assessment.
- NVIDIA plans to continue broad LPDDR use in its data centers.
- Huang said LPDDR deployment delivered twice the prior level of energy efficiency, without disclosing the comparison basis or methodology.
- NVIDIA is open to cooperation with Chinese memory makers, but no CXMT agreement or product deployment was disclosed.
- Huang identified operating-capital availability as the main AI-industry growth constraint.
Jensen Huang said a global memory glut is unlikely before 2029 as investment in AI infrastructure continues to expand. The NVIDIA CEO made the comments at NVIDIA GTC in Berlin amid market concerns over potential memory overproduction and pressure on sector equities following reports of capacity expansions and CXMT’s new DRAM technology.
Huang said demand for HBM and LPDDR continues to grow. NVIDIA intends to maintain broad use of LPDDR in its data centers, according to his comments. He said NVIDIA was the first to deploy smartphone-oriented LPDDR in data centers and that its use delivered energy efficiency twice the previous level.
For AI data-center operators and infrastructure suppliers, the comments point to memory availability and memory power efficiency as relevant variables in AI platform design. If NVIDIA’s view proves accurate, sustained deployment and equipping of AI infrastructure could continue to support demand for HBM and LPDDR, while more energy-efficient memory architectures could help manage AI data-center energy consumption.
Huang also said NVIDIA is open to working with Chinese memory manufacturers, including potentially using CXMT products. That openness should not be read as evidence of a contract with CXMT or of actual deployment of CXMT memory in NVIDIA systems.
On AI-cloud expansion, Huang said demand for services from new cloud providers, or neoclouds, has exceeded expectations. In his assessment, the sector’s principal growth constraint is a shortage of operating capital rather than insufficient demand. This could mean that the rollout pace of new AI cloud capacity is shaped by access to financing, not only by end-market demand or technology availability.
If Huang’s assessment is correct, continued AI infrastructure build-outs could sustain demand for specialized memory, particularly HBM and LPDDR. His comments also suggest that financing availability may influence the pace of new AI cloud-capacity deployment alongside memory supply and technology availability.
- TechNews 科技新報 / КитайскийПервоисточник →
Материал подготовлен редакцией ЦОД: Главное на основе указанных первоисточников.