Nvidia当前需求强劲但2029年内存市场预测仍存不确定性
Nvidia当前需求强劲但2029年内存市场预测仍存不确定性
Conclusion
Jensen Huang’s view is well supported for current Nvidia chip demand, but much less certain as a forecast of the memory market in 2029. Nvidia’s latest reported quarter showed rapid revenue growth and high profitability, and management was still guiding for growth constrained by supply. Memory suppliers have also warned of shortages lasting into the late decade. But those figures establish present demand—not that today’s shortages cannot turn into a glut after new capacity arrives. The other key vulnerability is whether AI customers can finance and earn adequate returns on the infrastructure they plan to build.
Data as of October 24, 2026. Nvidia’s latest reported quarter I could verify was Q2 FY2027, ended and reported on August 26.
What Nvidia’s financials say
- Revenue was , up and . Data Center revenue was , and GAAP gross margin was . This is strong evidence that demand was translating into shipped products and profits—not just a bullish narrative.
- Nvidia forecast Q3 FY2027 revenue of $108.0 billion, plus or minus 2%, with gross margin expected at 74.0%, plus or minus 50 basis points. The company said its outlook assumed no Data Center compute revenue from China. That guidance supports continued demand, while also making clear that sales depend on supply and that management did not include China compute in its forecast.
- Growth calculations: (current-quarter revenue − comparison-quarter revenue) ÷ comparison-quarter revenue.
Nvidia’s Q2 release described demand as accelerating and the AI infrastructure buildout as active. Reuters’ account of the earnings outlook reported that management expected about 70% revenue growth in fiscal 2028, while noting that memory-component shortages were limiting how quickly Nvidia could expand. This supports Huang’s near-term demand case, but it remains management’s forward outlook, not an independent guarantee of customer spending.
What the memory evidence says
Huang’s October interview was specifically about the possibility of a memory glut by 2029. He argued that customers can keep using more memory, including HBM and LPDDR, and said he saw no reason to expect oversupply by then. CNBC reported that the dispute had sharpened after concerns about a future supply wave weighed on memory stocks; it also noted a Chinese supplier’s new memory architecture as a source of competitive concern. CNBC’s report of Huang’s comments and the counter-argument
There is meaningful support for the shortage side of the debate: SK Hynix’s CEO expected memory shortages to persist through , while the company planned to begin HBM4E production in Indiana in . Samsung had also said shortages could extend into 2028. These are relevant supplier views, but they are forecasts from companies that benefit from tight supply—not proof of the eventual market balance. Reuters on SK Hynix · Reuters on Samsung
One important distinction: “memory” covers different products and uses. A glut in some memory categories would not automatically mean abundant HBM for AI accelerators; conversely, continuing HBM demand would not rule out oversupply elsewhere. Nvidia’s current financial results demonstrate demand for its systems, but do not settle the balance of every memory market years from now.
What could challenge the claim
-
New capacity arrives faster than demand. The current shortage encourages supplier investment and expansion. Those additions may take time to come online, but if several ramps coincide with slower AI spending, a late-decade surplus becomes more plausible. Nvidia itself was reported to be working with suppliers to expand capacity for several components—evidence of current tightness, but also of efforts to add supply. The Wall Street Journal on Nvidia’s supplier agreements
-
Customers may face financing or return constraints. At GTC Berlin, Huang said capital availability was a constraint on the growth of neocloud providers. CNBC also reported investor scrutiny of debt-funded AI infrastructure and concerns about the credit quality of some businesses in that ecosystem. Strong interest in compute does not necessarily become durable, funded orders if financing tightens or customers cannot monetize capacity. CNBC on neocloud financing and debt concerns
-
The 2029 “glut” case is a credible scenario, not a settled forecast. CNBC reported that a South Korean opinion column argued for a large future supply increase, and that CXMT’s technology announcement added to concerns about competition. Neither establishes that a glut will occur; together, they show why Huang’s confident demand extrapolation is contestable. A memory surplus could lower Nvidia’s component costs, but if it reflected weakening AI investment, the associated decline in accelerator orders could matter more than the cheaper inputs.
-
Competition and efficiency could reduce the amount of Nvidia compute customers need. Huang himself acknowledged China’s accelerating semiconductor efforts in the CNBC interview. If competing chips, domestic alternatives, or more efficient workloads capture more deployments, demand for AI infrastructure could grow more slowly than Nvidia’s current supply-constrained outlook implies.
Bottom line: Nvidia’s latest reported financials and guidance make “chip demand remains strong” a credible description of the business as of the data cut-off. The claim that memory will not be in surplus by 2029 is plausible and consistent with supplier warnings, but remains a long-range forecast. The strongest challenge is a combination of new capacity coming online and customers’ financing or returns falling short—not simply the existence of more memory.