Nvidia's Memory Call Meets a Supply-Chain Bill

Nvidia chief executive Jensen Huang has told reporters he does not expect a glut of memory chips before 2029, pointing to sustained demand for HBM and a coming wave of LPDDR deployments. The remark, carried by media outlets on Friday, is a demand-side signal rather than a clean bill of health for the financing and delivery risks hanging over the AI build-out.

Huang's view looks past the current data-centre construction cycle and rests on memory consumption staying elevated even as new capacity comes online. He draws a line between continued HBM demand and the broader LPDDR usage he anticipates. For valuation purposes the distinction matters: this is a forecast about how the market develops, not a reported sales figure and not a guarantee that every build-out project turns a profit.

Investors showed how twitchy they are about that question on Thursday, when Nvidia slid alongside other AI names. The Associated Press reported concerns over borrowing costs for data centres and AI chips, compounded by doubts that AI valuations had run too far.

Orders Versus Deliveries

A concrete read on appetite for Nvidia hardware arrived from Reuters on Tuesday: ByteDance, Alibaba and Tencent ordered more than 3 million RTX PRO 5500 graphics cards, a combined value above $35 billion. Chinese customers' orders overall topped 6 million cards, according to the report.

Should investors sell immediately? Or is it worth buying Nvidia?

Bookings are not shipments, though. Reuters also reported that Nvidia warned customers it could not guarantee it would fill the orders, framing the situation against the backdrop of US export restrictions on China. The scale of the interest is clear; whether it converts into deliveries is not.

Capital is already being committed to close the gap. On Wednesday Nvidia announced agreements with supply-chain partners to lift substrate capacity 2.5-fold over three years, and said it will advance KYEC a $1.1 billion capacity reservation payment for new test facilities in the US. The prepayment shows that locking in extra capacity carries a financial cost up front — the point where Huang's long-run optimism meets the market's scepticism.

Three Products, Three Stages of Readiness

Alongside the memory outlook, Nvidia is pushing deeper integration of optical components into its chip architecture. Its Optics-on-Interposer design moves the optical unit onto the interposer, putting the focus on how components are arranged inside the chip package. It is a prototype, and investors should treat it as such rather than as a product in wide deployment — no order or booked revenue attaches to it.

Software is further along. Nvidia unveiled its Minerva platform on Thursday, which fuses telemetry from infrastructure and compute workloads to monitor AI data centres. IREN is already using the system, according to the company, and IBM, Lambda and Tesla are among those planning deployments — announced rollouts, not completed ones.

Nvidia at a turning point? This analysis reveals what investors need to know now.

On the infrastructure side, Wednesday brought the DSX POD, a modular pre-fabricated reference design built on Vera Rubin. Nvidia says deployment at a prepared site could shrink from six months to as little as 14 days, with partners due to ship the units worldwide in early 2027. The time saving depends on the site being ready in the first place.

Taken together, the chip design, the monitoring software and the turnkey facilities sit at very different levels of maturity. What ties them to Huang's memory forecast is the same underlying bet: that demand holds up long enough for the capacity now being financed to pay for itself. His projection supports the demand side of that equation. It does not settle the delivery and funding questions on the other.

Ad

Nvidia Stock: New Analysis - 25 October

Fresh Nvidia information released. What's the impact for investors? Our latest independent report examines recent figures and market trends.

Read our updated Nvidia analysis...