Nvidia Wires Up the Back End: Test-Capacity Prepayment and a Per-GPU Software Trial Balloon
Published on 10/28/2026 at 08:01 | Editorial boerse-global.de
Nvidia is attacking the AI buildout from two directions at once — shoring up the factory floor that feeds its accelerators and floating a fresh way to charge for the software that runs on them. The common thread is preparation, not payoff: both moves sketch out future capacity and future revenue rather than booked results.
At the hardware end, the chipmaker said on October 21 that it and a group of suppliers aim to lift substrate capacity 2.5-fold within three years. Amkor, Ibiden, KYEC, Samsung Electro-Mechanics, SPIL and Unimicron are among the partners taking part, and the effort spans several stages of production rather than a single link in the chain.
Some of those partners have put numbers on their ambitions. Amkor is targeting a more than 300% increase in interposer-wafer capacity by the end of 2028, while SPIL intends to expand capacity for optical engines to 15 times current levels by mid-2027. Those targets are goals stretched over coming years, not capacity available today — a distinction that matters when weighing what the announcements actually deliver.
Cash on the Table at KYEC
One piece of the plan comes with money attached. Nvidia is putting up a $1.1 billion advance to KYEC to reserve capacity, earmarked for new test facilities in the United States. That goes beyond a memorandum of understanding: the prepayment is explicitly tied to locking in capacity, setting it apart from the expansion targets sketched by other partners. Here Nvidia is linking its roadmap directly to a financial commitment for a specific slice of the supply chain.
Should investors sell immediately? Or is it worth buying Nvidia?
The supplier plans line up with what customers say they intend to build. Oracle has outlined plans to scale Superclusters built on Vera Rubin NVL72 to multiple gigawatts, and Meta likewise intends to broaden its use of Vera Rubin and Vera CPUs. Those deployments give the manufacturing initiative a concrete backdrop — but they describe planned applications, while the supply agreements are meant to widen the industrial base beneath them. Neither translates into additional revenue already booked.
Huang's $2,000 Question
On the software side, CEO Jensen Huang used an October 22 appearance at GTC Berlin to raise the idea of licensing the company's agent software suite at roughly $2,000 per GPU per year. According to media reports, the figure represents a pricing approach that would tie software use to the hardware deployed — an additional earnings avenue, not reported income.
No customer contract and no recognized software revenue sit behind that number. What would ultimately determine its economic weight is how widely customers actually adopt such an offering. Alongside it, Nvidia pointed to its OpenShell Reference Architecture, unveiled October 21, which the company says is designed to deploy and manage AI agents across enterprise infrastructure and edge systems. Nvidia called the architecture immediately available and named partners spanning cloud, software, cybersecurity and systems.
The two steps carry different evidentiary weight. The architecture is an offering that exists now; Huang's licensing remark describes a possible route to monetization. Availability alone says nothing about commercial success.
Orders Outrunning Output
Deliverability remains the other test. Reuters reported on October 21 that ByteDance, Alibaba and Tencent placed orders for more than 3 million RTX PRO 5500 graphics cards, with a combined list value above $35 billion. Per that report, Nvidia could not guarantee it would fill the orders in full, because demand outstripped planned production. Big order interest is therefore not proof of revenue already realized — for investors, the gap between the two is where the valuation argument lives.
Read together, the threads trace different stages of the same journey. The substrate agreements aim to create the conditions for additional shipments, though announced capacity is not yet available production. The $1.1 billion KYEC prepayment buys reserved test capacity in the U.S. And the licensing idea points toward recurring software income if customers convert. The question for shareholders is not simply how fast demand grows, but whether Nvidia can serve it — and turn new software offerings into paying usage.
Ad
Nvidia Stock: New Analysis - 28 October
Fresh Nvidia information released. What's the impact for investors? Our latest independent report examines recent figures and market trends.
