Nvidia's Factory Floor: Rack-Scale Manufacturing, a Monitoring Platform and One Optical Prototype
Published on 10/25/2026 at 11:02 | Editorial boerse-global.de
Morgan Stanley is sticking with Nvidia as its top pick in semiconductors, and the reasoning has as much to do with production lines as with order books. The bank expects strong growth in 2027, citing supply-chain expansion, more efficient AI data centres and the Vera Rubin platform. Its confidence rests on Nvidia and its partners being able to build and deliver systems more efficiently — not simply on additional orders.
That thesis got support on Wednesday in Berlin, where CEO Jensen Huang said the company can now manufacture supercomputers at scale. Under the NVLink-72 design, rack assembly moves into the supply chain rather than happening on site at the data centre. Huang's account: delivery speeds up, and hundreds of data-centre projects worldwide can be served at the same time. Large computer manufacturers are involved in production, shifting a key step from the deployment site to manufacturing partners.
Demand, by Huang's own telling, still outruns Nvidia's supply capacity. In the BG2 podcast published Thursday, he described orders arriving almost weekly or even daily, with cloud providers, specialist AI clouds and equipment makers all buying heavily — and order volumes growing even as Nvidia raises output. That supports the growth case, but it is no guarantee that every ordered system goes straight into productive use. Huang himself listed power shortages, unfinished data centres and missing manufacturing components as obstacles to expansion.
How concrete those limits are shows up in reporting on Amazon. According to internal emails reviewed by Business Insider, the company accepts some Nvidia server racks before target sites have sufficient power, leaving the equipment waiting in interim storage. Amazon called the practice normal business. The scale and waiting times are not fully known, and a standstill across the entire GPU fleet is not established.
Should investors sell immediately? Or is it worth buying Nvidia?
Seaport Research takes a markedly more sceptical view of the infrastructure hurdles. Of 55 gigawatts of AI data-centre capacity planned worldwide through 2027, its analysis finds only 30 gigawatts could be built in time. Analyst Jay Goldberg points to a shortage of electricians, difficulties procuring gas turbines and delayed grid connections. The disagreement is therefore about execution: Morgan Stanley sees the conditions for strong growth, while Seaport doubts the necessary infrastructure will be finished on schedule.
Beyond the rack: monitoring software and a prototype
Nvidia is also pushing integration at the component level. An Optics-on-Interposer design moves the optical unit onto the interposer, putting the focus on how components are arranged inside the chip package. Nvidia is presenting a prototype here — a development stage investors should distinguish from a product in broad use. The news concerns technical development first and cannot be equated with an order or realised revenue.
More tangible is Minerva, unveiled Thursday. The platform combines telemetry from infrastructure and compute workloads to monitor AI data centres. IREN already uses the system, according to the company, and IBM, Lambda and Tesla are among those planning deployments — announced rollouts, not completed installations. The distinction matters: existing use gives investors a different reference point than a technical demonstration, while planned installations remain projects rather than finished implementations.
Nvidia is also targeting faster deployment of whole facilities. DSX POD, presented Wednesday, is a modular pre-fabricated reference design built on Vera Rubin. Nvidia says provisioning 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 potential time saving remains tied to the precondition of a prepared site.
The three efforts sit at different levels of maturity — chip architecture, monitoring of running systems and delivery of entire plants. For Nvidia shareholders, serial manufacturing is real progress, but no substitute for customer sites that are ready to operate. What counts is not only how fast Nvidia can ship, but how fast customers can put the systems to work.
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.
