Nvidia's Substrate Bet, Hugging Face Review and an Unverified China Report: Three Threads Investors Must Not Tangle

Nvidia's growth story is no longer just about how many chips it can sell. It is also about whether it can secure the materials to build them, win approval for a $12.93 billion acquisition, and manage a cloud partner network whose China exposure has become the subject of conflicting reports.

The company and its supply-chain allies laid out plans on 21 October to expand substrate capacity to 2.5 times current levels within three years. IBIDEN, Samsung Electro-Mechanics and Unimicron are involved in the effort. That is a production commitment, not a completed build-out — a distinction that matters as much as the headline number.

Downstream, Amkor intends to grow its advanced chip packaging capacity by more than 300% by the end of 2028. KYEC is planning new facilities in the United States, and Nvidia is backing that move with a $1.1 billion prepayment to reserve capacity.

A Deal Under the Microscope

Separately, the US Department of Justice has sought information on the competitive implications of Nvidia's planned purchase of Hugging Face, Reuters reported on 28 October. The price tag is $12.93 billion, with completion expected in the first half of 2027. The inquiry is not a finding of wrongdoing, and it does not mean the transaction has closed — nor does it amount to a decision against the deal. What it does signal is that regulators are examining whether Nvidia would gain control of a key distribution channel for artificial intelligence.

The acquisition sits on a different plane from the capacity announcements. Partnerships demonstrate operational intent; they say nothing about the regulatory status of the Hugging Face purchase. Investors who conflate the two risk misreading both.

Infrastructure Alliances Expand

The growth agenda stretches well beyond that one deal. Roughly a week before the substrate news, Nvidia and Oracle announced plans to scale Oracle Cloud Infrastructure superclusters to multiple gigawatts, using Nvidia's Vera Rubin NVL72 systems. The systems and Vera processors were also declared generally available on OCI.

Meta is deepening its ties as well. The two companies announced on 27 October an expanded collaboration around Meta's Muse model family, and Meta plans to deploy Nvidia infrastructure on a larger scale. These are partnership and infrastructure commitments — not updates on the Hugging Face review.

The China Question Remains Unresolved

Cloud partners may face restrictions of their own. The Financial Times reported that Nvidia has instructed customers to sell at least 51% of their services to non-Chinese companies. Reuters said it could not immediately verify the report, so the claimed threshold should not be treated as confirmed business policy. An earlier FT report — that Nvidia had told data-centre partners to scale back their engagement with Chinese customers — met the same fate: Reuters could not verify it either, and no confirmed restriction can be inferred.

17 November Is the Date That Counts

The next confirmed financial marker is Nvidia's earnings call on 17 November 2026, announced on 28 October for the third quarter of fiscal 2027. The reporting period ended on 25 October.

Until then, the three threads should be kept apart: a planned acquisition still under review, an announced infrastructure build-out, and reported customer directives that remain unverified. None of them substitutes for the quarterly figures — and the figures will not settle the regulatory or supply questions either.