Technology / AI Chips — At GTC Berlin, Nvidia committed $1.1 billion to chip testing, a 15-fold expansion of optics capacity and 2.5 times more substrate supply by 2028 — while CEO Jensen Huang argued its chips should be treated as a "financeable asset."
Nvidia used GTC Berlin on Thursday, October 22, to announce a supply-chain buildout behind its next Vera Rubin platform: six manufacturing agreements that together commit $1.1 billion to chip testing, target a 15-fold increase in co-packaged optics capacity, double chip-testing throughput and grow advanced substrate supply 2.5 times by 2028. In an interview with CNBC on the sidelines of the conference, CEO Jensen Huang made the strategic argument plainly: "One of the things that we're trying to do is to establish Nvidia's compute as a financeable asset. The residual value of our compute, because it's fungible, it's so productive, it's so durable, it is now becoming an asset class." Huang said demand from neoclouds — the new class of AI cloud providers — is "off the charts," with capital, not demand, as the constraint.
The bottleneck moved — so Nvidia followed it
The logic of the Berlin announcements only makes sense alongside Nvidia's DSX POD: a factory-assembled, containerised version of its Vera Rubin NVL72 rack designed to cut AI-factory deployment from five to six months to as little as 14 days. Compress deployment that far and the constraint stops being on-site integration and becomes upstream manufacturing — how fast chips can be tested, substrates produced and optical engines assembled. Substrate lead times explain the long dates: the Ajinomoto Build-up Film (ABF) substrates under every advanced AI accelerator need two to three years from capital investment to qualified volume, so the three substrate agreements starting in Q2 2027, Q1 2028 and immediately are not really Vera Rubin purchases — they lock capacity for the generation after it as well. The chip-testing agreement itself was signed October 13, 2026 with Taiwan's KYEC; its April 2027 U.S. date is a target, not a contractual guarantee.
Why this matters
Huang's "financeable asset" pitch is the other half of the same strategy, and it answers the question Wall Street asked on the very same day, when AI stocks sold off over capex fears and Meta's free-cash-flow deficit. If Nvidia's GPUs hold residual value like aircraft or power turbines, lenders can finance neoclouds against the hardware itself — turning Nvidia's $500 billion AI financing plan (with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR, announced in August) from a sales programme into an asset class. Huang also said U.S. regulations have accelerated China's domestic chip efforts, and that Nvidia remains open to working with Chinese companies, citing China's manufacturing and supply-chain strengths. Read together, Berlin was Nvidia arguing on three fronts at once: manufacturing capacity, financing structure and geopolitics — while its stock fell 2.4% in a sector-wide selloff it does not control.
Who gains, who is exposed
Korean, Japanese and Taiwanese substrate and testing partners gain multi-year, generation-spanning commitments — Samsung Electro-Mechanics and LG Innotek among them (see the separate substrate financing report). Neoclouds gain a financing story to take to lenders. The exposure sits with anyone who assumed the AI bottleneck would stay at chip design: if testing, substrates and optics are the real constraints into 2028, pricing power migrates to the least glamorous parts of the supply chain — and any demand pause leaves Nvidia holding commitments sized for a boom. Skeptics note that "financeable asset" is also what every hardware cycle claims at its peak; believers answer that GPU rental prices and resale values have, so far, backed the claim with data.
What happens next
Watch the Q2 2027 substrate start date, whether KYEC's U.S. testing capacity lands near its April 2027 target, and whether lenders actually begin pricing GPU-backed loans differently after the Berlin pitch. The Vera Rubin cycle will be judged less on its launch keynote than on whether these factories arrive on time.
Sources
TechTimes, "NVIDIA Commits $1.1 Billion to Chip Testing, Targets 15x Optics Capacity for Vera Rubin," October 22, 2026. CNBC interview with Jensen Huang at GTC Berlin, via Stocktwits market report, October 22, 2026. Global AI Watch, "Nvidia Partners with Korean Firms to Boost AI Chip Production," October 22, 2026.