Nvidia’s $119 Billion Commitment Backstop Faces Its First Real Test

Nvidia faces 12-month delay on flagship Kyber rack system until 2028, while standard Rubin chip remains on track. Competition from AMD, Google, Huawei looms.

The scale of Nvidia’s production pipeline is staggering: $119 billion in supply commitments, an operating cash flow that topped $50 billion last quarter, and a buyback authorization running into the hundreds of billions. Yet for all that financial muscle, the company’s product roadmap has hit an unexpected snag that is forcing investors to separate normal technical hurdles from genuine strategic risk.

A CNBC report has revealed that the Kyber NVL144 rack system, the ultra-high-end variant of Nvidia’s forthcoming Rubin Ultra architecture, is facing manufacturing difficulties that will push its arrival into 2028 — a delay of more than twelve months. The standard Rubin platform remains on track for mass production in the second half of this year, with partners such as Dell and Lenovo already building compatible servers and cloud giants AWS, Google Cloud, and Microsoft expected to integrate Rubin in the same period. But the absence of the flagship Kyber rack until deep into 2028 opens a window that competitors are already eyeing.

The Rubin chip itself is no slouch. Its GPU packs 336 billion transistors, a dramatic jump from the 208 billion in the current Blackwell generation, promising significant efficiency gains for both training and inference. Nvidia expects the cost of AI inference to drop sharply, and for model training the new architecture will require only a fraction of the compute cores. OpenAI is already planning to use the hardware, and Microsoft recently switched on new AI models powered by Blackwell for Azure customers — a sign that the upgrade cycle is accelerating regardless of the Kyber delay.

That momentum is reflected in the financials. Foxconn reported a 40% quarter-on-quarter revenue surge driven by AI server demand, while Nvidia’s own networking business grew 199% year-over-year, underscoring the stickiness of its ecosystem. The company commands roughly 80% of the AI chip market by revenue, and management has guided for near-doubling of year-ago revenue in the current quarter. Optimists argue that the Blackwell generation alone provides enough firepower to carry the stock, with the average analyst price target sitting at €264 — well above the May record high of €202.50.

Skeptics, however, see the Kyber delay as more than a scheduling hiccup. A two-year gap in the high-end rack lineup gives AMD a clear opening with its MI500X chip, Google with the TPUv8i, and even Huawei with the Atlas 350, all targeting the fast-growing inference market. Additional supply chain headwinds — particularly tight availability and rising prices for specialised memory chips — could squeeze margins even if Nvidia retains pricing power. Meanwhile, OpenAI’s decision to delay its IPO until 2027, reportedly to protect a trillion-dollar internal valuation, has prompted some investors to question whether the broader AI investment cycle is cooling. Nvidia’s implied volatility of around 39% suggests the market remains deeply skittish.

On the charts, the stock is treading water. It currently trades at €178.68, just below the 50-day moving average of €181, and roughly 11% off its peak. The 200-day line near €165 offers a floor, while the 100-day average at about €170 sits as an intermediate support level. A clear break above the 50-day line could reignite the rally toward the €202.50 record, but further supply chain hiccups in next week’s quarterly report risk a retest of the June lows.

The coming earnings release will be the definitive data point. If the cloud hyperscalers confirm they are sticking with Rubin deployments as planned, the delay looks manageable. If they start shifting orders to alternative architectures, Nvidia’s iron grip on the market could loosen for the first time in years.