Nvidia refutes: Counterattack on the roadmap, the market arbitrates between noise and signal

The "intact" Kyber rack according to Nvidia after a report pointing to a 2028 delay: the tempo of AI compute becomes a bet on the conductor's word.

The fact

On July 6, 2026, Nvidia publicly denied reports that its next-generation AI rack "Kyber" was facing production delays, asserting that the roadmap remains intact (Yahoo Finance). The initial report had pointed to a possible slip to 2028 due to manufacturing issues (Reuters, 03/07). In parallel, several Wall Street analysts mentioned that AMD "could close the gap faster than expected" (Barron's, 06/07).

Our analysis

Three readings intersect. First, the mechanics of the chain: between TSMC N3P/N2 wafers, CoWoS-L packaging, and HBM4, a six-month delay is plausible. Second, the mechanics of communication: Nvidia systematically denies delay rumors to preserve order books and avoid purchase delays. Finally, the competitive mechanics: AMD MI400 arrives at the end of 2026, and the smaller gap between the two CUDA/ROCm stacks reshuffles the deck for hyperscalers. The message for the investor: the catalysts are shifting from Nvidia alone to the ecosystem (AMD, Broadcom ASIC, Astera Labs on interconnect).

To watch

Nvidia SIGGRAPH day (late July); CoWoS TSMC guidance; HBM4 deliveries from Micron / SK Hynix; CoreWeave and Applied Digital feedback in Q3 on the effective availability of Blackwell Ultra.