Did Nvidia's 2028 Rack Delay Under Jensen Huang Open a Door for AMD and Google?

SemiAnalysis warns Nvidia’s next-gen Kyber NVL144 AI rack could slip to 2028 due to manufacturing issues with a key circuit board, despite Nvidia stating its roadmap is intact. The potential delay could give AMD and Alphabet more room to win high-end AI infrastructure customers, especially for inference/training-optimized systems. Net impact appears more incremental than catastrophic, but it meaningfully raises execution risk around Nvidia’s server-rack rollout and timelines.

Analysis

The investable issue is not near-term GPU demand; it is whether NVDA can still command a premium multiple if the market starts believing its system-level roadmap is vulnerable to manufacturing friction. A slip in a flagship rack platform mostly matters through confidence, because hyperscalers will still buy capacity, but they may diversify incremental allocations toward suppliers with more predictable delivery and simpler qualification paths.

AMD is the cleanest relative beneficiary because it only needs to win marginal inference and agentic AI share to improve its data-center mix; the software gap matters less where customers are optimizing for cost and memory density rather than absolute training leadership. GOOG/GOOGL is more of a “safe alternative” than a direct share-stealer: if customers want vertically integrated silicon with lower dependence on a single vendor, TPU economics become more attractive, and that helps cloud margin leverage even if external chip revenue is not the main story.

The contrarian take is that the market may be overpricing the delay headline relative to actual cash-flow impact. If this is a board/packaging bottleneck rather than a core design miss, NVDA can likely bridge it without losing much revenue, but the stock can still de-rate 10-15% if investors start discounting roadmap credibility. The key falsifier is any confirmation from customers or supply-chain checks that orders are being re-routed; absent that, the next 1-3 months are about sentiment, not fundamentals, while 6-18 months only matter if this becomes a pattern.

Key Decisions for Investors

• Relative-value trade: long AMD / short NVDA on a 3-6 month horizon, sized modestly. Thesis is not outright NVDA demand loss, but a higher probability that marginal high-end share and inference workloads diversify away from the incumbent. Falsifier: NVDA customer commentary or next quarter guide showing no adoption friction.
• Buy GOOG/GOOGL on any post-headline weakness as a smaller, lower-beta beneficiary of supply-chain certainty. The trade works if hyperscalers continue valuing vertically integrated TPU capacity; risk/reward is better as a defensive long than as a direct share-capture bet.
• Do not press a naked short NVDA on the headline alone. Wait for corroboration from OEM/channel checks or a second data point on delayed rack shipments; if none appears within 2-4 weeks, the setup is more likely to mean-revert than to trend.
• Set an alert on NVDA if the stock breaks the prior multi-week support after confirmation of the delay. That would signal multiple compression risk from roadmap credibility, not just transient newsflow.