Nvidia (NVDA) investors were more or less expecting another clean step forward in the AI upgrade cycle.Rubin was expected to continue pushing the roadmap ahead, customers would keep chasing capacity, and rivals would remain stuck trying to close an insurmountable lead.That confidence, though, has a new wrinkle.A reported delay linked to Nvidia’s next high-end rack system underscores that the tech behemoth’s biggest challenge might not be demand, but the physical complexity of scaling AI hardware. Moreover, the news comes at a point when Nvidia’s stock has been losing momentum, according to Seeking Alpha data, with shares down 0.46% over the past week and 12.56% over the past month. That compares with the S&P 500’s 1.70% gain over the past week and 1.67% decline over the past month, though Nvidia remains slightly positive over 6 months and year to date.Despite Nvidia still dominating the AI space, the next phase of that dominance might depend on parts of the system investors typically overlook.For its rivals like AMD (AMD) and Google (GOOGL) do not need Nvidia to stumble badly. They only need time.
Nvidia’s reported AI rack delay could give AMD and Google an openingAnnabelle Chih/Bloomberg via Getty Images
Nvidia’s reported delay is bigger than a product slip It seems Nvidia’s next big AI roadmap test is no longer just about how fast its chips can run.More Nvidia:Nvidia’s workplace culture sends Big Tech a warningNvidia’s $25B bond deal sends investors a clear signalBank of America resets Nvidia stock forecast after CFO meetingAccording to a report from TheFly, citing SemiAnalysis, Nvidia’s Kyber NVL144 rack architecture has reportedly been delayed by more than 12 months, pushing the system toward 2028. The setback comes only months after CEO Jensen Huang showcased Kyber at GTC as part of Nvidia’s next wave of AI infrastructure.What complicates things is that the point I mentioned earlier comes at a time when the AI trade is being heavily questioned by some of the market’s finest, with Nvidia stock under pressure as chip stocks sell off. Kyber is not a stand-alone GPU launch.It is designed around Nvidia’s Rubin Ultraroadmap and is meant to scale AI compute at the rack level, where hundreds of chips, switches, and interconnects behave like one larger system.So the delay is a lot more significant than a typical product slip-up.Nvidia has already moved beyond selling individual accelerators. The bigger ambition for it is to sell full AI factory systems to cloud giants and hyperscalers, racing to train and run larger models. A delay in the rack architecture takes a cut at the company’s next major growth engine, which is why its rivals may finally see an opening.Wall Street price targets for Nvidia stockBaird: $500. Baird raised Nvidia’s target to a Street-high $500 from $300 due to AI infrastructure growth, inference share gains, and faster Vera Rubin adoption.Bank of America: $350. BofA reiterated a Buy rating and $350 target, arguing Nvidia is becoming broader than GPUs as CPUs, AI systems, networking, and software lift its content per AI factory.Cantor Fitzgerald: $350. Cantor raised its Nvidia target to $350 from $300, keeping an Overweight rating as it pointed to strong AI demand, data-centre spending, and Blackwell backlog visibility.Morgan Stanley: $288. Morgan Stanley reiterated an Overweight rating and $288 target, keeping Nvidia as a top pick in processors and calling it one of the best value names in the group.JPMorgan: $280. JPMorgan raised its Nvidia target to $280 from $265, with analyst Harlan Sur keeping an Overweight rating after Nvidia’s strong Q1 FY2027 earnings.
Sources: Investing.com; TheStreet; GuruFocus/TradingView.
AMD and Google get a rare opening Nvidia’s reported Kyber setback gives its rivals time to work things out in the AI hardware race.According to the SemiAnalysis note, Nvidia currently lacks a proven solution to scale the Oberon Rubin Ultra domain, creating a potential opening for AMD’s MI500X and Google’s TPUv8i broadly to challenge Rubin Ultra on scalability.For perspective, AMD has been trying to turn its Instinct roadmap into a more credible alternative for hyperscalers that do not want to depend entirely on Nvidia. A delay at Nvidia’s end gives it more room to pitch customers on performance, supply availability and total cost.On the flipside, Google’s opportunity is different.Its TPU platform is already deeply tied to its own AI infrastructure, and Google Cloud can use that in-house design in competing for customers looking beyond GPUs. Additionally, if Nvidia’s system-level roadmap slows up, Google gets more time to prove its custom silicon can scale efficiently for larger AI workloads.However, this isn’t a clean handoff at this point.Nvidia remains the default AI hardware supplier for a reason. Its CUDA software stack, developer ecosystem, cloud relationships and installed base are tough to displace. Customers will test alternatives, but moving serious AI workloads away from Nvidia isn’t simple.For perspective, according to a Silicon Analysts report, Nvidia chips held a tremendous 80% of the AI chip market, while CEO Jensen Huang framed its AI chip opportunity at at least $1 trillion through 2027. Nvidia’s rack problem starts inside the system Nvidia’s reported bottleneck isn’t in the Rubin Ultra GPU itself but in the board that helps the whole rack work.The SemiAnalysis note says the issue is linked to the midplane PCB, which Nvidia calls the orthogonal backplane. Put simply, the board allows compute trays and switch trays to connect vertically at 90-degree angles inside the rack, which reduces the need for thousands of traditional cables.That matters when Nvidia is looking to make 144 GPUs operate as one connected AI system.SemiAnalysis said using conventional copper cabling for Rubin Ultra NVL144 might need over 20,000 cables, add over 30% to rack weight, and create serious signal integrity problems.Without this part, scaling AI compute becomes heavier, messier and harder to deploy.Related: ‘Big Short’ investor Michael Burry issues blunt 4-word warning on AI stocks