Command Palette
Search for a command to run...

Nvidia Delays Rubin AI Chip and Drops Revised Chip Design

aiai-infrastructureai-compute-chips 15 posts · 5 accounts

Nvidia's upcoming Rubin AI graphics processing unit faces engineering headwinds and a schedule delay exceeding a year, according to supply chain analysis of the company's roadmap. The chipmaker canceled the revised 2-plus-2 multi-chip module design for its Rubin Ultra variant, leaving only a smaller configuration that delivers roughly half the performance of the originally planned four-chip layout. The shift occurs as Nvidia navigates manufacturability challenges with high-bandwidth memory base dies, interposer redesigns, and thermal management for the next-generation architecture.

The adjustments complicate Nvidia's immediate data-center scaling plans. The company had intended to expand processing power across multiple Rubin Ultra units using an advanced optical interconnect, but the necessary components are deferred until the Feynman chip generation scheduled for later in the decade. An alternative back-to-back rack architecture designed to work around connectivity limits was also scrapped after cloud providers cited operational complexity. While some industry observers view the changes as signs of execution risk from Nvidia's aggressive development targets, others note that performance boundaries routinely shift during advanced AI hardware development.

From the sources (15 posts)

@jukan05

@JulienTechInvst AMD, yes. But I’m not so sure about TPUs.

@scaling01

this surely also pushes Feynman back and with that 100T models

@jukan05

@StreetSignal__ I agree. It also makes me wonder whether AMD might struggle to meet its own specs as well.

@jukan05

1. Bullish for the NPO supply chain. 2. The spec downgrade of NVIDIA Rubin Ultra signals the erosion of NVIDIA’s performance moat. 3. Bearish for the CPO supply chain. 4. Big winners: AMD and the TPU ecosystem?

@semianalysis_

As the NVIDIA roadmap indicates, CPO NVSwitch will not be available until Feynman. As a result, NVIDIA currently has no proven solution to expand the scale-up world size for Rubin Ultra, leaving a gap for competitors like AMD MI500X or TPUv

@semianalysis_

This news also comes as the 4-compute-die Rubin Ultra has been cancelled, leaving only the smaller 2-compute-die Rubin Ultra, which will deliver roughly half the real-world performance of the 4-die Rubin Ultra. 5/6🧵

@semianalysis_

NVIDIA will sell significantly more Oberon Rubin racks and Oberon Rubin “Ultra” racks to make up for this shortfall. We discuss the implications of these mass NVIDIA delays and cancellations for the memory, PCB, and ODM supply chains in ou

@semianalysis_

MASSIVE DELAY: Just 3 months after Jensen demoed Kyber NVL144 at GTC, it has faced major setbacks and has been delayed by more than 12 months, pushing it back to 2028. Below, we explain why Kyber has faced massive delays and why NVIDIA’s NV

@semianalysis_

Kyber NVL144 rack architecture has been delayed to 2028 as the PCB midplane remains challenging from a manufacturability standpoint. NVL576, which connects 8x Oberon racks over CPO between the NVSwitches, is also likely delayed or restricte

@semianalysis_

NVL72x2 back-to-back rack architecture was the new proposed architecture NVIDIA was developing as an alternative to Kyber. It was designed to increase the pure-copper NVLink scale-up world size by placing two Oberon racks back-to-back. Howe

@aaronwei3n

RT @Aaronwei3n: Yes, $NVDIA Rubin faces multiple headwinds. 1. HBM4 base die 2. Interposer Redesign 3. Heat Dissipation Redesign

@zephyr_z9

Having observed the Hopper & Blackwell and now the Rubin ramp over the last 4 years, plz don't expect Nvidia to meet their deadlines Jensen operates on very aggressive timelines and puts too much pressure on the supply chain AI servers are

@aaronwei3n

So true. It’s wild how the market panics over minor timeline shifts, ignoring the engineering realities. When $NVDA is inventing the state of the art and constantly redrawing the boundaries, execution hurdles are just part of the proces

@zephyr_z9

Just look at Rubin & Rubin Ultra GPU Now, even the 2+2 MCM design got canceled

@aaronwei3n

@Jespabe nope, its die has some issues too.

Preview built on a synthetic news corpus (16 weeks, Apr–Jul 2026). Impact calls are model reads, not price data.

About Archive