Jukan
Semiconductor & AI Infrastructure Research at Citrini
G2AIStack's reported component yield and capacity bottlenecks make NVIDIA's Rubin shipment outlook look dire. It says Rubin system expectations were cut from roughly 10,000 to around 5,000, citing GPUs, orthogonal backplanes and power systems; its ODM checks indicate shipments are falling short. x.com
The end of CUDA's moat is approaching, which is very bearish for Nvidia. Per DeepSeek's Liang Wenfeng in a leaked investor call, its own compiler and TileLang have sharply cut its reliance on Nvidia software, and porting them to Huawei chips could largely fix Chinese chips' ecosystem gap within a year, leaving production as the bottleneck. Chinese open-source models now run best on Nvidia GPUs, protecting its inference moat; Chinese compute independence would erode that. x.com x.com
SemiAnalysis
Independent Semiconductor and AI Research Firm
NVIDIA has revised Vera Rubin NVL72’s NVLink bandwidth from 3.6 TB/s to 3.0 TB/s per Rubin GPU package, a ~17% reduction. x.com
AlvaSupportsSemiAnalysis reports a 17% reduction in Rubin NVLink bandwidth and identifies signal-integrity challenges at the original speed, adding evidence that Rubin interconnect engineering constraints are affecting its specifications.Jukan
Semiconductor & AI Infrastructure Research at Citrini
SAMSUNG ELECTRONICS SAID, “WE EXPECT THESE TIGHT MEMORY SUPPLY CONDITIONS TO PERSIST INTO 2028 AS WELL.” x.com
AlvaSupportsExpected memory supply tightness into 2028 strengthens the risk that memory availability will constrain Rubin production throughout its ramp.
Satya Nadella
Chairman and CEO of Microsoft
We continue to scale our fleet with the latest innovations from NVIDIA, AMD alongside our own Maia systems. We have docked our first NVIDIA Vera Rubin GPUs and are very excited about it.
AlvaChallengesMicrosoft's receipt of its first Vera Rubin GPUs provides direct customer evidence that Rubin deliveries are progressing despite the reported component bottlenecks.
Vertiv Holdings Co
Critical digital infrastructure provider for data centers, communication networks, and commercial and industrial environments
We are making significant capacity investments globally to be well-positioned to serve our customers and execute on a robust backlog. ... These capacity additions are already contributing to our Q3 and Q4 growth. And it's not just our own capacity. We keep on proactively securing and expanding long term capacity agreements with key suppliers to ensure they scale in lockstep with our growth.
AlvaChallengesVertiv's capacity additions are already supporting growth, alongside expanded supplier capacity agreements, offering a path to easing the power-system supply constraints cited in the Rubin shipment thesis.
SK hynix Inc.
Memory Semiconductor Manufacturer
HBM4 is expected to cross over HBM3E in the fourth quarter in both bit shipment and revenue share thanks to mass production proceeding without disruption. ... The company is maintaining industry best level yield and both our fabs and back end processes in particular are ebac and yield refers to the entire process, not just individual steps like bump, stock or GSD test. On this basis, HBM4 already exceeded its planned oelde, so the target was revised up.
AlvaChallengesSK hynix says HBM4 mass production is proceeding without disruption and expects its shipments to overtake HBM3E in Q4, weakening the case that HBM4 supply will hold back Rubin output.
Reuters
Global News Agency
NVIDIA CORPORATION - VERA RUBIN NVL72 AND VERA CPUS NOW GENERALLY AVAILABLE ON OCI tradingview.com
AlvaChallengesNVIDIA and Oracle say Vera Rubin NVL72 is generally available on OCI, evidence that completed Rubin systems have reached a major cloud customer despite the reported shipment bottlenecks.
Fireside Alpha
Business, Technology and Investing Interview Digest
[Replying to Mims: "“Do you see that bottleneck easing at all?”"] “I think it’s going to get worse in 2027 based on what I see right now.” ... And so the demand is growing faster than supply. It takes four years to build these factories. So I don’t think that the shortage is going to quickly resolve itself.” x.com
AlvaSupportsDell expects the memory shortage to worsen in 2027 as AI demand outpaces supply and new factories remain years away, adding pressure to the memory supply needed for Rubin systems.
Jeff Pu
Technology Equity Analyst at GF Securities
Vera Rubin ramping with 85%+ first-pass yield; further acceleration expected in 1Q27 • 2026E: 67k GB200/GB300 + 6k VR200 • 2027E: VR200 + initial VR300 (Ultra) drive most of the growth (86k units) x.com
AlvaChallengesGF Securities estimates Rubin's first-pass yield above 85% and forecasts 6,000 VR200 racks in 2026 and 86,000 Rubin and Rubin Ultra racks in 2027, challenging the reported yield bottlenecks and dire shipment outlook.Jeff Pu
Technology Equity Analyst at GF Securities
Supply shortage expected to persist into 2029 (18%/16% shortage in 27/28) x.com
AlvaSupportsGF Securities forecasts ABF substrate supply shortfalls of 18% in 2027 and 16% in 2028, strengthening the case that GPU component capacity could constrain Rubin system output.
Amkor Technology, Inc.
Semiconductor Packaging and Test Services Provider
I think that has highlighted even more that packaging is becoming one of the key bottlenecks. So customers are really working to make sure they secure capacity into 2027 and even even into 2028.
AlvaSupportsAmkor's account of packaging as a bottleneck and customers reserving capacity into 2027 and 2028 strengthens the case that GPU supply capacity could constrain Rubin system output.Reuters
Global News Agency
LEASE DEAL WITH APOLLO TO DELIVER NEXT GENERATION OF NVIDIA VERA RUBIN CHIPS FOR SPACEX’S TERRESTRIAL COMPUTE tradingview.com
AlvaChallengesReuters reports an Apollo-led lease financing deal to deliver NVIDIA Vera Rubin chips for SpaceX's terrestrial compute buildout, adding a concrete customer deployment commitment that pushes against a dire Rubin shipment outlook.Jukan
Semiconductor & AI Infrastructure Research at Citrini
Korean media report that Samsung Electronics and SK hynix plan to begin mass production and volume shipments of 8-high HBM4 in Q2 next year. ... While the reporting outlet identifies Rubin Ultra as the target platform, we believe Vera Rubin will also adopt 8-high HBM4. x.com
AlvaMixedReported Q2 2027 HBM4 volume shipments set a timetable for added memory supply, while Jukan's expectation that Vera Rubin will use 8-high HBM4 conflicts with Daishin's report that the option was dropped.Jukan
Semiconductor & AI Infrastructure Research at Citrini
Correction: It’s 9.6, not 9.3. ... NVIDIA has officially lowered the HBM4 pin speed. ... 11.7-> 9.3Gbps x.com
AlvaSupportsJukan corrects the reported revised HBM4 pin speed to 9.6 Gbps, rather than the 9.3 Gbps previously shown; it remains below the earlier 11.7 Gbps specification.Jukan
Semiconductor & AI Infrastructure Research at Citrini
NVIDIA has officially lowered the HBM4 pin speed. 11.7-> 9.3Gbps x.com
AlvaSupportsJukan reports NVIDIA lowered its HBM4 pin-speed specification from 11.7 to 9.3 Gbps, a concrete relaxation of Rubin's memory requirements consistent with component qualification pressure.Jukan
Semiconductor & AI Infrastructure Research at Citrini
NVIDIA: Rubin’s specifications have changed amid base die supply constraints and requests from customers prioritizing bandwidth over cost efficiency. The original plan called for a mix of 8-high and 12-high HBM, but the 8-high option appears to have been dropped. x.com
AlvaSupportsDaishin Securities reports that HBM base-die constraints have already changed Rubin's memory configuration, strengthening the case that component supply is affecting its production plan.Fireside Alpha
Business, Technology and Investing Interview Digest
“Yes, I think on the semiconductor side, the biggest constraint, of course, besides memory, is substrates, PCB materials, substrate materials, all of these components were severely under-invested in. ... "And so we’ve seen, you know, tremendous bottlenecks there in terms of what can be made over the next two years, you know, both for accelerators as well as CPUs and every other component in the supply chain. x.com
AlvaSupportsPatel's assessment that substrate materials are underinvested and constrain accelerator output for the next two years strengthens the case that component supply could limit Rubin production.Jukan
Semiconductor & AI Infrastructure Research at Citrini
According to Korean media reports, Samsung Electro-Mechanics has received advance payments of $830 million from Tesla and $2 billion from NVIDIA to expand its ABF substrate production capacity. x.com
AlvaChallengesThe reported $2 billion NVIDIA advance to Samsung Electro-Mechanics for ABF substrate expansion points to added component supply capacity that could ease Rubin production constraints.
Dylan Bristot
Growth at Nebius
- KYEC, chip test: a $1.1B capacity reservation advance from nvidia for new US test sites, high volume in April 2027 ... - Nvidia's total test capacity: doubles from 2026 to 2027 ... my read: the names with a dated ramp are packaging, test and substrates, the unglamorous middle of the stack. the earliest date on the list is february 2027, so orders and prepayments move about a year before the capacity shows up x.com
AlvaChallengesNvidia's reported test-capacity expansion offers a potential path to easing one supply constraint, with new capacity arriving in 2027 rather than immediately.Fireside Alpha
Business, Technology and Investing Interview Digest
Jensen Huang holds up a Vera Rubin node on stage: what took two hours to assemble on Grace Blackwell now takes under five minutes, with no cables and 100% liquid cooling “So Vera Rubin is in full production. ... This is the first computer in the world that we’ve created. No cables, no wires, completely 100% liquid cooled. ... And this is one Vera Rubin computer. Okay, so this has two GPUs." x.com
AlvaChallengesJensen Huang's claim that Vera Rubin is in full production pushes against the dire shipment outlook, while the faster node assembly suggests one manufacturing step is becoming easier.
Elon Musk
CEO of Tesla, Founder & CEO of SpaceX
We're tracking to be able to bring roughly 200,000 GB300s online per month for every month through the remainder of this year and continuing that pace next year, with next year being primarily Vera Rubin compute.
AlvaChallengesThe plan to sustain the current monthly compute deployment pace into a primarily Vera Rubin buildout next year challenges a dire shipment outlook, though it is a target rather than evidence of Rubin shipments already delivered.