NVIDIA's GB200 AI Servers Face Overheating Rumors and Order Reductions Amid Conflicting Reports

Taiwan suppliers for NVIDIA GB200 AI servers, components: rumors of overheating GB200 are wrong

TL;DR: Rumors of NVIDIA GB200 AI servers overheating have been denied by Taiwan suppliers, who assert shipments are unaffected. Major cloud providers are reportedly reducing orders, seeking newer versions. The complexity of packaging with high-power chips is noted as a challenge. The transition to liquid-cooling is causing adjustments for some developers and suppliers. In a report from outlet The Information a couple of days ago, rumors of NVIDIA GB200 AI servers having overheating issues heated up again, but now Taiwan suppliers have come to the table saying these rumors are false, and asking "how many times is this rumor going to get repeated?" The 4 major cloud service providers (CSPs) with Amazon AWS, Microsoft, Google, and Meta (as well as others) are reportedly cutting back on NVIDIA GB200 AI server cabinets, wanting the newer versions according to reports. The NVIDIA GB200 supply chain in Taiwan "generally expressed helplessness" yesterday, in response to the rumors of GB200 overheating. They said "how many times will the same rumor happen" while emphasizing that shipments of NVIDIA GB200 AI servers are on schedule, and that the shipments are not affected by overheating issues. The issues from the GB200 AI server cabinets are coming from the complexity of the packaging, as there are far more higher-end, power-consuming chips inside GB200 NVL72 server cabinets compared to the previous generation. NVIDIA's new GB200 AI servers are liquid-cooled, and because of the shift from air-cooling to liquid-cooling, it's seeing AI developers and data center suppliers that haven't used water-cooling before, into a spin.

Taiwanese Server Manufacturers Deny Overheating Issues with NVIDIA's GB200 AI Servers, Says Production Is On Schedule

Taiwanese server manufacturers have claimed that NVIDIA's GB200 AI servers have no "overheating" issues and that production is on the right track. Rumors surrounding NVIDIA's GB200 AI servers witnessing design flaws have been circulating for quite some time now. They initially began back in Q4 2024, when NVIDIA was said to limit the initial shipments due to multiple server issues. Team Green claimed that the problems were resolved, but in recent coverage, we reported how the GB200 flaws have re-emerged, resulting in mainstream tech companies cutting down on orders. However, a report by Taiwan Economic Daily, citing Taiwanese server manufacturers, claims that these rumors are purely speculative, saying, "How many times will the same rumor happen?". They claim that production hasn't seen any influence yet and that the schedule is proceeding as planned, which does refute the general perception being built up against NVIDIA's GB200 AI servers. Previously, back in November 2024, it was revealed that Team Green's Blackwell servers were facing a design flaw, with the actual problem lying in TSMC's chip interconnect. Customers such as Microsoft, Amazon, and OpenAI are now cutting down on GB200 orders, which shows that the supply chain is generally facing an issue. Microsoft was previously said to deploy over 50,000 GB200 AI cabinets, but this figure is now tremendously reduced. Moreover, companies are now opting for NVIDIA's older Hopper generation solutions as well, so overall, the reception of Blackwell in the markets is certainly not looking too well, and NVIDIA might face difficulty in selling these cabinets to the industry, given that it fails to solve the persistent issues. Whether GB200 server flaws will translate into lower revenue for NVIDIA is a question for the future. Still, given how massively AI servers have scaled up in a matter of a few years, it is inevitable that the overall supply chain will face issues. However, seeing how pivotal it is for Team Green to maintain AI dominance, we are anticipating that the GB200 AI server flaws will be resolved quickly, given the economic and technological resources the firm holds right now.

NVIDIA Blackwell AI servers: overheating, architecture flaws see companies cutting orders down

NVIDIA's new Blackwell AI servers reportedly ran into issues last year with overheating and an architectural flaw, and it seems these issues haven't gone away, leaving big customers (paying big bucks) stranded, moving back to Hopper AI servers. In a new report from The Information, we're learning that the first significant shipments of NVIDIA's new GB200 AI servers have big customers experiencing overheating, and glitching issues, with the big problem being the "way chips connect". Big customers like Amazon, Google, Meta, and Microsoft have reportedly cut down their orders because of the issues. Back in October 2024, NVIDIA CEO Jensen Huang said "we had a design flaw in Blackwell" noting that it was "100% NVIDIA's fault" and not anything to do with the rumored issues with TSMC's new CoWoS advanced packaging. A few months later in December 2024, we reported that NVIDIA GB200 AI server mass production and its peak shipments could be delayed until Q2 or even Q3 2025... and here we are with more issues. It seems that cloud service providers (CSPs) are now delaying the move to Blackwell-based GB200 AI servers, and back to the solid Hopper AI GPU servers... I'm sure this story will continue to build, as more comments (hopefully from NVIDIA soon) pile on.

NVIDIA's Blackwell AI Servers Faced With Overheating & Glitching Issues; Major Customers, Including Microsoft & Google, Start Cutting Down Orders

NVIDIA's Blackwell AI servers are reportedly facing a supply chain bottleneck as Team Green fails to resolve the overheating and architectural flaws. NVIDIA Has Started To Delay Blackwell AI Server Orders Leading To Customers Preferring Older "Hopper" Generation Well, this certainly isn't the start NVIDIA would've hoped for with its Blackwell AI lineup, but it looks like Team Green is certainly being faced with a massive barrier. For those unaware, NVIDIA's Blackwell AI servers were initially expected to see volume production back in Q4 2024, but it was reported back then that the company's new AI architecture is faced with a design flaw, ultimately bringing in higher thermals. Despite Team Green claiming to have sorted the issue out, a new report by The Information has refuted this, saying that the Blackwell AI servers are "glitching" out. According to the report, it is claimed that the first significant shipment of NVIDIA's GB200 AI servers witnessed overheating and glitching issues, with the key problem lying in the "way chips connect." This problem has ultimately bothered mainstream customers like Microsoft, Amazon, Google and Meta, which is why the report claims that companies have cut-down Blackwell orders, and these firms are said to have placed orders worth over $10 billion. The situation is indeed alarming for NVIDIA and its AI business, given that supply chain issues in such products can be devastating for the firm's finances. While we are still unaware of the exact problem, it was claimed previously that the issue lies with TSMC's advanced packaging technology, i.e., CoWoS, which refers to the "chip connection" issue we mentioned above. NVIDIA did say previously that they had changed the Blackwell GPU mask made at TSMC, but this hasn't resolved the issue. Companies are now switching to NVIDIA's well-established alternatives, such as those from the Hopper generation, until Team Green sorts out Blackwell's problems. For now, we are unaware of how big of an impact the Blackwell design flaw will make on NVIDIA's revenue figures, but given that the company is unable to sort the issues out, Blackwell's success may be at stake here, which will prove to be troublesome for NVIDIA.

Nvidia's biggest customers delaying orders of latest AI racks, the Information reports

(Reuters) - Nvidia's top customers are delaying orders of the AI chip leader's latest 'Blackwell' racks due to overheating issues, the Information reported on Monday. The Santa Clara, California-based company's shares fell more than 4% in early trading. The U.S. government also said earlier in the day it would further restrict AI chip and technology exports, potentially hurting Nvidia's sales. The first shipments of racks with Blackwell chips have been overheating and exhibiting glitches in the way chips connect to one another, the Information reported. A rack, used in data centers, is a structure that houses chips, cables and other essential equipment. Major customers Microsoft, Amazon.com's cloud unit, Alphabet's Google, and Meta Platforms have cut some orders of Nvidia's Blackwell GB200 racks, according to the report. Nvidia, Microsoft, Google, Meta and Amazon did not immediately respond to Reuters' requests for comment. The so-called hyperscalers had each placed Blackwell rack orders worth $10 billion or more, the report said. Some of the customers are waiting to snag a later version of the racks or plan to purchase the company's older AI chips, according to the report. Microsoft was initially planning to install GB200 racks with at least 50,000 Blackwell chips in one of its Phoenix facilities, the report added. However, key partner OpenAI asked Microsoft to provide it with an older generation of Nvidia's 'Hopper' chips as delays popped up, the report said. It is unclear how the order cuts would impact Nvidia's sales as there may be other buyers for the "glitchy" GB200 server racks, the report said. The company is on track to exceed an earlier target of recording several billion dollars in revenue from Blackwell chips in its fourth fiscal quarter, CEO Jensen Huang said in November. Huang had also denied earlier media reports of a flagship liquid-cooled server containing 72 of the new chips experiencing overheating issues during initial testing.

Nvidia's customers are reportedly facing delays, shares extend decline By Investing.com

Investing.com -- Nvidia's stock dropped over 4% on Monday following a report from The Information about delays experienced by some of its major customers in deploying its advanced Blackwell AI chips in data centers. The delays stem from issues with the initial shipments of Nvidia (NASDAQ:NVDA)'s Blackwell-equipped racks, according to The Information. The publication explains that these racks have faced overheating problems and glitches in chip connectivity. The technical setbacks are said to have affected major customers, including Microsoft (NASDAQ:MSFT), Amazon (NASDAQ:AMZN) Web Services, Google (NASDAQ:GOOGL), and Meta Platforms (NASDAQ:META), prompting them to cut back on some orders of the Blackwell GB200 racks. In particular, Microsoft, which had planned to install a substantial number of these racks in its Phoenix data center, reportedly shifted to using an older generation of Nvidia chips, the H200, due to the delays. The Information noted that this adjustment significantly reduced the number of Blackwell chips deployed in the facility. Despite the reported issues, Nvidia remains optimistic about the potential of its Blackwell chips. The company had projected significant revenue from the Blackwell series, expecting several billion dollars in revenue in the January quarter alone. The chips are touted as being four times more energy efficient than their predecessor, Hopper, which is crucial for data centers with fixed energy capacities. However, the delays are said to be causing strain not only on Nvidia but also on its cloud provider customers and AI developers who rely heavily on Nvidia's chips for their supercomputing clusters. The Information reported that if Nvidia and its suppliers can resolve the issues, customers may reconsider and increase their orders.