AI chip dominance$NVIDIA (NVDA.US)$The two new cutting-edge AI chips have had a difficult start due to design flaws, leading to serious engineering challenges during their manufacturing process. These AI chips based on Blackwell architecture breakthroughs are crucial for Nvidia to further expand its leading position in the artificial intelligence computing infrastructure market. According to insiders, this delay has affected Nvidia's announcement in March of the highly anticipated Blackwell architecture chip lineup of B100/B200/GB200, with B100 possibly being eliminated due to design flaws and chip engineering issues, and replaced by the slightly lower-performing B200A. Although the delay in Blackwell has caused Nvidia's stock price to plummet, banks like Bank of America view this as an excellent opportunity for bottom layout.

Insiders revealed that NVIDIA is redesigning a version of AI accelerator AI chips, called to be better compatible with H100, a data center infrastructure based on the previous generation AI chip architecture Hopper. However, insiders pointed out that the data center hardware field targeted by B100 is only a relatively niche sub-market, and the real large-scale market targets are B200 AI GPU and GB200 targeted by NVIDIA to carry its self-developed Grace CPU on two B200 chips, respectively.

In addition, insiders also pointed out that due to some problems in chip manufacturing and CoWoS packaging technology support, the GB200 product that combines Grace CPU with NVIDIA AI GPU will not achieve large-scale listing as expected by the market.

These chip manufacturing challenges can be regarded as reflecting the challenges NVIDIA faces in accelerating innovation. NVIDIA's CEO, Huang Renxun, is pushing new chip design architectures and new chip connection technologies faster, aiming to maintain NVIDIA's almost monopolistic advantages in the core infrastructure of artificial intelligence. The company dominates the entire AI chip market, accounting for more than 90% market share, and AI chips are the most critical hardware behind the generation of generative AI tools such as ChatGPT. NVIDIA's performance and total market value have soared rapidly in the past two years, and its market value has once surpassed Apple and Microsoft, ranking as the "highest market value listed company in the world".

These news revealed by insiders were first reported by Information website, which can be regarded as reflecting the challenges NVIDIA faces in accelerating innovation. NVIDIA's CEO Huang Renxun is pushing new chip design architectures and new chip interconnection technologies faster, aiming to maintain NVIDIA's almost monopolistic advantages in the core infrastructure of artificial intelligence. The company dominates the entire AI chip market, accounting for more than 90% market share, and AI chips are the most critical hardware behind the generation of generative AI tools such as ChatGPT. NVIDIA's performance and total market value have soared rapidly in the past two years, and its market value has once surpassed Apple and Microsoft, ranking as the "highest market value listed company in the world".

NVIDIA has been deeply cultivating the global high-performance computing field for many years, especially its CUDA computing platform created from scratch has become popular all over the world, and it is the software and hardware co-simulation system preferred in the field of high-performance computing such as AI training/inference. This is also the core force driving NVIDIA's AI chip market share to be as high as 90%. CUDA computing platform is a parallel computing acceleration platform and programming assistance software developed exclusively by NVIDIA, allowing software developers and software engineers to use NVIDIA GPUs to accelerate parallel general-purpose computing (only supporting NVIDIA GPUs, unable to be compatible with mainstream GPUs such as AMD and Intel).

Blackwell production delay issue leads to NVIDIA stock price collapse.

However, NVIDIA refused to comment on recent media reports on the manufacturing issues of Blackwell architecture chips. The company spokesperson said that Blackwell's samples have been widely sent to customers, and the demand for its Hopper architecture AI chips in the market is still very strong. Hopper architecture AI chips currently include H100 and H200, which have been mass-produced. NVIDIA also stated that the production pace of Blackwell architecture chips in the second half of the year is expected to accelerate.

Reports of manufacturing challenges related to Blackwell architecture chips have caused NVIDIA's stock price to fall sharply by 6.4% on Monday following last week's correction, but the wider sentiment behind the technology stock crash has also affected its performance. At the same time, NVIDIA's strongest competitor in the AI chip field, AMD (AMD.US), rose against the trend by 1.8% during the US stock market downturn. Investors hope that this may be an opportunity for AMD AI GPU to capture more market share from NVIDIA.

NVIDIA's AI chips occupy a core position in the data center hardware of cloud computing giants such as Microsoft and Google, and these cloud giants are NVIDIA's largest scale customers. Currently, these cloud giants are spending billions of dollars to build super-large-scale AI data centers around the world to meet the huge cloud computing resources needed for their global customer layouts of artificial intelligence technology.

Nvidia CEO, Huang Renxun, announced at the earnings conference in May that the Blackwell architecture AI chip will be fully produced soon and is expected to deliver a certain scale of volume to cloud computing providers in the fourth quarter of this year. He predicted that the demand for new and previous Hopper architecture products would continue to exceed supply. "This year we will see Blackwell architecture sales," Huang Renxun said during an earnings conference call with analysts. The company plans to release its next quarter earnings report on Aug. 28 East Coast time.

Nvidia's next-generation architecture AI chip, the Blackwell series based on Blackwell AI chips, will see a new high-performance boost. Tech giants such as Amazon, Dell, Google, Meta and Microsoft will heavily configure Blackwell AI GPU in their latest data center AI server systems. Wall Street analysts generally believe that the demand for Nvidia hardware from these tech giants will far exceed market expectations. Recently, industry insiders revealed that due to strong demand for Nvidia's upcoming Blackwell architecture AI chips worldwide, Nvidia has increased its AI chip foundry orders with TSMC by at least 25%.

Nvidia's hottest AI chip, H100/H200 GPU accelerator, is based on Nvidia's breakthrough Hopper GPU architecture and offers more powerful computing power than its previous generation, especially in floating-point operations, tensor core performance, and AI-specific acceleration. Even more impressive is that the Blackwell architecture AI GPU performance is far higher than that of the Hopper architecture. On GPT-3 LLM benchmark with 175 billion parameters, Blackwell architecture GB200 inference performance is 7 times that of the H100 system, and provides 4 times the training speed of the H100 system.

Nvidia remains the mainstream call on Wall Street and the "buy on dips" demand is strong.

Although delays may affect Nvidia chip manufacturer TSMC's chip manufacturing process and production capacity planning, forcing TSMC's tight CoWoS capacity to shift to chip giants such as AMD, Wall Street analysts have mostly taken a calm attitude and still insist on a bullish trend for Nvidia's stock price.

Based on the fact that Nvidia will launch a new generation of Blackwell AI GPUs and the demand for Nvidia H100/H200 AI GPUs is still extremely strong, some Wall Street analysts expect this to stimulate a new round of revenue and stock price growth for Nvidia. Therefore, they are more firm in their bullish view of Nvidia's stock price trend within 12 months, believing that the current decline is an excellent opportunity to "buy on dips".

"Given the acceleration of innovation, production capacity disruptions may continue," said Matt Ramsey, an analyst at TD Cowen, in a research report.

He emphasized that even if delays in delivery occur for a few weeks, this may not have a significant negative impact on Nvidia's explosive revenue growth rate or long-term performance growth rate this year. However, Ramsey said this largely depends on how Nvidia ultimately fixes these issues and how fast it delivers chips to major customers. The analyst reiterated a "buy" rating on Nvidia and reiterated a 12-month target price of $165 (Nvidia shares fell 6.36% to $100.45 on Monday).

Morgan Stanley recently stated that Nvidia hopes to "redesign" the Blackwell architecture AI chip to further improve the stability of the chip and servers equipped with Nvidia AI chips. Due to the CoWoS-L packaging yield issue, B200A is expected to switch to CoWoS-S packaging. Although the Blackwell architecture chips may be temporarily halted at TSMC due to the redesign, TSMC's chip and packaging capacity is expected to recover the delayed capacity after the expansion in the fourth quarter, and ultimately deliver Blackwell chips to cloud giants on schedule. Morgan Stanley reiterated its 12-month target price for Nvidia of up to $144.

Bank of America emphasized that Nvidia's recent stock price decline provides an excellent opportunity to buy on dips and reiterated a target price of $150 and a preferred investment target for the chip industry. Regarding reports of a delay in Nvidia's next-generation Blackwell chip launch, Bank of America stated that Blackwell may not be included in performance until the fourth quarter, and even if production capacity is delayed, it may not affect Nvidia's Blackwell-related revenue. Nvidia and TSMC may solve chip design and engineering problems together in the short term. At the same time, Bank of America emphasized that the strong life cycle of Nvidia's previous generation Hopper products will continue to drive high-speed performance growth for Nvidia, and that the challenge is not market demand but the supply side.

Editor/new