NVIDIA Blackwell costs twice peers

Nvidia goes into its May 20 earnings report with two facts in view: demand for Blackwell remains strong, and supply is still a constraint. Forbes reported on May 18 that Nvidia’s issue is not weak appetite for its newest AI systems, but the infrastructure and fulfillment gap between customer orders and recognized revenue. Morgan Stanley’s latest view, cited by Wccftech on May 18, adds a second point. The bank said Blackwell is priced at roughly twice the level of Google and Amazon custom AI chips, but argued the premium is justified by better performance per watt. Wccftech said Morgan Stanley estimated Nvidia’s advantage at 2x to 8x over custom ASIC alternatives in that metric. (forbes.com) ### If Blackwell costs more, why are customers still buying it? Morgan Stanley’s argument, as reported by Wccftech, is that hyperscalers care less about sticker price than about how much useful compute they can deliver per watt at scale. The report said that calculation supports paying more for Nvidia systems if they reduce power and infrastructure burden across large AI deployments. (wccftech.com) CNBC reported in November that Google, Amazon, Meta, Microsoft and OpenAI were all developing or using custom AI chips as alternatives to Nvidia, with Google’s TPUs leading the ASIC field. But CNBC also said those chips are typically narrower in scope than Nvidia’s GPUs, which remain the standard for a wider range of AI training and inference workloads. (wccftech.com) ### What is the bottleneck if demand is already there? Forbes said the constraint sits between bookings and delivery. Its May 18 report described the risk as infrastructure-related, meaning Nvidia can have customer demand in hand while still facing delays in turning that into shipped systems and booked revenue. (cnbc.com) Nvidia’s investor relations page shows the company will report first-quarter fiscal 2027 results on May 20. That release and call are the next scheduled point for investors to hear management address Blackwell supply, Hopper demand and data-center revenue directly. ### Where do Google and Amazon fit into this comparison? (forbes.com) Google and Amazon are the reference point because both companies have pushed custom silicon to reduce dependence on Nvidia and lower AI infrastructure costs. Wccftech’s summary of Morgan Stanley’s note framed Blackwell against Google TPUs and Amazon Trainium, saying Nvidia still commands a premium despite those lower-cost in-house options. (investor.nvidia.com) CNBC’s broader comparison of Nvidia GPUs with custom chips said the appeal of ASICs is lower cost and tighter workload specialization. The trade-off, according to CNBC’s reporting, is that Nvidia still offers a more general-purpose platform with a larger software and deployment ecosystem. ### Why does performance per watt matter so much now? (wccftech.com) Power has become a first-order constraint in AI infrastructure. Morgan Stanley’s view, as cited by Wccftech, is that a chip that delivers more output for the same energy budget can justify a higher purchase price, especially for cloud operators building at very large scale. (cnbc.com) That logic pushes attention away from unit price alone and toward system efficiency. For production teams, that means the economics of AI serving increasingly depend on batching, model routing, quantization and workload placement across different hardware pools — an inference drawn from the premium Morgan Stanley assigns to higher performance per watt. (wccftech.com) ### What should readers watch on May 20? Nvidia is scheduled to release first-quarter fiscal 2027 results after the bell on Wednesday, May 20, and to hold its conference call at 2 p.m. Pacific time, according to its investor relations page and related event listings. Investors will be looking for specific comments on Blackwell shipments, data-center demand and whether supply constraints are easing. (investor.nvidia.com) (wccftech.com)