It’s All About the GPUs — 70% of Nvidia’s AI Supercomputer Cost Goes to Vera Rubin Chips
Inside Nvidia's newest rack-scale AI supercomputer, a single line item on the bill of materials is swallowing the lion's share of a $4 million build, and it tells you exactly who wins and loses as hyperscalers race to spend $1.3…
This post may contain links from our sponsors and affiliates, and Flywheel Publishing may receive compensation for actions taken through them.
Number That Reframes the Vera Rubin Trade
Inside an NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) VR200 NVL72 rack-scale AI supercomputer, 70% of the roughly $4 million bill of materials goes to one line item: the Rubin GPUs themselves, excluding their high-bandwidth memory. That figure comes from an HSBC estimated bill of materials for the VR200 NVL72 platform, and it is the cleanest single data point yet on where the AI infrastructure dollar actually lands. DRAM adds another 13%, HBM 9%, NAND 2%, NVLink 2%, cooling 2%, power supply 1%, and the CPU 1%.
What 70% Actually Tells You
In an era where every hyperscaler is trying to bend the AI capex curve, the VR200 breakdown says the GPU is still where the value concentrates. Nvidia is selling the rack, and the silicon it designs takes seven of every ten dollars a customer spends to fill it. That maps to what CEO Jensen Huang told investors on the fiscal Q2 2027 call: “Today, we’re not just selling the best chips. We’re selling a full-stack AI factory platform.”
The scale of that platform is showing up in the reported numbers. Nvidia’s data center revenue hit $89.02B in Q2 FY2027, up 117% year over year, on total revenue of $96.22B with a non-GAAP gross margin of 75.0%. Management guided fiscal Q3 revenue to $108.0B plus or minus 2%, and expects Vera Rubin to account for about 20% of data center revenue in Q3. Huang described the platform economics bluntly: “For Hopper, we were at about 18 billion per gigawatt. For Grace Blackwell, we’re about 25 billion per gigawatt. And for Vera Rubin, it’s about 40 billion per gigawatt.”
Market Reaction
NVDA trades at $212.35 intraday, down 7.71% over the past week and down 5.58% over the past month, as tech leaders publicly floated an AI slowdown narrative (MarketWatch, Sept. 14, 2026). Zoom out and the picture looks different: shares are up 14.13% year to date and up 854.66% over five years. Peer memory suppliers are riding the same wave. Micron Technology (NASDAQ:MU) trades at $929.17, up 225.76% year to date, and SK hynix (NASDAQ:SKHY) trades at $175.21.
Bull Case
The 70% figure is a value-capture story. Every rack Nvidia ships routes the majority of its BOM back to Nvidia’s own silicon, and the demand for those racks is already booked. Huang told analysts Nvidia has “already received purchase orders from every major hyperscaler, AI cloud, and system OEM” and expects Vera Rubin to be the fastest product ramp in company history. Supply is the ceiling: “We expect to grow revenue by approximately 70% in fiscal 2028” while customer forecasts imply doubling, because “at this moment, we have supply for 70%.”
The demand pool behind that constraint is unusually concrete. Top-five hyperscaler capex is expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027, AWS is deploying an additional 2 million GPUs, and OpenAI’s commitments represent roughly 12 gigawatts of NVIDIA compute. Nvidia backed the visibility with balance sheet: total supply obligations surged to $279.0B, largely tied to memory procurement for Vera Rubin production.
The 24% of BOM sitting in DRAM and HBM is the reason Micron and SK hynix keep printing records, and it is why Nvidia has locked in a multiyear technology partnership with SK hynix for next-generation memory. Micron’s fiscal Q3 gross margin hit 84.9% on revenue of $41.5 billion, with management flagging tight DRAM and NAND conditions beyond calendar 2027. That pricing pressure is real for Nvidia, and the company acknowledged margins bottoming in Q4 in the 71% to 72% range before recovering. But the 70% BOM share means the price hikes Nvidia is passing through to customers matter more to the P&L than the memory bills it is paying. The other 30% of the rack (memory, cooling, power, networking) is its own investable theme, and we profiled seven suppliers behind that buildout in a free report on the AI stocks that aren’t chipmakers.
Bottom Line
For long-term holders, the number reframes the debate. The AI slowdown chatter dominating this week’s headlines assumes the GPU tier is commoditizing. The VR200 BOM says the opposite: Nvidia’s silicon is still where the AI capex dollar lands, and the platform is expanding into CPUs, networking, and software around it. At a forward P/E of 24 with Q3 FY2027 guidance of $108.0B and Vera Rubin ramping into every hyperscaler order book, the setup rewards patience over the next few gigawatts of buildout, not just the next earnings report.
Contact [email protected] for any questions or corrections.
After two decades of patrolling the dark corners of suburbia as a police officer, Rich Duprey hung up his badge and gun to begin writing full time about stocks and investing. For the past 20 years, he’s been cruising the markets looking for companies to lock up as long-term holdings in a portfolio while writing extensively on the broad sectors of consumer goods, technology, and industrials. Because his experience isn’t from the typical financial analyst track, Rich is able to break down complex topics into understandable and useful action points for the average investor. His writings have appeared on The Motley Fool, InvestorPlace, Yahoo! Finance, Money Morning, and, of course, 24/7 Wall St. He has been featured in both U.S. and international publications, including MarketWatch, Financial Times, Forbes, Fast Company, and USA Today.
