Bernstein: Surging Memory Prices May Strengthen Nvidia's Pricing Power
0xBroomberg
Bernstein estimates memory costs will exceed one-third of Nvidia's next-gen Vera Rubin system price, with HBM prices more than doubling by volume shipment; the firm believes Nvidia can pass the increase to customers, keeping margins intact.
How expensive is memory in the next-gen system?
Bernstein estimates the Vera Rubin NVL72 rack will cost roughly $9.1 million, with memory and storage alone at about $3.2 million — over one-third of the total.
The main driver is HBM — high-bandwidth memory purpose-built for AI chips — which the firm expects to more than double in price by the time Rubin ships at scale.
This means → in the next generation of AI systems, memory is no longer a supporting component — it is the single largest cost item.
Who absorbs the price hike?
Bernstein believes Nvidia has enough pricing power to pass memory cost increases through to customers rather than absorbing them.
The report states: "We believe Nvidia likely has some dynamic pricing mechanism to pass this on to customers rather than absorb it through lower margins."
In plain terms = Nvidia sells a full AI compute stack its customers cannot easily walk away from, so higher input costs end up on the buyer's bill — Nvidia's own margins stay largely intact.
How much more will a data center cost?
A 1-gigawatt Vera Rubin data center will cost roughly $47 billion to build, up from $40.5 billion for the prior chip generation, per Bernstein's estimate.
Beyond memory, rising networking, cooling, and power costs also contribute to the increase.
This reflects a broader trend: AI infrastructure cost inflation is no longer just about the chip — it is the entire supply chain repricing simultaneously.
What are the big customers saying?
Meta CEO Mark Zuckerberg and Microsoft CFO Amy Hood both cited rising chip and data-center memory costs as a key driver of higher AI spending in recent earnings calls.
The trend has already driven sharp gains in memory-related stocks.
This means → large downstream buyers are already paying the higher prices, and the market has begun pricing this pass-through logic into equities.
How is Nvidia locking in supply?
Nvidia CEO Jensen Huang has warned that memory shortages could last "quite a few years."
This week Nvidia and SK Hynix announced a multi-year partnership to co-develop next-generation memory for Vera Rubin and future systems; SK Hynix's market cap crossed $1 trillion in late May.
In plain terms = Nvidia is locking down the scarcest resource early through long-term contracts — a move that is both defensive (securing its own supply) and offensive (making it harder for rivals to access the same grade of memory).
Where is the limit of that pricing power?
Bernstein's core thesis: Nvidia can fully pass through cost increases to customers.
That thesis holds only as long as customer AI investment appetite stays strong and no sufficiently compelling alternative platform emerges.
This means → whether memory costs can be smoothly passed on will be the critical test of Nvidia's pricing-power boundary — if customers begin deferring purchases or shifting to rival platforms, this logic chain breaks.
Content is for reference only, not financial advice.