Work the margin guide in dollars and it stops looking like a rounding error. Nvidia reported a 75% gross margin for the quarter, in line with estimates [8], then told investors to expect 74% in the current quarter, a floor of 71-72% in fiscal Q4, and 72-73% across the coming fiscal year, with chip price increases already assumed in those figures [10]. Peak to trough, that is 300 to 400 basis points [1]. Against the $108 billion guided for the current quarter [3], three points of gross margin is about $3.2 billion of gross profit in a single quarter [2]. The guidance beat itself was roughly $3.4 billion above the $104.57 billion consensus Business Insider cited [3]. The margin being conceded is about the size of the upside just delivered.
Two lines from the call make the destination of that money readable. Nvidia said memory costs are "extreme," that they surpassed even its own expectations, and that it was therefore "resetting expectations today" [9]. Its margin also does not return to 75% once the price increases land; it settles two to three points below where it sits now [7]. A vendor that raises prices and still earns less on each dollar of revenue is passing through an input it does not price. The buyer pays the increase and the memory supply chain keeps it.
That changes what a 2027 compute plan is negotiating over. Data center revenue was $89.0 billion against an $85.83 billion estimate [6], about 93% of the company total [4], and up roughly 18% from $75.2 billion the prior quarter [5]. There is no second segment to absorb a memory squeeze. Nvidia paired guidance of around 70% fiscal 2028 revenue growth with a warning that supply stays bottlenecked [5], and gave full-year guidance for the first time [16]. Growth is not the open question in that pairing. Input cost is.
The financing loop sharpens it. Nvidia has joined major investment firms on platforms intended to mobilize more than $500 billion for AI infrastructure, according to Forbes [12]. CFO Colette Kress met the label head on: "We recognize the scale of this support, and we know some will call this circular financing," she said. "We see it differently" [11]. Whichever way that reads, a memory-driven price rise now travels through a relationship in which the vendor is helping fund the buildouts that buy its parts.
The planning consequence is unglamorous. Costed in GPU units, 2027 is an allocation problem, and allocation is what large buyers have spent two years learning to win. Costed in dollars per delivered system with memory contracted, it is a procurement problem with a different counterparty across the table, one whose pricing Nvidia has now said out loud it could not forecast for itself.