Nvidia Q2 FY2027: The $279 Billion Supply Commitment Says the Memory Crunch Is Structural
Nvidia reported $96.2 billion of revenue for the second quarter of fiscal 2027, up 106 percent year over year and 18 percent sequentially, with data center at $89.0 billion and gross margin holding at 75.0 percent on both a GAAP and non-GAAP basis. The third-quarter guide landed at $108.0 billion against roughly $104 billion of consensus. None of that is the disclosure worth reading twice. The number that carries information is $279 billion, the company’s supply obligations, against the $145 billion figure disclosed three months earlier. Nvidia nearly doubled its forward purchase commitments in a single quarter and management tied the bulk of the increase to memory procurement for the Vera Rubin ramp.
A company does not commit $279 billion to a component it expects to get cheaper. That is the entire argument, and everything else in the release corroborates it.
The margin path is the corroboration. Gross margin was 75.0 percent in the quarter, guided to 74.0 percent for Q3, guided to bottom at 71 to 72 percent in Q4, and guided to settle at 72 to 73 percent for fiscal 2028 once executed price increases take effect in the first quarter. Read the last number carefully. Nvidia’s own recovered margin, after the price increases land, sits below the level it just delivered. The company is modeling elevated memory input costs through the end of fiscal 2028. That is close to a two-year view, and it comes from the buyer with the deepest HBM relationships and the largest checkbook in the industry. Colette Kress framed memory scarcity as a byproduct of the AI buildout itself, which is the structural claim stated plainly: the demand that drives accelerator volume is the same demand that starves accelerators of memory.
The mechanism is not new but the quarter dates it. HBM production consumes wafer starts that would otherwise produce conventional DRAM, so the server DRAM sitting alongside the accelerator in the same rack reprices on the same shortage. Server DRAM contract pricing roughly doubled in the first quarter of calendar 2026. TrendForce carries an estimate in the region of 260 percent for the full year. Memory now accounts for something near a quarter of what a high-end AI rack costs to build, against a much smaller share two years ago. SK Hynix sold out its 2026 output before the year opened. DRAM capital spending is rising far faster than wafer capacity, which is the tell that the money is buying HBM conversion rather than net bit supply. Nvidia’s response was not to wait for that supply. It was to write a commitment large enough to secure allocation and then push more than 15 percent of price increase onto AI server buyers into early next year.
That reshapes the competitive question in a way worth separating from the margin arithmetic. Nvidia’s moat has been the CUDA install base and the rack-scale system, both of which are software and integration advantages. This quarter added a third, and it is financial. A $279 billion purchase commitment is itself a barrier to entry. AMD named HBM4 supply as a constraint on the MI450 and Helios ramp on its own second-quarter call. Custom accelerator programs stand in the same queue. Design differentiation does not ship without memory allocation, and allocation now goes to whoever will commit balance sheet years in advance of revenue. That converts part of the accelerator race from an engineering contest into a financing contest, which advantages Nvidia, Broadcom, and the hyperscalers who can prepay, and quietly disadvantages every entrant that cannot. The cost of holding that position is that the marginal margin point migrates upstream. The three points Nvidia gives back at the Q4 trough are not destroyed. They land in the income statements of Samsung, SK Hynix and Micron, which is why the correct read of a Nvidia gross margin cut, for anyone holding the memory names, is thesis confirmation rather than warning.
Shares closed at $210.65 ahead of the print and were essentially unchanged on the release itself, then ran roughly 4 percent to $219.53 once Huang put approximately 70 percent fiscal 2028 revenue growth on the table against consensus nearer 45 percent, with management describing that figure as supply-constrained and actual demand as materially higher. Market capitalisation sits at $5.16 trillion on about 33 times trailing earnings, inside a 52-week range of $164.07 to $236.54, against an average street target near $310 across 59 covering analysts. The market bought the growth line and discounted the margin guide. That is the right trade only if the trough is genuinely a trough.
The base case is $200 to $240. Q4 lands at 71 to 72 percent as guided, the price increases execute in the first quarter, fiscal 2028 tracks near the 70 percent growth number, and gross profit dollars compound straight through the margin compression: the Q3 guide implies roughly $80 billion of gross profit against roughly $72 billion in the quarter just reported, an 11 percent sequential increase while the percentage falls. The bull case takes the stock back through the $236.54 record toward $270 to $300, triggered by a first-quarter print showing gross margin recovering above 73 percent with the increases holding, or by any relaxation of the China exclusion, which currently contributes zero data center compute revenue to the outlook. The bear case is $165 to $185 and it is a cohort derating rather than a Nvidia-specific failure. If the Q4 trough breaks below 71 percent, or the fiscal 2028 range moves down a second time, the entire memory-exposed AI complex reprices together and Nvidia at 33 times becomes the most liquid instrument for expressing that view.
The decision-relevant disclosure is the supply obligations line in the third-quarter 10-Q. If it prints above $279 billion, Nvidia is still buying into scarcity and the crunch has further to run. If it flattens, the company believes it has bought enough. That single line will say more about 2027 memory pricing than any spot quote.