InvestNot yet confirmed elsewhere1 publisher2 min readPublished Updated
AMD's 432GB answer to Vera Rubin turns the accelerator fight into a memory-cost fight
The MI455X carries 50% more HBM than the chip it replaces, and AMD's memory supplier just posted a record 76% operating margin. Rack buyers are now budgeting memory, not compute.
The Investor · Invest desk
What happened
- AMD's next flagship accelerator, the Instinct MI455X, is specified at 432GB of HBM4.
- The current MI355X carries 288GB of the previous memory generation, making the step up exactly 50%.
- AMD guided third-quarter revenue to about $13 billion with adjusted gross margin held near 56%.
Why it matters
- cost With AI-server memory leading SK Hynix's price increases, the memory market now shows up in rack quotes rather than in a component footnote, and the buyer signing for 2027 capacity is underwriting...
- constraint Su's 2027 allocation secures parts on schedule but not their cost, so AMD cannot contract its way out of memory inflation, only manage around it.
- decision By guiding margin flat, AMD has pre-committed to either passing HBM4 costs to customers or absorbing them through efficiencies; the next few margin prints settle which one it chose.
- exposure At about 31 times next year's expected earnings, shareholders near $479 own the assumption that the pass-through works.
Set AMD's two comparison numbers next to each other and the pitch resolves. Against Vera Rubin, AMD's own product page claims 50% more memory capacity and a 15% edge in peak performance on one AI math format [4]. The capacity claim is more than three times the size of the compute claim [16]. That is a company telling buyers which line of the spec sheet to shop on.
The same table reads a second way. At 288GB, Nvidia's new generation lands, by AMD's accounting, at exactly the capacity AMD already ships in the MI355X [17]. Both vendors have settled on gigabytes as the axis, which means both are buying more of the same input from the same short list of suppliers.
At rack scale the arithmetic stops being abstract. Helios is specified at 31 terabytes of HBM4, about 72 chips' worth [10], and 31TB across 72 sockets works out near 430GB each, consistent with the chip spec [18]. Build that same 72-socket rack at the previous 288GB and it holds roughly 20.7 terabytes, so Helios carries about ten terabytes more HBM in the same box [19]. Ten terabytes is the size of the new cost line, per rack, before anyone negotiates a discount.
The supplier of that increment is having a better quarter than its customer. SK Hynix, the leading HBM maker, reported a 76% operating margin on revenue up 257% year over year [5], while AMD's adjusted gross margin was 56% on record revenue of $11.5 billion [7]. Operating margin normally sits below gross margin; here the memory maker's operating line runs about 20 points above its customer's gross line [21]. SK Hynix said high-performance products for AI servers led its price increases, with DRAM and NAND both up significantly from the first quarter [6]. Holding gross margin flat through a roughly 13% sequential revenue step [20] means pass-through, offsetting efficiencies, or both [14]. And the segment on the hook is now most of the company: data center revenue was $6.7 billion in the second quarter, 58% of the total [8].
The Anthropic agreement is where the exposure runs longest. Up to 2 gigawatts of MI450-series chips in Helios racks [1] is a commitment sized in years, and Lisa Su has described shipments starting this quarter and ramping into 2027 [11]. Anything quoted that far out has a memory price embedded in it that neither side has seen yet.
What to watch
- Whether AMD's fourth-quarter or 2027 gross margin guidance slips below the 56% line once Helios ships in volume.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence56
- Adoption48
- Hype gap+18
- Incentives64
- Confidence57
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
AMD has announced a partnership under which Anthropic will deploy up to 2 gigawatts of MI450 series chips in Helios racks.
- [2]
AMD's Instinct MI455X, the flagship of the MI400 series introduced this year, is specified at 432 gigabytes of HBM4, the newest generation of high-bandwidth memory.
- [3]
AMD's current top chip, the MI355X, carries 288GB of the prior memory generation; the step up to the MI455X is exactly 50%.
- [4]
AMD's product page stacks the MI455X against Nvidia's Vera Rubin generation: 432GB of memory against 288GB, slightly higher memory bandwidth, and a claimed 15% edge in peak performance on one AI math format. These are AMD's own marketing numbers.
- [5]
SK Hynix, the leading maker of high-bandwidth memory, reported second-quarter revenue up 257% year over year with a 76% operating margin, a record for the company.
- [6]
SK Hynix said high-performance products for AI servers led its price increases, with DRAM and NAND prices rising significantly from the first quarter.
- [7]
AMD's companywide second-quarter revenue reached a record $11.5 billion, up 50% year over year, with non-GAAP gross margin of 56%.
- [8]
AMD's data center segment produced $6.7 billion of second-quarter revenue, 107% more than a year earlier and 58% of the company total.
- [9]
AMD management guided third-quarter revenue to about $13 billion, up about 41% year over year, with adjusted gross margin holding near 56%.
- [10]
Helios, AMD's rack-scale system, packs 31 terabytes of HBM4, about 72 chips' worth, and AMD markets that figure alongside the rack's computing power.
- [11]
CEO Lisa Su said on AMD's early August earnings call that Helios is in production with initial shipments on track to begin later that quarter and ramp through the fourth quarter and into 2027 to meet very strong customer demand.
- [12]
Su said on the same call: "There is very good visibility into HBM allocation for what we expect to deliver in 2027," describing years of work with AMD's memory partners.
- [13]
The Motley Fool's analysis argues that allocation is volume, not price: a supply agreement means the memory arrives on schedule, while what it costs and how much AMD can pass along to customers is the open variable, leaving AMD able to manage but not control the cost side of this generation.
- [14]
The same analysis holds that guiding gross margin flat while revenue ramps implies AMD expects to pass higher input costs through, find offsetting efficiencies, or both.
- [15]
AMD stock trades near $479, at about 31 times what analysts expect the company to earn next year, a price with a smooth ramp already baked in.
- [16]
AMD's claimed memory-capacity advantage over Vera Rubin is more than three times the size of its claimed peak-performance advantage.
- [17]
The 288GB capacity AMD attributes to Nvidia's Vera Rubin generation equals the 288GB AMD already ships on the MI355X.
- [18]
Helios's 31 terabytes across about 72 chips works out to roughly 430GB per socket, consistent with the MI455X's 432GB specification.
- [19]
A 72-socket rack built at the prior 288GB per chip would hold about 20.7 terabytes of memory, roughly ten terabytes less HBM than Helios's specified 31 terabytes.
- [20]
AMD's roughly $13 billion third-quarter guide is about a 13% sequential increase on second-quarter revenue of $11.5 billion.
- [21]
SK Hynix's 76% operating margin sits about 20 percentage points above AMD's 56% adjusted gross margin.
Sources
1 independent publisher whose own reporting we read for this story.
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.