Rambus Posts First-Ever $200 Million Quarter as CPU-Driven Memory Demand and Hyperscaler IP Wins Fuel Confidence Into 2027
Q2 2026 earnings call, July 27, 2026
Rambus delivered its strongest quarter on record, crossing $200 million in revenue for the first time and beating the high end of its own guidance, as CEO Luc Seraphin used the call to lay out a more detailed roadmap for how CPU-centric AI workloads, not just GPU accelerators, are reshaping demand for the company's memory interface chips and IP. Revenue came in at $207.4 million, up 20% year-over-year and 15% sequentially, while non-GAAP earnings per share of $0.77 rose 24% year-over-year. It was also the first call for new CFO Sumeet Gagneja, who succeeded the prior finance chief and said his priority is "drive profitable growth through disciplined financial execution, allocate capital thoughtfully and provide shareholders with transparent and consistent communication."
Hyperscalers Are Now Writing the Memory Spec Themselves
The most significant new disclosure was a design win with a Tier 1 U.S. hyperscaler for next-generation HBM controller IP tied to a future AI chip. Seraphin described a structural shift in how the silicon IP business is originated: hyperscalers are increasingly defining memory subsystem architectures internally, ahead of finalized industry specifications, and then handing those specs to ASIC partners or building silicon themselves. "Their role in defining the architectures on complex subsystems like the memory subsystems is becoming more and more important," he said, adding that once a hyperscaler settles on an implementation, "that proliferates into their own ecosystem." This is a license-based, not volume-based, revenue stream — meaning Rambus can recognize IP revenue well before the underlying chip ever ships. Seraphin confirmed the HBM win falls into this category, saying "we see the revenue much earlier than that" relative to the 12-to-24-month lag before customer products actually reach the market. Silicon IP revenue is expected to keep growing 10% to 15% annually, and management said its confidence in that range is "continuing to build."
CXMT's Big Splash Confirmed as a Rambus Positive
Asked directly about China's CXMT, which made headlines the same day as the earnings call, Seraphin was unambiguous about Rambus's licensing exposure: "Every company that builds memory has to have a license agreement with us, and they're one of them. So we're very pleased with the success, and that's going to be a good thing for us in the long run as well." It is a notable data point for investors tracking how Chinese DRAM capacity additions flow through to Western IP licensors regardless of geopolitical friction elsewhere in the semiconductor supply chain.
Inventory Build Is Strategic, Not a Warning Sign
With memory supply tightening across the industry, analysts pressed on whether hyperscalers or OEMs might be over-ordering ahead of shortages. Seraphin pushed back on that framing for Rambus's own customer base but confirmed the company itself is stockpiling: inventory rose $16 million in the quarter. "We don't see any signs of our customers building inventory for the concerns that you expressed. This said, however, we are building some inventory on critical products that we believe are going to ramp here in Q3, Q4 and early next year as we do see our lead times lengthening given the tightness in the supply chain." Management confirmed no capacity constraints hit Q2 shipments, but lead times across the supply chain are stretching, and that dynamic is expected to persist through 2027.
MRDIMM Timeline Pushed to 2027, Not This Year
On the next-generation memory module transition, Seraphin tempered near-term expectations. With AMD's Venice CPU platform ramping and server shipments expected to begin in the fourth quarter, investors have been looking for signs of an accelerating MRDIMM mix. Management's answer: not yet. "The contribution for Q4 is going to be minimal... a more material contribution is going to happen in 2027 when both platforms from the CPU guys ramp in the market in earnest." MRDIMM modules carry roughly four times the silicon content of a standard DIMM, making the eventual ramp a meaningful margin and revenue lever, but Rambus is explicitly not building 2026 guidance around an early inflection.
CXL Gets an IP Endorsement, Not a Product Commitment
Pressed on CXL following Meta's public endorsement of the interconnect protocol, Seraphin drew a sharp distinction between Rambus's IP business, where CXL remains strategically relevant, and its chip business, where the company is deliberately staying on the sidelines. "We continue to see a fragmented market from a product standpoint with many deployments looking like ASIC-like or customer-specific products," he said, adding that Rambus would "rather invest into standard products at this point in time" on the chip side while continuing to support CXL through IP licensing. It's a clear signal that Rambus views CXL adoption as further out and less standardized than MRDIMM, which benefits from being a JEDEC-defined product that slots into existing infrastructure.
LPDDR and SOCAMM2 Framed as Additive, Not Cannibalistic
With SOCAMM2 chipsets for LPDDR5X now shipping, analysts questioned whether lower-content LPDDR-based server modules could eat into Rambus's higher-value RDIMM and MRDIMM franchise. Seraphin rejected that framing: "I wouldn't say that the industry is shifting to LPDDR... DDR will remain dominant where server-grade scale, capacity, reliability, serviceability are required." He characterized LPDDR and SOCAMM as complementary products for power-constrained deployments rather than substitutes, reiterating prior guidance that near-term SOCAMM2 revenue is modest but the strategic positioning is important as AI infrastructure becomes more heterogeneous.
Unit Growth Estimates for Servers Revised Higher
Rambus's own read on the server CPU unit market has shifted meaningfully more bullish over the past quarter. Seraphin noted that where the company previously characterized unit growth as "mid- to high single-digit," it is now tracking double-digit growth, citing Gartner's 12% estimate as a reference point and stating Rambus believes it is outgrowing that market rate. That reassessment lines up with AMD's more aggressive x86 CPU CAGR guidance cited by Baird's Tristan Gerra during the call, and supports management's view that channel-count expansion (AMD moving to 12 channels, Intel moving from 8 to 12, with the broader market eventually shifting toward 16) remains a durable, multi-year unit tailwind independent of any single platform ramp.
Guidance Discipline Amid Bullish Underlying Demand
Product revenue guidance for the third quarter calls for $110 million to $116 million, a 14% sequential increase at the midpoint, with total company revenue guided to $210 million to $216 million. New companion chip products, which management said represented a "low double-digit percentage" of product revenue in the first quarter, are expected to reach "mid double digits" by the end of the fourth quarter. Despite the positive setup, Seraphin repeatedly declined to extend guidance beyond a single quarter, citing both platform ramp timing risk and the ongoing supply chain tightness as reasons for caution. Gross margin is expected to stay within the company's long-standing 60% to 65% model, though Gagneja noted recent quarters have run toward the lower end at 60% to 63%, with quarterly variability driven by mix rather than pricing power, since Rambus's standard-product business model does not allow for the kind of price increases seen elsewhere in the memory supply chain.