ALAB Q2 2024: Robust AI Demand and Gen6 Design Wins Propel Growth
- Increasing content per GPU: Executives repeatedly highlighted that PCIe retimer content per GPU is set to grow—not only in current AI platforms but even more so with the shift to next-generation architectures such as Nvidia’s Blackwell GPUs (e.g., GB200) and PCIe Gen 6, suggesting inflation in average revenue per system.
- Robust pipeline and early design wins for next-generation products: The company is already shipping preproduction quantities for its PCIe Gen 6 products and has secured significant design wins in both AI and internally developed accelerator platforms, positioning it well for accelerated growth as these high-performance systems move into full production.
- Diversified market penetration across AI and general compute segments: The discussion points indicate expanding opportunities not only within specialized AI servers but also in general purpose compute systems, as design wins in GPUs and ASICs evolve. This diversification reduces risk and enables multiple growth vectors for revenue expansion.
- Slower penetration in general-purpose servers: Analysts noted that the growth in retimer adoption for general compute has been slower than expected, suggesting that revenue diversification away from AI-specific platforms might be delayed or limited.
- Reliance on preproduction orders for next-generation products: The company’s Gen 6 products are still in preproduction with design wins that have yet to convert to full volume production, creating uncertainty around future revenue realization.
- Margin pressure from a shifting product mix: Q&A comments indicated that a product mix shift toward more hardware-heavy solutions (e.g., Taurus modules) is expected to compress non-GAAP margins, which may negatively impact near-term profitability.
-
Growth Outlook
Q: Confidence in growth from AEC ramps?
A: Management expressed strong confidence in quarter-on-quarter growth driven by new product cycles, robust AI demand, and increased hyperscaler capex—even though guidance doesn’t extend beyond one quarter. -
Server Penetration
Q: Retimers in general servers potential?
A: They expect adoption of their retimer solutions in general-purpose servers to pick up in 2025, leveraging existing design wins as OEMs refresh fleets. -
Competitive Dynamics
Q: How’s the PCIe Gen5/6 competition?
A: Management emphasized their integrated, software-enabled approach that eases transition from Gen5 to Gen6, underscoring early preproduction shipments and secure design wins. -
Gen 6 Market Share
Q: What share for PCIe Gen6 design wins?
A: They stressed being first to market on Gen6 opportunities by smoothly upgrading from Gen5, thereby capturing key market share as hyperscalers accelerate production. -
800-Gig Timing
Q: When will 800‑gig deployments start?
A: The move to 800-gig systems is forecast for late 2025, following ongoing 400‑gig deployments and the emergence of advanced back‑end connectivity options. -
Accelerator Mix
Q: Are ASICs growing versus GPUs?
A: Management noted a rising mix of internally developed AI accelerators that drive higher retimer content, signaling a significant shift from GPU-dominant configurations. -
Content Skew
Q: Is GPU content uniform across designs?
A: They explained that connectivity content per GPU varies by topology—with complex, high-performance systems like Nvidia’s Blackwell seeing a higher component count—though specifics differ across deployments. -
Optical Inflection
Q: When does PCIe optical become dominant?
A: They indicated that while optical connectivity shows promise over longer distances, a majority shift depends on wider hyperscaler adoption and evolving network architectures. -
Module vs. Stand-Alone
Q: Modules or stand-alone retimer revenue?
A: They clarified that the bulk of Aries revenue comes from stand-alone retimers, with Taurus modules also contributing, collectively supporting robust AI server growth. -
Leo Ramp Timing
Q: When will Leo’s ramp materialize?
A: A material production ramp for Leo is expected around 2025, once CXL 2.0–compatible CPUs become available, spurring memory expansion applications. -
Revenue Segments
Q: Revenue split by product segments?
A: Although details were not broken out, management highlighted that strong Q2 performance was driven by Gen5 AI servers, with upcoming boosts expected from both Aries and Taurus segments. -
400-Gig/SCM Scale
Q: How broad is the 400‑gig/SCM ramp?
A: The ramp is supported by diverse hyperscaler wins in multiple cable configurations, initially led by select partners and expected to broaden as demand for higher speeds escalates. -
Content Framework
Q: How does internal versus third‑party content compare?
A: They mentioned that while it’s hard to generalize, both internal and third‑party platforms benefit similarly when employing PCIe protocols, with connectivity scaling to meet varied form factors. -
Blackwell Diversification
Q: Did Blackwell spur architecture diversification?
A: The Blackwell transition has led to a broader array of deployment configurations, with increased connectivity content addressing challenges in power and thermal management. -
Blackwell Follow-Up
Q: Impact of Blackwell architectural changes?
A: Management refrained from commenting on rumors but reiterated their commitment to supporting all preferred deployment models—from full racks to HGX servers—ensuring design win flexibility. -
Revenue Drivers
Q: What balances overall revenue growth drivers?
A: They pointed to balanced contributions from Taurus modules, incremental Aries modules, and rising Gen5 deployment content, which together underpin Q3 growth expectations. -
Lockup Expiration
Q: Confirm early lockup expiration?
A: They confirmed that the early lockup has already expired and is no longer a factor, with details confirmed in today’s release.