CompaniesAstera Labs

Astera Labs

#company

Thesis

  • The verdict:
    • Close industry connection and knowledge are required for tracking this stock.
    • Risks: Merchant scale-up switching gets crowded or integrated. Customer concentration (Amazon). Customer in-sourcing. Competition from Broadcom, platform migration to Ethernet fabric.
    • Need to watch: broadening customer base for Scorpio?
    • Valuation allows for no margin of error.
  • The business: AI fabric switch.
  • Moat:
    • Qualified silicon at each generation, with a design win to volume lag of up to 2 years.
    • Seats in PCI-SIG, CXL, NVLink, UALink as hedge for standard shift.
    • COSMOS lock-in (?) - fleet telemetry embedded in customer operations -> content compounding.
    • Scarcity of SerDes analog talent, evidenced by Qualcomm buying Alphawave rather than building.
  • Non-moat:
    • The absence of any hyperscaler in-house retiree program is not a sign of barrier.
    • The amazon relationship supported by a warrant is a purchase framework, not an order book.
    • A differentiated platform (owning four categories plus software), but no scale advantage (as evidenced by the high growth at Broadcom and Credo).
  • Growth drivers:
    • Accelerator unit growth, rising networking content per accelerator, mix shift into scorpio switches and hardware modules (higher revenue and lower margin), optical interconnect and custom silicon (a possibility for 2027 and beyond).
  • Economics:
    • $1.25bn of cash, no debt, self-funding.
    • Growth margin decline driven by mix ( shipped into scorpio and hardware modules) and (Amazon) warrants.
    • GAAP earnings are impacted by stock compensation dilution (negatively) and negative effective tax rate (positively).
  • Risks:
    • Merchant scale-up switching gets crowded (e.g. competition from Broadcom) or integrated (by top and customer).
    • Customer concentration - top 1 >70%, top 3 86%.
    • Platform shift to Ethernet; CPO removing the copper being retimed.
    • Roadmap slip, e.g. Leo was promised for 2025 and is now guided to 2027.
  • Unknown:
    • What is the trend of total connectivity content per accelerator?
    • Revenue split by product family.
  • Astera versus Broadcom
    • Lack of scale -> much higher R&D intensity (55.3% versus 23.8%) + paying up to amazon for volume -> lower OM.
    • Much higher customer concentration and break risk from one customer's decision.
    • Much less power over suppliers / higher supply shock risk.
    • Broadcom has custom accelerator co-design competency, and the right to bundle the socket.

Reports

Reference

Built with LogoFlowershow