UBS supply-chain checks are pushing back hard on the "Broadcom loses Google TPU after v8" thesis that's been floating around.
Timothy Arcuri's research note, based on a recent management meeting plus channel checks, says Broadcom stays deeply engaged past v8i. Management is calling TPU worries "misguided" and says engagement remains solid for the next two generations after the v8i ramp.
Here's what the checks are saying:
• Google is replacing the prior v9 Pumafish plan with a larger multi-chip 3nm design. That part is expected to drive most 2028 volume after a late-2027 ramp.
• UBS then sees a 2nm v10 shipping late 2028 and taking most TPU volume in 2029, with up to 16 dies, Broadcom 3.5D packaging, and 448G SerDes.
That's the opposite of recent rumors that MediaTek or another partner takes the socket after v8.
Still a research-note check, not a confirmed Google or Broadcom disclosure. Doesn't settle final share or volume. But if v9 is now a bigger 3nm multi-chip part, does that push MediaTek's role to I/O dies, or off the program entirely?
$GOOGL $AVGO
Timothy Arcuri's research note, based on a recent management meeting plus channel checks, says Broadcom stays deeply engaged past v8i. Management is calling TPU worries "misguided" and says engagement remains solid for the next two generations after the v8i ramp.
Here's what the checks are saying:
• Google is replacing the prior v9 Pumafish plan with a larger multi-chip 3nm design. That part is expected to drive most 2028 volume after a late-2027 ramp.
• UBS then sees a 2nm v10 shipping late 2028 and taking most TPU volume in 2029, with up to 16 dies, Broadcom 3.5D packaging, and 448G SerDes.
That's the opposite of recent rumors that MediaTek or another partner takes the socket after v8.
Still a research-note check, not a confirmed Google or Broadcom disclosure. Doesn't settle final share or volume. But if v9 is now a bigger 3nm multi-chip part, does that push MediaTek's role to I/O dies, or off the program entirely?
$GOOGL $AVGO