#CPO Last night's CPO hot take, I see it differently. The hype cycle for AI should be stretched out over time; pumping it all at once isn't great for maximizing our trading profits.
Now the CPO hype is getting pushed back. Prices shot up too fast, and I'm hesitant to buy. If they say it won't really explode until two years later, that's not a good sign, right?
Last night, SemiAnalysis reported that:
NVIDIA's native single-ended 800VDC mass production has been pushed to after 2028, not the earlier ramp the market was hoping for. The ±400VDC rollout is still on track, mainly for hyperscaler self-developed ASIC deployments, and sidecar orders are expected to land in the second half of 2026, ramping up production in 2027.
Rubin Ultra / Kyber related sidecar volume was originally expected to ramp up sooner, but now that's pushed to the 2028 window.
CPO expectations have been downgraded: the 2027 scale-up CPO shipment expectations seem overly optimistic, and SemiAnalysis believes these will be delayed compared to current Street expectations.
The real large-scale scale-up CPO ramp might not happen until 2029, instead of what the market is shouting about for 2027/2028.
Doesn't this tell us that the CPO hype lifecycle has been extended?