Kioxia FQ2 27 – Canary?

Kioxia’s issues are Kioxia specific.  

• Kioxia reported good results but guided badly for the coming three months, immediately raising concerns that demand is softening and that AI rollout is starting to flatline.
• However, Kioxia had a reasonable explanation for its guidance and Sandisk did not corroborate the commentary implying that this is a Kioxia specific event rather than the sign of something systemic.
• FQ2 27 revenues / EPS were JPY2.48tn / JPY836, broadly in line with expectations but guidance missed.
• Here, FQ3 27 operating profit is expected to be JPY2tn, offering slight growth QoQ and 14.9% below the consensus operating profit forecast of JPY2.35tn.
• FQ3 27 Revenue guidance was also 11.6% light at JPY2.50tn compared to consensus at JPY2.83tn.
• There are several reasons for this.
  • First, pricing: where the company expects pricing will increase by 3-4% QoQ, well down from the double-digit growth it experienced in the first two quarters of fiscal 2027.
  • The immediate question that leaps to mind is whether supply and demand are moving towards balance, which would be the first indicator of the top of the memory cycle.
  • This is counterintuitive because the company assured the market that it remains sold out for all of calendar 2027 and that 50% of calendar 2028 is already covered by its long-term contracts.
  • Assuming the company is being truthful on demand, this looks like the first impact of the long-term contracts that the memory industry has entered into, where they have sold further short-term pricing upside in return for longer-term guarantees of volume and pricing.
  • Management also stated that pricing for half of the shipments for FQ3 27 was still being negotiated, and so there is some scope for this forecast to be beaten.
  • Furthermore, CEO Hiroo Ota stated in an interview that prices had risen enough and that further price increases could hurt demand, as is happening in the device market, which makes up roughly half of its revenues.
  • He stated that he had instructed his sales teams not to push for very aggressive price increases.
  • Sandisk also sees modest pricing increases for Q4 26 but is guiding to double-digit revenue growth as a result of its volume expectations for the coming quarter.
  • This brings us to the revenue miss (see below).
  • Second, deferrals and pull-forwards were blamed for the revenue miss, where the weaker revenues are a result of revenue recognition timing, not a fall in demand.
  • In FQ2, some orders were fulfilled early, I suspect as a result of being offered a higher price, while some contracts that were expected to fall in FQ3 have been pushed out to FQ4.
  • Hence, the company stated that it expects growth to be higher in FQ4 as a result of delays as opposed to cancellations.
  • These contracts are typically data centre contracts which take longer to fill and face the risk of delays given that 45%+ of all US data centres encounter permitting, power or political delays.
  • Furthermore, half of its revenues come from consumer electronics where demand has already fallen hard due to the increases in costs from memory, and it looks to me that it is this market that Ota-san is trying to protect.
• I think that the majority of the evidence points towards this being a Kioxia-specific anomaly rather than the sign of the top of the memory cycle.
• This is because both Kioxia and Sandisk remain sold out for calendar 2027, there has been no let-up in data centre capex forecasts, which keep rising, not falling, and there has been no impact on the DRAM makers.
• The situation in NAND is not nearly as severe as it is in DRAM and so there is no direct read across to be taken.
• Furthermore, Samsung and SK Hynix are completely ignoring the 17% decline in Kioxia’s share price this morning.
• Hence, I don’t think that this is a sign of the top, but its is pretty clear that the very rapid rises in prices are going to slowdown materially in 2027.
• This fits with the notion that 2027 is going to be fine but there are questions about 2028 and when the market will begin to price this risk in.
• This is exactly the kind of signal that will indicate the peak of the cycle but in this instance, I think it is noise.