To The Moon

When AI bears look at the low multiples $SK hynix(SKHY)$, $Micron Technology(MU)$, $SanDisk Corp.(SNDK)$ and Samsung then the response is usually some version of “earnings multiples are meaningless for cyclical companies at peak demand and peak margins.”

The reason these suppliers trade at such low multiples is precisely because the market already assumes today’s earnings are temporary... people aren't valuing current margins as permanent so the debate is really about duration.

If earnings peak this year and begin collapsing shortly afterward then the low multiples are traps and I agree with the bears but if supply remains tight for several more years (which I do) then those valuations are WAY understated with how much cash the companies will generate before the cycle turns.

SK Hynix has said the shortage could remain severe through the end of the decade while Samsung has also pointed toward meaningful tightness continuing through at least 2027.

These companies obviously benefit from presenting a strong outlook so their forecasts shouldn't be accepted blindly but they also have the clearest view into customer commitments, equipment orders, wafer capacity, packaging constraints and the production schedules of the entire industry.

The double-ordering argument also deserves a bit more context because a weak fill rate can sometimes indicate ghost demand but it can also mean customers are trying to secure supply they genuinely cannot obtain.

The stronger evidence for me is from the structure of the agreements since customers are signing multiyear contracts, accepting pricing floors and ceilings, providing prepayments and in some cases helping suppliers fund additional capacity (temporary demand usually doesn't commit capital years ahead or help finance the supplier’s expansion).

This cycle also looks different from a traditional cycles because the physical bottleneck is much harder to solve since HBM consumes substantially more wafer capacity than standard DRAM while each new generation becomes more difficult to manufacture and package.

As the memory suppliers move from HBM3E toward HBM4, HBM4E and later generations, part of every capacity increase is absorbed by the greater manufacturing intensity of the product itself.

That same narrative supports $Taiwan Semiconductor Manufacturing(TSM)$ and $ASML Holding NV(ASML)$ because more advanced AI chips require leading edge wafer capacity, EUV tools, advanced packaging and years of coordinated capital investment before supply can meaningfully expand.

TSMC doesn't add capacity overnight while ASML machines sit at the beginning of nearly every leading edge production roadmap which makes the supply response slower and more capital intensive than a normal short-cycle industry.

$NVIDIA(NVDA)$, $Advanced Micro Devices(AMD)$, and $Broadcom(AVGO)$ sit one layer closer to the end demand but the same duration question applies since bears say $Meta Platforms, Inc.(META)$, $Alphabet(GOOGL)$, $Amazon.com(AMZN)$ and $Microsoft(MSFT)$ are overordering accelerators and custom silicon which eventually creates excess inventory and falling margins but my takeaway is that training demand is being joined by inference, agentic AI and custom silicon programs that extend the cycle way longer than people expect.

The argument that these companies are at peak demand also becomes less convincing when the bottlenecks continue moving across the stack since one quarter the shortage is GPUs, then its HBM, then advanced packaging, then optical connectivity, then power, cooling or data center capacity.

Again that doesn't mean shortages last forever but it does suggest the buildout is still expanding across several physical layers rather than reaching a clean and obvious peak.

I agree that cyclicality hasn't disappeared since semiconductors will eventually reach a point where supply catches up with demand, pricing weakens and margins decline but I really think people need to compare the expected duration of the shortage with current valuations, contract structures, capacity timelines and the risk that AI demand slows before new supply reaches the market.

The industry will eventually normalize but if the market is valuing many companies as though earnings collapse next year while supplier commentary and fab timelines point to tightness lasting much longer then the low multiple may represent opportunity rather than warning.