INTC

Unrelated workloads, same dead silicon

I originally cleaned up these notes as a formal research write-up when applying to D.E. Shaw, mostly to force myself to separate what was genuinely knowable in real time from where I got lucky on macro timing.

Throughout early 2024, weird crash logs kept piling up around flagship Intel 13th and 14th-gen Raptor Lake chips. Looked at in isolation, any single report was easy to brush off as aggressive motherboard power defaults, unstable DDR5 profiles, GPU out-of-memory errors, or buggy game code. Read side by side, those excuses stopped holding up.

RAD Game Tools posted an unusually blunt engineering note documenting repeatable Oodle decompression faults on those exact processors.[1] Level1Techs then dug through crash telemetry and found roughly 50% failure rates on workstation W680 server motherboards, hardware built specifically for locked-down stability with zero overclocking.[2] Days later, Alderon Games published their own failure numbers across dev machines, dedicated servers, and player PCs, announcing they were migrating servers off Intel altogether.[3]

Different codebases, locked server boards, controlled thermals, same CPUs degrading over months of use. If silicon was physically wearing out rather than merely misconfigured, a future microcode patch wasn’t going to heal chips already in customers’ hands. Intel was looking at RMAs, warranty charges, and datacenter trust damage that equity consensus was treating as forum noise.

What actually moved the stock

Intel closed at $33.37 on July 22 after acknowledging that a microcode algorithm was requesting elevated operating voltages.[4] Ten days later, on August 1, Q2 earnings hit: weak guidance, a 15% headcount reduction, a $10B cost-cut plan, and a suspended dividend sent shares from $29.05 to $21.48 overnight.[8][10]

Raptor Lake defects didn’t cause Intel’s foundry burn, and any honest post-mortem has to separate the two. What the CPU failures did was erode margin and customer goodwill in client computing, the one division supposed to subsidize everything else, right as the broader balance sheet cracked.

Being early looks identical to being wrong

My first batch of puts expired at zero. The technical evidence was already sitting in public bug trackers and server logs, and I priced my expirations as if a company Intel’s size would confirm a hardware defect within weeks.

Instead of walking away when those contracts died, I kept reading every motherboard BIOS advisory and kernel thread. None of the underlying failure data had been explained away; the W680 workstation numbers made “gamer overclocking” impossible to take seriously. I opened a second put position ahead of late July.

On August 5, Intel extended warranty coverage on affected boxed 13th and 14th-gen processors by two years, up to five years total.[5] On September 25, Intel published its root-cause post: Vmin Shift instability in an IA core clock-tree circuit that permanently ages under elevated voltage and temperature.[7] Its 2024 10-K later noted the defect weighed on sales and future warranty reserves.[9]

By year-end, INTC had dropped from $47.80 to $20.05 while QQQ rose from $402.59 to $511.23. Public engineering logs had the story months before investor relations did.