To The Moon

hey everyone, Wonton Investor here. sorry for the radio silence lol, i've been busy with school and internships. but i'm back, and i wanted to start with my best position ever, AMD, which i've held since 2022 and wouldn't buy today.

why i bought it (it wasn't genius)

my reasoning in 2022 was basically "Nvidia is obviously the AI chip winner, but it's expensive and everyone owns it, so i'll buy AMD, the other one". the classic retail move of buying the second-place horse because the first-place one costs too much.

then DeepSeek happened in early 2025. everyone panicked that cheap Chinese AI meant US tech was finished, and i doubled my position instead. my argument was that cheaper AI means people use MORE of it, not less (the Jevons paradox), and that the shift towards inference would favour AMD.

grading my own homework

the anti-panic call was right. it even happened twice more this year - Kimi K3 in July got called "a fresh DeepSeek moment" and markets dipped, then recovered again. and the numbers now back it up: total AI capex guidance for 2026 is around $730B, 158% above what forecasts said two years ago.

but my reason was a bit off. i expected inference to run on CPUs, so AMD's server chips would win. what actually happened is that AMD's Data Center revenue doubled to $6.7B in Q2 2026 (now 58% of the company) because AMD finally had GPUs that big customers like OpenAI, Meta and Anthropic were willing to buy.

and my old mental model of AMD's GPUs, "cheaper but not as strong", was only half right. they were cheaper. but chip for chip, AMD often matched Nvidia, especially on memory (192GB on the MI300X vs 80GB on Nvidia's H100). the weak part was everything around the chip: software that's years behind Nvidia's CUDA, and no answer to Nvidia's rack-scale systems until this year. that's why Helios, AMD's first rack-scale system, matters more than any faster chip.

on my substack i go much deeper on this, including what i completely missed in my DeepSeek post (spoiler: the bottleneck moved from chips to the "plumbing" connecting them) and which AI infrastructure names i've been buying because of it. it's at thewontoninvestor.substack.com.

the earnings gamble

back in early August, before AMD reported, i wrote down that i'd top up into the print as a pure gamble, with separate money and separate rules. but i also wrote that "will they beat?" and "will the stock go up?" are different bets.

AMD beat, raised Q3 guidance to $13B vs $12.5B expected... and fell 7% the next day, because gross margin came in at 54% instead of 56%. and it's not the first time:

my rules for these punts, which i'm keeping:

size it like a lottery ticket

use swing-trading money, never long-term money

no averaging down if it goes wrong

write the exit down before going in

then, because the market loves humbling anyone who writes things down, AMD drifted to the $470s... and ripped through a $1 trillion market cap in the last week of September.

why i hold it, but wouldn't buy it

at 50x forward earnings and now a trillion dollars, the market already agrees with everything i believed in 2022. so i don't add to it as a long-term position.

but i also don't sell, because "would i buy it?" and "would i sell it?" are separate questions. not adding cost me some upside. selling because "it's expensive" would've cost me way more, right before its best month.

so should YOU buy AMD now? i'm not a financial advisor, so i can't tell you. but at a trillion dollars, it doesn't need to be wrong to disappoint, it just needs to be slightly less right than everyone expects.

the full version, with the DeepSeek scorecard, the AI "plumbing" i've been buying, and part 2 on why i think those names are about to get commoditised, is on my substack at https://thewontoninvestor.substack.com/p/everyone-pivoted-to-ai-i-still-hold-6b6?r=g6v1z.

peace out,

The Wonton Investor

if you enjoyed this, follow me here for more. i'm also on X at @thewontoninvest.

disclosure: i hold AMD. not financial advice - just my own reasoning about my own money.