Daily Alpha · Substack
· Post-Market Alpha · by Buzzberg Research
Substack signal centered on whether sparse attention changes AI memory demand and on retail positioning in metals and US equities.
Themes on this desk
Sparse attention memory
SemiAnalysis's GLM-5.3 sparse attention item addresses HBM memory usage and DRAM/NAND implications.
Metals hold
Rooster refuses to sell gold/silver despite an ugly chart, citing a 6% term deposit as the alternative.
US ETF allocation
Rooster is adding a final 10% ($24k) into a US ETF portfolio.
SemiAnalysis: GLM-5.3 sparse attention and HBM memory usage
The preview headline examines how GLM-5.3 sparse attention affects HBM memory usage, listing KV cache offloading, HiSparse, DeepSeek sparse attention, IndexShare, and asynchronous optimization as topics.
Sparse-attention efficiency changes could alter HBM/DRAM/NAND demand intensity from AI inference, a key variable for memory suppliers and AI accelerator economics.
Watch Whether the full analysis concludes sparse attention reduces or preserves HBM demand; invalidation of the demand thesis if memory usage per model falls materially.
Source →Rooster deploying remaining 10% ($24k) into US ETF portfolio
The author states they have $24k left, equal to 10% of the portfolio, and are buying into a US ETF portfolio.
A concrete sizing disclosure: 10% of the portfolio is being added to US equity ETF exposure, implying a bullish incremental allocation to US equities.
Watch Whether the full 10% is deployed and which US ETFs are named; invalidation if the author reverses or holds the cash instead.
Source →Rooster refuses to sell gold/silver despite ugly chart
The author states that although the gold/silver setup 'looks shit,' they would not sell, and repeatedly cites a 6% term deposit as the alternative benchmark.
Signals a holder refusing to capitulate on precious metals and anchoring to a 6% cash/term-deposit hurdle, implying the author sees metals upside exceeding that yield.
Watch Whether gold/silver recovers from the 'looks shit' level; invalidation if the author exits or abandons the 6% term-deposit comparison.
Source →