Daily Alpha · Substack
· Midday Alpha · by Buzzberg Research
In Tae Kim's interview, d-Matrix CEO Sid Sheth argues AI inference is increasingly constrained by memory bandwidth and capacity.
d-Matrix's Raptor will stack a custom DRAM tile on its SRAM compute tile instead of using HBM
d-Matrix argues inference is bottlenecked by memory bandwidth and capacity, not just compute; its Corsair product is moving to rack-scale GA with Arista, Supermicro and Broadcom, and Raptor co-packages a custom DRAM tile with its SRAM compute tile in a 3D stack.
If d-Matrix executes, it validates a 3D-DRAM inference supply chain anchored by TSMC/Powerchip packaging and Nvidia's NVLink Fusion rack ecosystem.
Watch Raptor's progress toward latency-sensitive inference by end-2027.
Source →QTR flags public companies with billions riding on OpenAI and Anthropic private valuations
The post says some of the world's largest public companies have billions riding on private AI valuations that may not survive public-market scrutiny, as AI stocks come under pressure and circular-financing concerns mount.
If the OpenAI and Anthropic IPOs flop, these stocks could suffer, per the headline.
Watch OpenAI and Anthropic IPO pricing.
Source →Rooster says any LITP retrace should be bought
The newsletter frames pullbacks in the LITP ETF as buying opportunities, without naming entry levels, price targets or a catalyst.
It is a bullish stance on lithium-miner exposure via LITP.
Source →Rooster questions a dollar collapse because EURUSD is about 50% of the DXY and prefers precious metals
The author notes a USD collapse would require EURUSD to rally and asks why, calls the monetary system a 'paper money game' and states a preference for precious metals.
It positions metals as the alternative to fiat rather than giving a specific FX forecast.
Source →