Market Thoughts (and Global Strategist Worries), Parsing META News, Updating our Tech HW Model, Options Market Tea Leaves

Research evidence excerpt

Market Thoughts (and Global Strategist Worries), Parsing META News, Updating our Tech HW Model, Options Market Tea Leaves

ort – if even true - is about monetizing legacy compute (as FundaAI suggests), selling access

to Spark and its other models, increasing scale to lock up compute and/or reduce costs, or more ominously an admission that –

only a few months after talking up their model roadmap, signing LTAs with memory makers and reducing headcount (while

hiring for compute) – they’re now improbably saddled with excess compute. The company hasn’t commented at all on the article,

but my conversations yesterday revealed a strong divide between generalists & internet folks (who view this as META being

a good steward of capital/giving itself an out, i.e. incrementally -ve for the AI capex story) and semi/hardware folks like me

(who view it merely as a de-risking of cloud concerns, perhaps – as Jeff Pu argues - also making it easier to raise capital to keep

spending).

I also want to re-flag that JPM’s options traders on Tuesday shared an interesting recap of the quarter, underscoring that

2Q was a strange spot up/vol up environment (e.g. stocks and options got more expensive simultaneously – stocks rallied, but fear

rose…reflecting an insatiable eagerness for AI upside); long vega was the winning trade. For most of the past year, investors have

bought vol in semis/memory (NVDA, MU, MRVL, SMH) and sold in hyperscalers (MSFT, GOOGL, AMZN, META)

to fund those bets. The market has already priced a lot of AI optimism into long-dated options, which are now expensive. Equities

are making big moves day to day — but not in one clean direction — so short-dated options are paying off if you actively trade

around them; and hyperscale vol also looks interesting to own.