# chip earnings — X 热门讨论 (2026-09-17 14:28 UTC)
## @Straits_Trader (Straits_Trader) · 09-16 06:52 · ♥303 ↻3 💬2 ⚠️SMIC 0981.HK has bounced back. Is it still worth buying? What should we focus on? https://x.com/Straits_Trader/status/2100115606502547461
## @Straits_Trader2 (Alt of@Straits_Trader) · 09-16 06:52 · ♥261 ↻2 💬2 ⚠️SMIC 0981.HK has bounced back. Is it still worth buying? What should we focus on? https://x.com/Straits_Trader2/status/2100115706440187987
## @ShanuMathew93 (Shanu Mathew) · 09-17 10:38 · ♥34 ↻4 💬2 Ok I did recaps for Jensen, Elon/Gwynne, and Satya on @theallinpod. Have to say, @altcap was my personal favorite but b/c he talked about compute & markets so I'm obvi biased there :). The presentation is a must watch and his framing on market activity and setup from here was spot on.
>Chip suppliers have captured much of the upside. Hyperscaler capex has surged alongside semiconductor free cash flow, a relationship he describes as nearly dollar-for-dollar in his chart. Semis account for 70% of Nasdaq returns; the companies spending on AI have seen much less of the benefit.
>Earnings have outgrown share prices, compressing multiples. He cites earnings growth of 26% and Nvidia at 14× next-year fully taxed GAAP earnings. His argument against a bubble depends on those earnings proving durable.
>The required revenue ramp for the labs is steep. He estimates combined annualized lab revenue around $100B exiting July and wants at least $180B by year-end to sustain the trade. His longer-term illustration runs from roughly $200B to $450B, then $800B–$1T, against annual capex approaching $1.5T. Customers ultimately need enough revenue to pay for the infrastructure (e.g., someone has to pay for the compute! can't rent expensive compute and not generate return off that expense)
>He's less bullish on deployment than the 43 GW 2027 forecast he attributes to SemiAnalysis. His estimate is closer to 25 GW because of power, permitting, grid connections and labor (I've been consistent on here I think mid-teens to 20 this year, 20-30 bull case type of figures). Roughly half would go to Anthropic and OpenAI, which he still thinks can support the required revenue growth (I've also made this prior point, most of the meteoric AI $ rise we saw came at a time of 1 to several GW. They're moving into 5-10GW soon so should have plenty of compute to grow biz against the near infinite TAM they market otherwise it is definitional deceleration)
>His portfolio exposure is medium. Stronger lab revenue and lower oil would bring more exposure; weaker evidence could mean less. Higher rates, regulation and a potential Anthropic IPO delay remain risks. Owning AI was enough in 2023–25. From here, the revenue, deployment and financing assumptions have to hold. https://x.com/ShanuMathew93/status/2100534746681938176