Steve Hou
87.1K posts

Steve Hou
@stevehou
Research @Bloomberg, opinions my own. @UMICH/@ETH_en/@UVA alum. Perpetually curious, but also "incredibly unsophisticated" (according to Chamath Palihapitiya)
New York, USA Katılım Kasım 2016
4.7K Takip Edilen50.8K Takipçiler
Steve Hou retweetledi

Perhaps even more notable than Anthropic's $44B ARR figure is the reported 70% gross margins on inference.
If you do some back of the napkin math and assume:
- 75% of total rev. comes from AWS, GCP marketplaces for which Anthropic pays 25% rev. share ($8.25B total rev. share costs)
- 2.5GW of total compute at $12B per GW ($30B total), with $13.2B on inference (to back into 70% gross margins) and $16.8B on training
- 5,000 FTEs at $0.75MM in annual compensation per head ($3.75B total)
- $1B of other OpEx (e.g. marketing, office rent, finance, legal etc.)
... Then total run-rate COGS + OpEx is $43B, implying $1B of EBITDA.

English

@madridraptor Yes, blood pressure, blood test, urine test, some vaccination, plus a whole conversation with the GP. It’s a lot
English

@stevehou What kind of checkup? Like the ones you do every year in China? 全身体检?
English
Steve Hou retweetledi

One of my favorite of our recent blog posts: @cerebras made GPT‑5.3‑Codex‑Spark so fast that we had to rethink how Codex uses the Responses API – leading us to build WebSocket support for ultra fast latency.
Cerebras speed is just 🤯
OpenAI Developers@OpenAIDevs
⚙️ We made agent loops faster with WebSockets in the Responses API As Codex got faster, the bottleneck moved from inference to inefficient API calls WebSockets keep response state warm across tool calls, helping workflows run up to 40% faster end to end openai.com/index/speeding…
English

@xeophon @infornomics What are the top3 differences and similarities?
English

@jaybee484 @Noahpinion @TheStalwart @ezraklein @AlexThomp Just gotta be more pragmatic and do things faster. I think Americans care a lot more about procedures.
English

@stevehou @Noahpinion @TheStalwart @ezraklein @AlexThomp Sounds like a really interesting case study then 🙂 I think One Medical was trying for this efficiency lane but was struggling before Amazon bought them
English

@jaybee484 @Noahpinion @TheStalwart @ezraklein @AlexThomp Insurance. They checked everything incl what’s covered not covered. Super experienced from other patients presumably.
English

@stevehou @Noahpinion @TheStalwart @ezraklein @AlexThomp Do they take insurance or do you pay cash? A bunch of our red tape has to do with insurance
English

It’s an immigration medical exam but I was also due for a regular annual checkup. I asked them whether it’s possible to combine the two exams so I could save a trip. I was fully expecting that they say no since that’s what I’d expect from an US hospital. They went sure no problem. That is Asian pragmatism and how you get abundance.
English

@Noahpinion @TheStalwart @ezraklein @AlexThomp how about that for “abundance”? Just be more efficient all around? Fewer formalities, processes, red tapes. I straight up was shocked myself as I had gotten used to US medical system inefficiency.
English

@Investor_NICK_ @WarrenPies Recessions are rare bc they are by definition tail events
English

@stevehou @WarrenPies They’re probably rare because our government and the Fed makes them rare. But it gives us the best situation. Capitalism without risk of recessions is essentially heaven without the possibility of hell. Probably best for most folks because recessions really do suck.
English

Yup estimates of earnings esp at the index level are rarely off by a couple percents.
Warren Pies@WarrenPies
Common pushback: "Estimates aren't earnings" Historically, though, analysts rarely miss on estimates outside of recession...With so much focus on guidance, analyst accuracy is only increasing in recent years...Odds are that companies BEAT these estimates.
English

@stevehou @WarrenPies Outside of recession … so essentially they never even consider a recession and revise earnings only after the recession has already hit. Damn. Now I understand why they say no one ever sees a recession coming.
English

DAU growth curve for generative AI apps is flat. What's happening? bigtechnology.com/p/are-ais-cons…
English





