Tech Executive and Analyst. Founder of FourWeekMBA, The Business Engineer & The AI Supercycle
London, England
- Tweets 23,686
- Following 4,111
- Followers 5,916
- Likes 27,624
Pinned Tweet
Top 20 Mental Models For Business, by The Business Engineer
businessengineer.ai/p/the-bu…
1
1
6,764
Microsoft and Hugging Face's ThinkingBox grades AI agents on the database they leave behind, 20 runs per task. The post says Claude Opus 5.5 and Opus 5 both pass 241 of 507 tasks every time, at $7.80 and $13.3…
fourweekmba.com/ai-microsoft…
$MSFT @Microsoft @huggingface #AIAgents #AI
29
A Moonshots clip says Sonnet 5.5 beats Opus 5.5 on Terminal-Bench 4.0 at half the price. Anthropic's own table confirms 70.6% vs 66.4%, and shows Opus 5.5 ahead on the other seven rows (our count). Input and output prices are half; cache reads are the same $0.2…
@AnthropicAI #AI
49
Chase on take-or-pay in AI compute: commit to 100 MW and you pay for 100 MW whether you use it — and Crusoe’s GPU rental agreements are typically take-or-pay.
@HarryStebbings @ChaseLochmiller
20VC with Harry Stebbings — Who REALLY makes money from AI & the future of Data Centers | Crusoe CEO
piped.video/watch?v=Ko7nU1Im…
1
51
How do you quickly find what’s the next form factor for AI? You open source the firmware so anyone can experiment with the hardware. Cool!
🚨 side project alert 🚨
Announcing Muse Gadgets, an open source ESP32 firmware and Linux sdk so that you can make hardware devices that work with Muse.
Grab an API token from gadgets.muse.ai and point your favorite coding agent at the github repo to build your own peripherals for Muse.
57
Mistral says it is opening a Munich hub for Physics AI and will build one gigawatt of European compute capacity by 2030. The post gives no cost, location or financing for the gigawatt. Emmi AI's own May announcement says its team of more than 30…
@MistralAI #AIInfrastructure #AI
56
Cool!
I've wanted this for so long! Scan any menu and see your options right in front of you.
Insane how quick this was to make with CLAD and @specs. 🤯
3
152
Today wasn't 20 headlines. It was one story: the AI stack is being carved into permanent layers — and every company, big or small, is betting its survival on which layer it owns.
1
1
1
78
The AI Supercycle is accelerating the sorting. The companies that survive this phase won't be the ones who built the best model. They'll be the ones who made it structurally hard to leave. Full breakdown — link below.
1
1
10
The full frameworks behind today's moves:
businessengineer.ai/p/the-ma…
#AI
10
jensen just cited five sources for AI-factory ROI. ornn puts A100 returns at 80%. headlines say capex is vindicated.
it isn't. it's a menu.
nvidia picked numbers like a landlord quoting only the best lease. caveats, engine labels, comparisons — left out.
you're not reading returns. you're reading a prospectus.
businessengineer.ai/p/beyond…
NVIDIA's AI-factory returns post cites SemiAnalysis, Sprout, Barkr, Silicon Data and Ornn. We opened three. SemiAnalysis says the engine label is essential when quoting the gain. Ornn's 80% for A100 five-year rental sit…
fourweekmba.com/ai-nvidia-ai…
$NVDA @nvidia #Semiconductors #AI
67
Aleph Alpha open-sourced Kolibri, a 78B-parameter English-German model, under Apache 2.0 and says it supports 1M tokens of context. Its own model table gives the longest trained length as 262,144; 1M needs a serving override. Benchmarks are its own harnesses; we verified no…
#AI
50
The Pentagon's FY2027 budget justification asks for $6.421M for 'Polygraph Next', adding AI/ML scoring and non-contact sensing to polygraph testing. The five requested years sum to $30.3M (our arithmetic). The contract method and performer are listed as TBD. The agency's ow…
#AI
1
45
Crusoe's CEO says on 20VC that prices for using Hoppers are now higher than three years ago when the chips were new. A spoken claim, transcribed by us from a clip. The only independent data we found covers six months of 2026: Ornn's spot index up 70% to 111%…
#Semiconductors #AI
59