Sponsored by
Hello readers,
Welcome to the AI For All newsletter! Today, we’re talking about how AI is helping develop superbug-smashing disinfectants, agentic AI in the physical world, and more!
AI in Action: The AI killing machine that could save your life
Bacteria are winning. The same antimicrobial compounds that have scrubbed kitchen counters and hospital floors for over a century — called quaternary ammonium compounds, or QACs — are losing their edge as resistant "superbugs" evolve faster than chemists can respond. Designing new versions is painstaking work: build a molecule, synthesize it in the lab, test it against pathogens, repeat. A team of researchers at Emory University, George Mason, and Villanova decided to let AI take the first swing — and in doing so, produced what they believe is the first example of AI-generated disinfectant discovery.
The problem was that AI had no good map of this chemical territory. QACs have a distinctive shape — a positively charged core with multiple hydrophobic "tails" that essentially spear through a bacterium's membrane and cause it to fall apart — but that structure is rare enough in existing molecular databases that standard AI models couldn't learn it. The team built a custom model from scratch, training it on just 603 hand-tested compounds from their own labs. Their first attempt generated hundreds of candidate molecules and handed them to an expert chemist for review. The verdict: 9% were worth making, and 21% weren't even valid chemical structures. Good proof of concept, not yet a practical tool.
So they added a filter. Before any molecule reached the chemist's desk, a second AI model screened it for predicted antibacterial activity. That one change transformed the results: synthesis-worthy candidates jumped from 9% to 38%, and invalid outputs dropped to zero. From the 29 most promising molecules they actually built and tested in the lab, 11 turned out to be genuinely novel disinfectants with confirmed activity against multiple dangerous bacterial strains — including one that worked against all seven strains tested, gram-negative bacteria included. Those are the hardest to kill. The private sector has taken notice, and the team is now running more AI-generated candidates through the pipeline, with undergraduates doing the synthesis work.
🔥 Rapid Fire
Analysis: Am I Meant To Be Impressed?
Microsoft, Amazon, Google, and Meta are expected to spend up to $900 billion in AI capex in 2026 and over $1 trillion in 2027
Microsoft and Amazon disclosed their annualized AI revenues
Microsoft: $37 billion revenue run rate ($3.08 billion per month)
Amazon: $15 billion revenue run rate ($1.25 billion per month)
To be clear, this is revenue, not profit
Microsoft’s AI revenue is 1.05% of its $294B in AI capex so far
Amazon’s AI revenue is 0.42% of its $298B in AI capex so far
To be clear, these numbers are bad
Google and Meta still refuse to disclose their AI revenues
But wait, it gets worse: Microsoft and Amazon’s AI revenues are heavily dependent on circular financing — Google’s too most likely — with two companies: OpenAI and Anthropic
OpenAI represents an estimated 70% of Microsoft’s AI revenue and 80% of Microsoft’s AI compute capacity
Anthropic represents an estimated 80% of Amazon’s AI revenue and 75% of Amazon’s AI compute capacity
OpenAI and Anthropic represent 48% ($748 billion) of Microsoft, Amazon, and Google’s revenue backlog (all revenue, not just AI)
To be clear, 48% of Microsoft, Amazon, and Google’s future revenues are meant to come from two companies that cannot and will never be able to pay them $748 billion, so Microsoft, Amazon, and Google must continually invest in these two companies to A) keep them alive and B) feed money back to themselves — this is unsustainable
Analysis: The AI Compute Demand Story Is A Lie
The “demand” for AI compute is an illusion created by A) circular financing and B) capacity constraints that are the result of slow data center construction and OpenAI and Anthropic taking up 70% of compute
OpenAI and Anthropic make up 85% of AI compute spending
Big Tech needs $3 trillion in new AI revenue by the end of 2030
NVIDIA, Google, and AMD have had to backstop multiple data centers
Like NVIDIA, Google is creating SPVs to sell TPUs to itself
GPU depreciation is eating into Microsoft, Google, and Meta’s profits
Depreciation is projected to eat into 38% to 58% of profits by 2030
Banks seek to offload risk to avoid ‘choking’ on data center debt
Rising AI costs are becoming a problem for even investors
Write docs 4x faster. Without hating every second.
Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.
Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.
Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.
📖 What We’re Reading
The idea of an AI agent that doesn't just detect a supply chain bottleneck but actively re-routes logistics, or an agent that doesn't just flag a temperature spike but actively recalibrates the machine, represents the holy grail of automation.
But as we move these agents from the digital sandbox to the physical world of the Internet of Things (IoT), the risk profile changes appropriately. A hallucination in a chatbot is embarrassing; a hallucination in a predictive maintenance system can be catastrophic.
