Chamath Palihapitiya

Step 1. Everyone experiments. Step 2. The leaders start commoditizing features and make them much better for themselves. Step 3. Turn the crank on costs and make each capability high margin. As goes Databricks, others will follow. As a result, token consumption will explode but cost per token will crater because these companies won’t tolerate rent seeking behavior from models on performing generalized tasks that can be done by many providers.

If an alien visited earth and asked you: “Show me one chart that sits upstream of everything your species has accomplished or will ever accomplish.” What would you show them? I would show them the chart below. From my perspective, as long as we keep this chart going, we will figure it all out…

Zuck never mentions it because he knows the great engineers want to work at a space/search/ai company… Not an advertising company.

Lol But what will the nuclear plants power then?!

🔥🔥

I’m very excited to see the Ant and OAI S-1s. Here are some of the questions I am very interested in: 1. What is the best guess of the steady state number of tokens to accomplish a fixed benchmark task that actually represents real work in a company. To understand what is worth knowing: fix the real world task and then measure the token consumption and cost across the important combinations of the following: Model + Harness (native and 3rd party) : Task What is it today and considering intelligence scaling, what is our best guess for its descent down the cost curve? 2. If 3rd party harnesses drive token efficiency and most tokens today …

wow - what a miss…

On Control Planes… Four companies shipped an enterprise AI control plane in the last two weeks. They all do one thing well which is govern which model answers a request and what it costs. That’s a real problem and I’m glad it’s getting solved. But what none of them capture is why the software was built that way in the first place, or which requirement a given change was meant to satisfy. Now years later, when an auditor or regulator or senior exec asks for exactly that record, the engineer who made the decision will be gone and the model that wrote the code will have been retired. Now what?? We live this at 8090. At CMS, for example, we…

Yup

Build-vs-Buy Math is Now Squarely in Build For the entirety of the SaaS wave, it was always cheaper and faster to buy software vs build what you needed. But the buying cycle continues to get more complicated. Gartner says the buying group for a complex enterprise software purchase is 6-10 people. 77% of buyers describe their most recent purchase as very complex or difficult. That process exists for well known reasons: someone runs the security review, another negotiates the terms and another signs off on the risk. But because of AI, the “build” side of that comparison has now moved materially. When a “buy” evaluation still takes two or t…

The insurance industry has priced the AI risk. Carriers went to state regulators this year asking to exclude AI-related damages from ordinary general liability policies. 

Regulators approved more than 80% of those requests, and the standard forms behind roughly 82% of American property and casualty coverage now carry a generative AI exclusion that gets attached at renewal. Underwriters are telling you they can't price a loss they can't inspect or audit . This is the same problem all large enterprises have with examiners, regulators or auditors when they ask “Show me what happened and demonstrate that your controls actually worked.” Imagi…

Agreed

50 to 80% of a software system's lifetime cost arrives after launch. At launch, a system integrator's delivery team rolls off to the next engagement, so the people who made design decisions are gone before the expensive years begin, and the support begins. Ask who takes the call three years in, when a regulation changes or the workflow the system was built around no longer exists. The engagement model was designed to end exactly when most of the cost begins. 8090 stays accountable for the code in production after go-live and keeps it current as the business changes. 

Learn more: https://t.co/vIw9cXYzkS

This is a really good development. As model costs converge, the next likely step will be the use of independent 3rd party harnesses/control planes (8090 Software Factory, Palantir, Nous’ Hermes etc). A 3rd party harness + the ability to use any model quasi-interchangeably is something all enterprises will ask for within 36 months as they realize the importance of data/IP leakage and keeping complete sovereignty of their encoded judgement.

It took longer than expected but Perimeter is changing the nature of cancer surgery…

Every year Gallup measures how many people report to a single manager across the American workforce. The number jumped from 11 to 12 in a single year, the fastest move in the decade they have been tracking it. The driver can be heard on every restructuring call: companies are cutting middle management roles and distributing the remaining employees across fewer remaining managers. Gallup's own data says engagement peaks when a manager has eight or nine direct reports. Because that is roughly the number where a person can still give real feedback and know what their people are actually working on. Past fifteen it falls off a cliff. The ques…

Tesla is going all in on nuclear. Actually it’s going all in on solar but they are the same thing - the sun is a nuclear reactor, just farther away. https://t.co/XuuRCJ1oRS

Were the polls actually completely off or did a bunch of mainstream media decide to lie and try to push Wisconsin towards socialism? Seems like it failed regardless.

The market is reducing itself to: Technology people who knows how this stuff works Vs Everyone else who has no idea and so gets short because it violates their non-technology priors

Making progress every week… Try it here: https://t.co/UAxNYUVIps

Retiring spreadsheets in favor of software… 8090 built the system that now runs part and model numbering inside BISSELL’s product development organization. The rules that a handful of expert reviewers once enforced by hand are now written into software, requests that follow them are approved in seconds. Reviewers only give their attention to the requests that need it. Also, because the system is built with the help of our Software Factory, it is well documented, auditable and can change on a dime by simply editing the rules. The system now runs at about half the annual cost of the process it replaced.

There is a bipartisan bill in the House right now that would require federal agencies to produce an inventory of their legacy systems along with a five year plan for each one. Think about what it means that this has to be legislated. These systems were built by a rotating cast of contractors across 30 to 40 years. A map of what was built and why was never even a deliverable. Now the code outlives the reasoning behind the system. That is the real modernization problem, and you do not solve it by signing another contract that produces more code and still no record. Every system an agency runs should be able to explain itself: what it does, …

How Two Companies Built a $71+ Billion Drug Class

In 2025, two drugs earned $71 billion between them, more than double the combined revenue of OpenAI and Anthropic that year. The drugs are semaglutide, which Novo Nordisk sells as Ozempic and Wegovy, and tirzepatide, which Eli Lilly sells as Mounjaro and Zepbound. Tirzepatide became the best-selling pharmaceutical in the world. Though AI revenue is accelerating and the gap is closing, through 2025, two molecules out-earned the two leading frontier AI companies combined, and they did it while reaching fewer than 2% of the roughly 800 million people who would qualify for them globally. Despite mainstream adoption and public testimonials from…

Speaker lineup for this year’s Summit is 🔥

I love my friends. 🥰🥰

This is a no brainer policy

And if reports are accurate, with zero distillation to jumpstart it. Thus proving many things that are more value destructive than value accreting…

Power is THE binding constraint. Data centers are being shut down, GPUs are sold out, models are being commoditized and spot rates are rising all leads to power being critical. Not fanciful plans for power, future forecasts of BTM or distributed batteries blah blah blah but energized power today. This means the following hierarchy is developing from greatest to least value: 1. Hyperscaler 2. Neocloud 3. Model maker Ideally, you are 1+3 (Google, SpaceX, Meta) where you own massive power today and have a leading set of models to keep API pricing from 3rd parties honest enough to benefit them vs the model maker. But even if you are just (1),…

I remember it well. https://t.co/YvK8jzWpKV

Tactical Game Theory: Meta Scorched Earth This should have been Meta’s play two years ago. That said, they are in an even better position to do it now considering the power and compute constraints that are emerging.