.@chamath on the pricing shift he thinks will make the AI market explode: “When Google invented and scaled AdWords and AdSense, the market exploded… because we went from buying 1,000 ads to just getting an outcome. Right now we buy a million tokens. We will eventually move to an outcome. And I think when that happens, this market will just go absolutely crazy.”

The two constraints on AI are effectively electricity and chips. China and the U.S. have different advantages regarding both: - China wins on raw electric power (added 543 GW of new capacity last year), but their ability to produce advanced chips is still a few years behind. - The U.S. is ahead on chips, dependent on an international supply chain. NVIDIA and AMD design leading processors, while Taiwan’s TSMC manufactures many of them. This means access to Taiwanese factories is a critical part of American AI leadership. The bottleneck for America is clear: we are short on electricity.

In March 2026, Anthropic accounted for 69.5% of business AI token spend on Ramp, more than ~2x OpenAI's share. The gap has since narrowed, with Anthropic at 51.0% and OpenAI at 44.5%.

According to ARK, the cost per mile of a personally owned vehicle has remained at ~$1.10 for 90 years (adjusted for inflation). By 2035, autonomous taxis could reduce that cost by 77%.

.@chamath on superintelligence being on the scale of the Industrial Revolution: “[The President] says that whoever wins superintelligence wins. When you think about it, he's completely right. There is no world in which you're going to have this multinational corpus of people trying to allocate wins and losses on this thing. This is on the order of the Industrial Revolution."

.@chamath on why AI job doomerism is a political strategy: "There's an incredibly important incentive for folks that are not in the power structure today, i.e. Democrats, to take up the doomerism call to action, because it is a very straightforward way to whip up fear and then catalyze that to power in the midterms and then in '28." "All the data up until now tells us there will be more jobs and more to do. More work to be done, more judgment, more taste to be imposed. This narrative does not come from a place of data-driven fact. It comes from a political calculus that doomerism is a way to foment distrust and displeasure and then use that at the polls."

Microsoft CEO Satya Nadella wrote that a company renting an AI model “essentially pays for intelligence twice”: Once in money, and again in the proprietary knowledge it has to reveal to make the model useful. This mechanism is called the learning loop: every prompt and correction captures some of the company's know-how. That knowledge is distilled and, with a closed model, the provider keeps that information. If your usage data is worth more than the token cost to serve, providers might end up paying you to use their models. The precedent exists: OpenAI created an opt-in program that gives accounts up to 1M free tokens a day on certain models in exchange for training rights on that traffic. Chips can be rented and algorithms can be leaked or distilled, but the prompts and corrections flowing through a model compound privately for whoever holds them.

Slowing aging by just one year would add ~$400B per year to U.S. GDP. Slowing it by five years would add $2T - more than twice the annual U.S. defense budget. In 2024, the U.S. federal government spent roughly $2.6T on Social Security and Medicare (about 38% of total federal spending), and most of it was concentrated on older adults 65+. By 2029, this spending will grow to $3T every year. This structural and fiscal pressure of aging creates economic opportunity where even a small shift in biological aging could be transformative.

A timeless rule of competition: commoditize your complement. Every product has complements. When a complement gets cheaper, demand for your product can rise. So companies often try to drive the price of key complements toward zero. Open-weighting a model is basically this move at a frontier scale. OpenAI and Anthropic position the AI model itself as the product you pay for. For Meta, dependence on a model layer controlled by competitors would threaten its ability to profit from its own AI products. So Meta gave Llama for free. They can afford to give away the model because they don’t have to sell it. Instead, they sell advertising to 3B users. The model is Meta's complement, while its data and distribution are the moat. Once you start to see the pattern, other open-model releases become more clear: - Google (Gemma) and Microsoft (Phi) sell cloud computing and software. A cheap model layer sends more workloads to their clouds. - OpenAI (gpt-oss): When open competitors set the price, you release something open too. - Nvidia agreed to acquire Hugging Face for $12.9B, preserving openness and support for competing hardware and cloud providers. Making models more accessible encourages more applications and greater usage, which expands demand for computing infrastructure. Essentially, companies open up one layer when doing so strengthens the part of the stack where they make money. For companies whose only business is selling access to closed frontier models, cheaper and more capable open-weight models mean more competition at the model layer and more pressure on token pricing.

Capex has exploded over the last ~2 years. It is incorrect to look at this trend and assume that FCF has collapsed because operating cash flow has collapsed. The hyperscalers are deep into an investment cycle, so they are consuming a significant portion of their operating cash flow.