Alex Kantrowitz

Alex Kantrowitz reads economist Steve Howe's chart, which puts Meta ahead on ad revenue in 2028 by a good margin after Google led in 2024, because Google has not figured out how to target ads well within AI mode while Meta pulls away on ad revenue despite lagging on large language models. He is not sure everyone assumes Google will win the AI race. Google has the number one model on the benchmarks, a new model called carbon released the same week.

Howe wrote that everyone assumes Google will win the AI race, a complete 180 from a year ago, and left open whether winning AI is commercially good for Google. Amazon has gone from a distant third to a company challenging Google for second, althou

Ethan Mollick says you do not get organizational productivity gains instantly from individual ones. A person can produce 100 times more code than before, which is great for a solo entrepreneur. Inside a company running two-week sprints, coordinating with marketing, and holding daily stand-ups, that 100 times gain becomes a barrier. Nobody has time to deal with all of that code.

Shipping stats show the same gap: 20 times more code produced, and 30 percent more products shipped. Organizations have not adapted to AI yet. Alex Kantrowitz cites App Store figures of more apps and fewer reviews. A flood of novels, games, and apps is one pattern. Studies of products shipped show a huge increase in

Ethan Mollick says AI token prices face real downward pressure because a huge move for token efficiency is underway and costs will decrease even as more compute comes online. Lower cost means AI does more stuff for less money in a very short period of time, so long-term token costs do not worry him.

A chart from Andreessen Horowitz compares the price of compute with the price of tokens. Compute has stayed steady and risen a little, while the price of tokens is falling dramatically. Alex Kantrowitz calls the drop the most deflationary technology he can remember in the price paid for the technology. Use is still rising because agent-style work requires more tokens, and a Fable class model can

Sundar Pichai says Google partnered with Anthropic because, very early, the founding team looked world-class and had been involved in some of the earliest GPT models. Google could tell quickly that they would be at the frontier. The habit is to identify companies at the cutting edge of technology and partner when the chance is there. SpaceX is the other example he names. Some of those partnerships also took shape around Google's TPUs.

Alex Kantrowitz puts Google's commitment at tens of billions of dollars invested. Pichai does not discuss that figure. He describes the decision as recognizing a frontier team and taking the chance to partner.

Sundar Pichai says Google's cloud business is already giving up short-term revenue, and that the industry is doing some form of the same trade. At any moment the cloud business forgoes that revenue because Google makes compute trade-offs to support frontier AI development. He says that choice is already in place. Alex Kantrowitz frames it as giving researchers the biggest computer even when it hurts today's cloud profits, and he recalls Pichai once calling AI as profound as fire and electricity.

Pichai does not treat the trade as a blank check. Constraints push creativity and more efficient algorithms, including ways to use less memory. What Google is committed to is investing what is neede

Sundar Pichai says Google's ground rule is to invest in enough compute to stay at the AI frontier, and that priority is non-negotiable. In 2015 he told the company to be AI first, so AI runs across search, cloud, and Waymo. Google DeepMind gets cutting-edge compute access for research and frontier model development before other plans.

Each business works from its own long-range plan. The cloud team brings projections, demand signals, and committed contracts, and those signals shape a long-term compute path. Google builds its own compute and its own data centers. Lead times are long, so the compute on hand today comes from decisions he made in 2024. Google and other companies are short, and

Kevin Roose says that during the five days after Sam Altman was fired, OpenAI's board looked for a new chief executive and passed on two options. Alexander Wang, whom Roose describes as Altman's former roommate, told the board he was interested in the job, according to Roose's sources. Alex Kantrowitz notes that Wang now runs Meta's Superintelligence Lab.

The board also held a middle-of-the-night call with Dario Amodei and Daniela Amodei, the chief executive and president of Anthropic. They had left OpenAI about three years earlier, in Roose's account, because they could not trust Altman. Each side was sounding the other out. One idea on the table was merging OpenAI and Anthropic under Dari

Kevin Roose says that if AI starts improving itself, new models do not have to wait on human researchers, and he thinks the labs should stop before that loop closes. He believes recursive self-improvement is possible, the labs are racing to do it, and it is dangerous.

Alex Kantrowitz cites an Anthropic figure that Claude is now the lead on 26 percent of Anthropic's AI research and development projects, not a team member, a pair programmer, or a coding assistant. OpenAI has put out versions of the same kind of disclosure. Roose sees no reason the share would not rise to 50 percent, 75 percent, or someday 100 percent. OpenAI's goal is to build an automated AI researcher. If the labs close the

Kevin Roose says Dario Amodei and his six co-founders left OpenAI because Sam Altman kept telling them one thing and someone else something different, or making a promise he did not follow through on. There was no single inciting incident. The pattern runs back to the founding days of OpenAI.

Altman is very good at reading the room and understanding what people want him to do in a given moment. That skill is a huge asset when he raises money from Masa Son and other investors. It let OpenAI scale quickly and get back in the good graces of the Trump administration by saying things that administration wanted to hear. The same habit is a superpower, a liability, or something in between, dependi

Dylan Patel says most neoclouds have terrible security, and open-source models can hack one or reach other customers' data on it. He points to GLM 5.3, a Chinese open-source model, as especially good at that. SemiAnalysis published a report he calls Most Neoclouds Suck at Security. Alex Kantrowitz read an Ilya Sutskever post warning that neoclouds have limited cybersecurity, and Patel says the post echoed that report.

On one cluster, the firm accidentally saw other clients' storage and Slurm jobs, including work for intelligence agencies of a non-American country in the top ten by GDP. They reported it. A malicious customer on that shared cluster could have seen what those clients were trai

Dylan Patel says a trillion dollars spent today on an AI data center and GPUs needs about $1.5 trillion of revenue over six years if payback is a straight line, about $250 billion a year. The six-year clock is the accounting life of the assets. The buildings may last longer than the GPUs. Early years can bring in less and later years more, as long as the total pays the project back.

Patel calls published estimates of the required revenue too low for the pace of the build-out, and he expects revenue to come in higher than those forecasts by 2031. Alex Kantrowitz cited a Bain estimate that hyperscalers need $6 trillion a year by 2031 to break even. Meta can stop buying new capacity and still