In a week when US equity markets went risk-off, there was some hopeful news on the technology front, pointing to future efficiency gains and, hence, higher unit velocity. There are also reasons to be optimistic about future intelligence, increasing value velocity.

What Got My Attention This Week

• Quote of the week: Elon Musk, "AI is compression and correlation." See the thread here.

• There was a ton of energy data floating around over the last couple of weeks: the incredible investments in China, almost all new energy in the USA is from alternatives, Australia is giving away free electricity to some homeowners during short windows, and more. Energy is the force that drives productivity. For some time, the energy density of fossil fuels was that force. However, my gut feeling is that the world is at an inflection point. Between batteries, alternatives, and increased focus on energy per capita, what has looked like a disaster may be about to turn the corner.

• The other thing that really got my attention was Meta discussing capex going to 40% of revenue. A lot is going on here: protecting themselves from an invasion by OpenAI, using AI to drive their ad business, and the big question: is a company that is not a cloud provider going to get a return on an investment of that size? I would have to dig in a lot more to have an informed view. For now, it got my attention, especially in the context of my initial take on where they are positioned

AI Unit Velocity

NVIDIA has announced that its Dynamo software platform (link) provides 10x improvements in throughput, per watt, per dollar, and, of course, revenue.

OpenAI also released research on smaller, sparse models. While not the first to work on sparsity, and this research has only produced rudimentary capabilities, many will likely be happy to hear OpenAI discuss something other than just scaling laws.

While significant efficiency gains have been achieved, I am optimistic there is much more to come. I am not yet convinced that AI has had its "transistor moment." In an industry that is wafer, chip, and energy constrained, it's reasonable to expect engineers to focus heavily on efficiency gains in 2026. This is especially true for China, which is being guided down that road by having less access to leading chips (though 2026 may also be the year Huawei starts to move closer to top-tier offerings).

AI Value Velocity

This was a week where Google "drops"—and suspected drops—got me and many others excited for the release of Gemini 3.0. There is also growing expectation for Grok 5, alongside a truckload of progress from Chinese open-weight models.

If that wasn't enough to raise expectations, interesting research is also coming out on better RAG (Meta), efficient self-evolving agents (Alibaba), multi-agent scientific research (Tencent et al.), and more.

While I am not one to scream AGI or ASI from the rooftops, better GenAI is coming.

AI Stock Market Velocity

Not a good week for risk-on investments, whether in crypto or high-risk equities. Several items dropped in a similar time period that spiked concerns: the perception that OpenAI was fishing for government backing; "big short" Mike Burry being public about his depreciation bets; spread action on Oracle debt; further analysis of Meta's capex as a percentage of revenue; Coreweave customer news that required explanation; and declining conviction in a December Fed Funds rate cut.

Too big to fail? Too much to swallow in such a short period. And so, the big rotation was on.

In addition, small businesses are in a painful place. As for the consumer, it's a genuine mixed bag: credit card use continues to rise, while credit card delinquency is rising; trucking volumes are down, real estate is in a funk, and more. The coming week will have some more retail news, but the "two Americas" narrative continues to have steam.

See my curated list of news here, along with an AI-generated summary.