The dot-com boom ran on debt. The AI boom runs on balance sheets. That difference matters. Good take by Goldman Sachs.

AI is reshaping digital advertising by unleashing a wave of creative supply.

Teams can now generate thousands of ad variants quickly and cheaply. Targeting is getting smarter as Meta, Google, and others push more compute into the ad stack to serve the right message to the right person at the right moment. Feedback loops are tightening: every impression and click trains the system, compounding performance gains. Gen AI boosts the volume and variety of creative, while GPU-accelerated models sharpen targeting and timing. The net result is more relevant ads and higher conversion rates.

Where could this go next? Advertisers will generate far more creative, feeding models that continuously personalize and optimize. Conversion rates should increase meaningfully as models learn which creative works for whom and when. We expect revenue growth to follow — better matches drive more sales.

As AI transforms digital advertising, it’s also reshaping how consumers discover and buy products online. We believe this shift positions e-commerce as one of the major beneficiaries of the AI revolution.

For more on this C:\Take, watch Michael:

The GPU is eating software.

We noticed this chart from Apollo is getting airtime...

We call it chart crime! Zero-return decade was unique to .com.

Of course, cherry-picking from one crisis to the next looks terrible.

Sports betting is just the beginning for prediction markets...

The king is lagging the kingdom...

No "Code Red" here! ChatGPT traffic historically dips this time of year.

The AI revolution will require a massive build-out of new infrastructure.

That means more chips, more data centers, more electrical equipment… and far more power. As businesses continue to grasp AI’s potential, we continue to see more and more announcements with facilities now measured in gigawatts (GWs).

OpenAI alone has announced partnerships with the semiconductor world representing nearly 30GW of new data center demand, and most believe we’re still in the early innings of the build-out. The question is, where will all of that power come from?

We see three options:

• Build new generation. That's already happening. But given the explosion of demand, if you want a new gas turbine, you won’t see it until after 2029. New nuclear will take even longer, extending into the 2030s. And renewables, while faster to install, are not able to provide the primary power required for data centers given their intermittent nature.

• Contract with a regulated utility. Regulated utilities have predetermined Return on Equity they can earn. Generally, working with a regulated utility means longer interconnection studies, potential political pushback, and strict regulatory guidelines that dictate who can connect and when.

• Partner with an independent power producer (IPP). Over the past 12 months, many hyperscalers have turned to IPPs, which own generation assets, set their own electricity prices, and can often move faster than regulated utilities. We are seeing hyperscalers willing to pay 3-4x the regulated or market rate for expedited access to these large energy resources.

The universe of IPPs is not very large, and we are seeing in early agreements that many are choosing to go nuclear. Nuclear generation is baseload energy (24/7), clean, and produces a lot of power in one location. Amazon announced a transaction with Talen Energy in March 2024; Microsoft partnered with Constellation Energy in September 2024 to bring the Crane Clean Energy Center (formerly Three Mile Island) back online; and as recently as this October, NextEra Energy unveiled plans to restart its Duane Arnold nuclear plant in conjunction with Google.

For more on this C:\Take, watch Max:

Markets repriced fast!

When private multiples are lofty, the right question isn't "Is the price high?" but "Is the growth durable?"

The chart above shows adoption and usage growth for two code-first models:

• Claude Code (Anthropic) —> 30x growth in NPM downloads over the past five months

• Codex (OpenAI) —> 2x total installs a month after GPT-5 launch

The steep usage curves indicate strong product-market fit for AI coding assistants and agents. By driving 20–40% in developer productivity, these tools justify premium pricing and accelerate seat expansion. The growth reflects compounding improvements in model capability and productization — reasoning, longer task length, agentic workflows — that are already translating into willingness to pay.

If revenue compounds at high rates, today’s 20–40x forward revenue multiples could compress quickly. For example, sustained 70–100%+ growth with improving gross margins can drive a seemingly expensive entry multiple into the single digits within just a few years.

So when we’re asked whether private markets are too expensive, we focus instead on growth and durability. Through that lens, select opportunities with sustained adoption curves can be attractive, even at elevated initial valuations.

For more on this C:\Take, watch Lucas:

Major hyperscalers like Alphabet and Meta boosted their capex forecasts alongside their latest earnings reports, once again surpassing Wall Street’s already lofty expectations.

For much of the year, sell-side analysts have been playing catch-up, repeatedly raising their AI infrastructure spending projections with each new company update. Following this latest round of upward revisions, consensus capex forecasts for 2026 now stand at ~$518 billion, up 29% year-over-year - and an astonishing 65% above what analysts had expected at the beginning of the year.

These new spending plans underscore not only the sustained momentum behind the AI buildout, but also how much investors and analysts may still be underestimating the true scale and staying power of the overall AI investment cycle.