CM #8: Did Google just win the AI race?

Why Google is likely to pull ahead with Gemini 3

It’s been a good month for the Caffeinated Memos series. Last month’s write-up BKSY 0.00%↑ is trading at $31 from initial coverage at $21, up ~50%.

This month, there’s a time-sensitive catalyst on a more familiar name, GOOGL 0.00%↑ Google is expected to release Gemini 3 by the end of October. Based on early indications and rumors in the AI community, Gemini 3 may not only cement Google’s lead in the LLM wars but could be a significant step change in AI capabilities. Additional hype around Gemini 3 could send Google past the $300 mark by December.

From AI loser to AI winner

Reviewing the chart of Google over the past two years, one may notice a clear “character change.” From Nov 2024 to April 2025, Google was flat, despite growing revenue 10%+ and earnings 30%+, the same performance that investors are used to from this long-time compounder.

Investors were initially worried about Google’s profitable search business transitioning to the AI era. Major websites have seen visits fall YoY including Forbes (-55%), Business Insider (-38%), and the Washington Post (-33%). Apple’s Eddie Cue said in May Google’s safari searches had fallen YoY.

A few notable events have changed the narrative:

1. Google’s earnings in July showed search revenue up 12% YoY

2. Waymo started expanding outside of San Francisco. Waymo does 250K+ trips a week, in 15+ cities, and has started proving that Waymo’s are safer than human drivers

3. Google not only avoided a breakup in the antitrust case, but was allowed to keep most of its existing distribution deals. On the antitrust case, Google was actually saved by the potential AI threat to its business.

Since Nov 2024, Google has became the leader in the AI race and Gemini reached 400M+ monthly active users.

Google’s initial attempt to compete with ChatGPT- “Bard” failed to compel investors or gain traction. Google falling behind OpenAI in the model race always seemed odd. It’s the same kind of odd when Arch Manning was bad at football. You know Arch couldn’t possibly be as bad as he was playing earlier this season. There’s been some reporting on why Bard was so bad. Google’s AI talent was split in two (Brain/Deepmind). Google’s management might have been afraid of releasing something too good (innovators dilemma).

Whatever the reason for the initial struggle, Google has all of the advantages. Near-infinite cash flow, ($134B OCF), one of the deepest benches of AI researchers (even without Ilya), a custom ASIC (TPUs), tons of training (YouTube), distribution (Chrome/Search), hyperscaler know-how (Google Cloud).

Since Bard, Google combined all its AI teams into one, released the Gemini series of models. Even Sergey returned in pursuit of building a “god” model.

It’s been showing in the data. On the LLM arena, which uses blind tests of real users, Gemini 2.5 models are #1 in text, #1 vision, and #4 in WebDev. Keep in mind Gemini 2.5 pro was released in March and is beating out OpenAI and Anthropic models from Aug/Sep respectively. So Google has a 4-5 month lead here for Gemini 3.

The market is unsurprisingly giving Google a 73% chance of finishing the year with the best model Yet investors may still be surprised by just how good Gemini 3 is.

Gemini 3: The Next Leap Forward

Gemini 3 is expected to release in the next few weeks. An intern supposedly leaked that Oct 22nd is the target launch.

In the last week, Google has rolled out beta testing for Gemini 3. While we don’t have benchmarks, there are limited samples of Gemini 3 output. The mechanism is within AI studio, one out of every x prompts is being fed to an “advanced model.”

Some of the things users have built so far:

A virtual windows environment including a working snake and calculator game.

An SVG of an Xbox controller:

SVGs are a bit trickier than generating images as they require distinct shapes. Here’s Anthropic’s flagship model 4.5 sonnet trying the same thing.

or here’s a rough emulator of a Nintendo Switch, made in under a minute.

or even a simulation of a blackhole.

These are also mostly “one-shots” as one would need to wait 1/x prompts for Gemini 3 to appear again on the a/b test. Which means either (a) Gemini 3 is legitimately incredible, or (b) someone with access is really good at cherry-picking demos. My money’s on (a)

Google already had a ~4 month lead on the competition. With the release of Gemini 3, Google has at least kept its lead or even stretched its lead to 6 months +. This is terrifying for the competition. xAI had to go $12B in debt for the 1GW Colossus 2 data center. OpenAI has been busy doing financial gymnastics to fill its own funding gaps. Even Meta is going $26B into debt for a data center.

Unlike some of its competitors, Google can fund its model development directly from its $134B in operating cash flow, ahead of all other companies outside Microsoft. Unlike Meta or Microsoft, Google doesn’t have to give up 70% gross margins to Nvidia on chips. Unlike OpenAI, Google doesn’t need to promise some margin to Oracle on data centers. Google is able to get better ROI on its AI spend due its vertically integrated stack. While OpenAI will have its own custom chips eventually, that’s still two years away and OpenAI will be on a first-gen product which has its own issues.

Valuation and Risks

Google is trading at a forward PE of 25, below the 27 Forward PE of the Nasdaq 100 as a whole and within Google’s historical range of 20-30 times earnings. This means investors are on average neither bullish or bearish Google compared to the market. If Gemini 3 establishes clear AI leadership, multiple expansion should be in play.

If AI really is a 10 trillion market and Google is 1st, great! If AI falls apart and Anthropic/xAI go bust, search will be fine. One potential scenario, is 2-3 AI players reach profitability at scale and share the AI market, similar to today’s cloud market. The pure margins of Windows/Office licenses are likely better for Microsoft than Azure, but the sheer market size of cloud has made up for it and more. The same could be true when thinking about Google search vs. AI mix.

The bottom line is AI is the most important story in the market, Google is already leading today and is likely to lead even more after Gemini 3. If Google gets too far ahead, the “vibes” at Meta might not be so good.

Investors are likely to recognize if Google’s AI lead is strong enough, the next logical question is why shouldn’t Google be the most valuable company in the world?

Disclosure: Long GOOGL Nov 21 255 Calls. Not investment advice - just ideas that seemed brilliant at 2am. Do your own research.