Why Google was a genius and Microsoft is a villain

$3.1 billion. That’s what Google paid for DoubleClick in 2007, a company most consumers had never heard of that controlled the plumbing underneath online advertising.

The mistake most investors make right now is assuming Microsoft’s OpenAI arrangement is fundamentally different from that deal. It isn’t. Both are distribution giants buying the layer that sits beneath everything else — the infrastructure layer that competitors can’t easily replicate or route around.

In 2007, analysts called DoubleClick visionary. Google had roughly 55% of U.S. Search traffic and about $12 billion in annual revenue, tripling from $4 billion just two years earlier. The acquisition looked like a company growing into its natural shape. Smart infrastructure thinking. Nobody blinked.

Microsoft integrates OpenAI into Azure, Bing, Copilot, and Office — structurally the same bundling play — and the word “monopoly” shows up in every headline.

The deals are twins. The only variable that changed is the cultural moment. We went from “move fast and build” to “move fast and face a subpoena.”

You need to hear this: the difference in public reaction doesn’t reflect a difference in the underlying transaction. It reflects a difference in when the transaction happened, and who got there first.

It’s easy because precedent is invisible until it becomes liability.

When Google bought DoubleClick, it also acquired YouTube for about $1.6 billion in 2006. It launched Chrome in 2008 as part of its push toward cloud applications. Each move looked logical in isolation. Nobody drew the through-line from “we now control both the ad inventory and the ad server” to “we’ve just built the foundation of a future DOJ monopoly case.”

First time here? Browse past issues

That case arrived fifteen years later.

The tell was right there in the filings, if anyone had bothered to compound the antitrust surface area forward.

Here’s what most analysis misses — the hidden cost of an acquisition isn’t the purchase price. It’s the regulatory debt that accrues silently for a decade, then comes due all at once. The DOJ said Google and Yahoo together accounted for 90% or more of each relevant market back in 2008 when it prepared to block their ad partnership. Google was already brushing the antitrust tripwire before most investors even understood the DoubleClick integration was complete.

The regulatory hangover from an acquisition doesn’t show up in the next earnings call. It shows up in the next administration. By then the deal looks inevitable — and so does the lawsuit.

“But Microsoft didn’t acquire OpenAI outright. It’s a licensing and investment structure, not a full buyout.”

Fair point. And it’s the argument Microsoft’s lawyers will make for the next decade. But the structural reality is that Microsoft controls OpenAI’s distribution surface through Azure exclusivity, has board observation rights, and captures the majority of commercial inference revenue. Call it what it is: a deal that walks and talks like an acquisition while wearing the legal clothing of a partnership. The DOJ has shown it’s willing to scrutinize exactly these kinds of arrangements when they produce monopoly outcomes.

The right way is to separate the investment thesis from the regulatory thesis and price them independently.

Google’s DoubleClick deal was a generational return on capital — for about a decade. The acquisition compounded Google’s ad dominance, helped drive revenue from $12 billion in 2007 to over $280 billion by the time regulators caught up. Then the clock ran out.

The DoubleClick acquisition wasn’t a mistake — it was a masterpiece with a built-in expiration date.

Microsoft’s OpenAI clock started the same way. The return window is real. So is the eventual liability. Pricing only one of these is the analytical error most investors are making right now.

Both views are correct — they just operate on different time horizons. The investment thesis and the antitrust thesis aren’t in conflict. They’re sequential. The print looks great for years. Then it doesn’t.

Here’s what it looks like now: investors see Microsoft’s OpenAI integration as a capex cycle play — Azure inference revenue growing, Copilot attach rates climbing, operating margins expanding. The read-through is straightforward. Spend now, harvest later. Nobody is drawing the through-line forward, because nobody ever does until the subpoena arrives.

Here’s what it looks like after: a decade from now, the same integration that drove those margins becomes the centerpiece of a government complaint. Remedies could force unbundling, licensing changes, or structural separation. The compounding returns start running in reverse.

Let me put a number on it. If you’re modeling Microsoft’s AI revenue contribution, you need to carry at least five hidden costs:

1. Regulatory defense spend — Google’s antitrust litigation has cost hundreds of millions in legal fees alone. Microsoft should expect the same, starting roughly five to seven years out from today.

2. Remedy risk to bundled revenue — if forced to unbundle Copilot from Office or decouple Azure from OpenAI’s models, the unit economics of each product degrade. This isn’t theoretical; it’s the exact remedy structure the DOJ pursued against Google’s search distribution.

3. Opportunity cost of regulatory distraction — management bandwidth consumed by depositions, document production, and consent decree compliance is bandwidth not spent on product. Google’s leadership has been in and out of courtrooms for three years running.

4. Partner defection premium — as antitrust scrutiny intensifies, enterprise customers start hedging. They sign secondary contracts with Anthropic, Mistral, or whoever’s next. That hedging costs Microsoft pricing power it won’t get back.

5. The political optionality discount — antitrust enforcement changes with administrations. You can’t model which administration will be in power in 2033. That uncertainty itself functions like a cost — and the deeper the integration, the harder it becomes to unwind without triggering the very outcome you were trying to avoid.

I’ve watched this playbook run twice now, once from a sell-side desk and once from the outside. The cadence is identical.

This isn’t complicated. The question isn’t whether Microsoft’s OpenAI bet generates returns. It will — probably enormous ones, for five to eight years. The question is whether you’re pricing the full lifecycle cost, or just the front half.

Google’s DoubleClick deal earned its cost of capital many times over before regulators arrived. But if you’d modeled only the upside, you’d have missed the largest single legal liability on Google’s balance sheet today.

Watch the cadence, not the guide. The clock is already running.