Broadcom's AI Surge: Analyzing the OpenAI Deal vs. Ken Griffin's Exit

It’s not often you get a clear look into the playbook of a financial titan. But every quarter, 13F filings give us a glimpse. And in the second quarter of 2025, the filings from Ken Griffin’s Citadel revealed a move that looked like a powerful, declarative statement on the future of artificial intelligence hardware. The firm dumped a massive portion of its stake in Broadcom (AVGO) and plowed capital into the undisputed market leader, Nvidia (NVDA).

This was a classic "smart money" signal. Citadel, with its army of analysts and near-limitless data resources, was making a choice: it was backing the general-purpose GPU king over the specialist in custom silicon. The move seemed logical, a textbook case of taking profits after a massive run and rotating into the sector's gravitational center. For weeks, that was the story.

Then, on October 13th, the narrative was completely upended. Broadcom announced a mammoth $10 billion custom chip deal, widely understood to be with OpenAI, and the stock surged 9% (Broadcom stock pops 9% on OpenAI custom chip deal, adding to Nvidia and AMD agreements). The specialist that Griffin had just sidelined suddenly looked like the most strategic player in the game. It begs the question: Did one of the sharpest investors on the planet misread the chip market, or is there a different story hidden in the data?

The Logic of the Pivot

To understand Citadel’s decision, you have to look at the data available in the second quarter. Broadcom’s stock had been on an absolute tear, up roughly 90%—to be more exact, 91% in the prior year—pushing its market cap into the stratosphere (at around $1.63 trillion). It was trading at a demanding 50 times forward earnings. From a risk management perspective, trimming a position that has performed that well is not just prudent; it's standard operating procedure.

The underlying thesis for caution had merit. Broadcom’s custom chip business, while incredibly lucrative, was concentrated among a few hyperscale clients like Google and Meta. If AI infrastructure spending were to slow down, or if one of those key clients decided to shift its strategy, Broadcom’s revenue could be disproportionately affected. Selling 82% of the position looks like a calculated move to de-risk and lock in substantial gains.

The capital then flowed to Nvidia. Citadel more than quadrupled its stake, buying into the one company that serves as the foundational layer for the entire AI ecosystem. This is the ultimate pick-and-shovel play in the AI gold rush. While Broadcom is like a master craftsman forging bespoke tools for the two or three largest mining operations, Nvidia is the one selling the essential, high-margin equipment to every single prospector on the frontier. The trade was a pivot from concentrated, customer-specific risk to a broader, market-wide bet on AI proliferation. On paper, it was a defensible, even intelligent, reallocation of capital. What possible reason would a fund have to believe the bespoke toolmaker was about to land the single largest contract in its history?

An Inconvenient $10 Billion Footnote

The problem with even the most well-reasoned trade is that it can be instantly invalidated by new information. And the information that emerged about Broadcom post-Q2 was nothing short of monumental.

First came the blowout fiscal Q3 results. AI-related semiconductor sales had surged 63% year-over-year, and the company was sitting on a record backlog of $110 billion. That alone should have raised eyebrows. A backlog of that magnitude suggests a level of forward visibility that contradicts the idea of a precarious, overly concentrated business. Then came the bombshell: a confirmed $10 billion order for custom AI chips from a new cloud customer, set for delivery by 2026. This wasn't just another order; it was a validation of Broadcom's entire business model. The market’s most important new player, OpenAI, had chosen the custom ASIC route for its future hardware needs, and it chose Broadcom to execute it.

And this is the part of the data that I find genuinely fascinating. The very risk Citadel appeared to be trading away from—customer concentration—morphed into Broadcom’s greatest strength. Landing OpenAI doesn't just add a massive revenue stream; it cements Broadcom’s status as the go-to partner for tech giants building proprietary AI systems. Analysts who had been bullish, like Mizuho’s Vijay Rakesh, doubled down, reiterating his "King of AI Custom Silicon" thesis and slapping a $410 price target on the stock. The market agreed, sending the stock popping.

It’s impossible to know the internal calculus at Citadel, which famously operates as a "pod shop" where individual portfolio managers have significant autonomy. Was this one manager's call? Or did the firm's overarching view miss the accelerating trend of hyperscalers seeking custom silicon to gain a competitive edge over off-the-shelf GPUs? What metrics could they have been tracking that failed to signal the magnitude of the OpenAI deal simmering just below the surface?

A Tale of Two Theses

Let's be clear: Ken Griffin is not going to lose sleep over this. Citadel is a market-making and high-frequency trading behemoth, and this long-only position is just one piece of an impossibly complex puzzle. We can't see their hedges or their short positions. But from the perspective of an investor analyzing publicly available data, the conclusion is unavoidable. The move to dump Broadcom for Nvidia looks like a significant, data-driven miscalculation. It was a trade that swapped a unique, defensible moat in custom silicon for a more crowded, albeit powerful, position in general-purpose hardware, right before Broadcom proved its moat was deeper and wider than the market understood. It underscores a critical lesson: sometimes the "smart money" is just making a smart-sounding bet based on yesterday's data.