The 13F filings that arrived last month describe a coordinated shift. Stanley Druckenmiller’s fund added to its positions in Amazon and AMD during the second quarter while selling or eliminating stakes in Broadcom, Intel, and Micron. Chase Coleman’s Tiger Global trimmed several of its largest technology holdings and added to AMD and SpaceX. David Einhorn’s DME built a new position in PayPal and exited Victoria’s Secret and Peloton. Three funds with three very different styles landed on the same conclusion: the AI trade is moving from chip makers to the companies that use their chips.
The filings, which reflect positions held at the end of June, show the direction of the rotation with unusual clarity. Druckenmiller, whose macro and technology calls have made him one of the most watched investors in the market, increased his exposure to Amazon, the largest cloud and e-commerce company, and to AMD, the second-largest designer of AI accelerators. At the same time, he cut or exited Broadcom, which supplies networking and custom chips to the AI buildout, Intel, whose turnaround remains incomplete, and Micron, the memory maker at the center of the HBM boom.
The pattern within the pattern is about where value sits in the AI stack. The first phase of the AI trade rewarded the companies selling picks and shovels: Nvidia’s GPUs, memory from Micron and SK Hynix, networking from Broadcom. The second phase, in the eyes of these managers, rewards the companies that deploy AI at scale, including the cloud platforms, application software, and consumer internet giants whose products are being rebuilt around machine intelligence. AMD’s position in the rotation is a bet that the accelerator market is becoming a two-company race, with AMD gaining share from Nvidia at the margin.
Tiger Global’s moves tell a similar story with a different emphasis. Coleman’s fund, which built its reputation on technology growth investing, reduced exposure to several large technology companies while adding to AMD and to SpaceX. The SpaceX addition fits the fund’s history: Tiger Global was an early and large investor in private space and technology companies, and its willingness to hold through the listing period reflects a longer time horizon than its hedge fund peers. The AMD position, meanwhile, is a bet on the compute layer that Tiger Global has been building across its portfolio.
Einhorn’s PayPal purchase is the outlier that confirms the theme. PayPal is not an AI company in the way Amazon or AMD are, but it is a large platform whose operations, fraud detection, and merchant services are being rebuilt around machine learning, and whose stock trades at a fraction of the multiples of its AI peers. Einhorn, known for value investing and for public letters criticizing expensive momentum trades, appears to be treating PayPal as the application-layer value play of the AI rotation. His exits from Victoria’s Secret and Peloton, both struggling consumer businesses, complete a shift toward platforms built around AI and away from retail-facing names with weak growth.
The coordination is more likely a matter of similar analysis than collusion. All three managers run concentrated portfolios, read the same earnings reports, and face the same question: after two years in which chip stocks carried the market, which part of the AI economy offers the best risk-adjusted returns from here? The answer visible in the filings is that applications and compute alternatives look more attractive than the most crowded semiconductor trades.
The filings come with caveats that matter. The 13F form is filed with a 45-day lag, so the positions reflect June 30, not the present. Managers can and do change their minds quickly, and the second-quarter filings say nothing about what the funds hold today. The forms also report only long positions in U.S. equities, hiding options, short positions, and foreign holdings, all of which could alter the picture. Analysts who study the filings caution against reading too much precision into a snapshot taken two months ago.
The rotation also carries a valuation logic. Semiconductor stocks, after their enormous run, trade at multiples that assume several more years of hypergrowth. The companies deploying AI, by contrast, have seen their earnings estimates rise while their stocks have been more selective, leaving the application layer relatively cheaper on forward earnings. For managers whose mandates reward finding value within growth, the arithmetic favors shifting some exposure down the stack.
The risk in the rotation is that it is early. Chip companies continue to report record results, and the memory, networking, and accelerator makers are still the companies where the AI revenue is most visible. If the hardware cycle extends, the managers who sold Broadcom, Micron, and Intel will have sold into strength, and the applications they bought will lag. The test of the trade will come in the next few quarters, when the earnings reports of both groups reveal whether the value has moved from the suppliers to the deployers.
For now, the filings offer a rare look at convergence among sophisticated investors. Three managers with different styles, different histories, and different client bases reached the same conclusion about the second quarter: the pure semiconductor trade is crowded, and the better risk-adjusted opportunities are in the companies applying the technology. Whether the market agrees will be decided by the earnings reports, the stock prices, and the next round of filings, which will show whether the rotation was a repositioning or a change of heart.
