All data herein is representative of the Institutional Share Class. Please see important disclosures at the end of this commentary.


Dear Shareholder:

The FPA Crescent Fund – Institutional Class (“Fund” or “Crescent”) gained 7.13% for the quarter and 14.75% in the trailing twelve months.

It captured 62.3% of the global market’s return (i.e., MSCI AWCI, the “ACWI”) over the trailing twelve months, slightly underperforming its own 63.8% average net risk exposure during the same period.

Performance versus Illustrative Indices (%)1

Portfolio & Market Discussion

In the six months ending June 30th, equity markets delivered strong returns (11.25% for the ACWI and 10.21% for the S&P 500) and became increasingly narrow.  Exciting AI capabilities and data center supply shortages drove the market and economy.  We continued to find opportunities in attractive mid-sized companies that are largely AI and data center-agnostic.

The following pictures illustrate the market’s narrowness as seen in the indices.  

AI-related stocks have captured investors’ minds and wallets, driving the majority of recent stock market returns and, by some counts, accounting for 50% to more than 75% of the S&P 500’s total gains since late 2022.2  According to J.P. Morgan, “AI” companies now constitute approximately 50% of the S&P 500, exceeding even the Information Technology sector’s 35% weight at the then S&P 500 peak in March of 2000.

Weighting of AI-Related Companies in the S&P 500 Index3

Semiconductors and tech hardware now represent nearly 30% of the market, the highest level recorded.

Weighting of Tech Hardware Companies in the S&P 500 Index4

Over the last six months, semiconductors and tech hardware accounted for nearly 90% of the S&P 500 Index’s return.5 

Until very late in the second quarter, momentum had been the strongest indicator/driver of returns and relatively more pronounced than in the past.

Ratio of US Momentum stocks to US Minimum Volatility stocks6

From many vantage points, the market, as expressed by the indices, is quite narrow.

Good reasons underpin the market’s narrowness.  Progress in AI, first with OpenAI/ChatGPT and more recently with Anthropic/Claude/Mythos, has provided a credible peek into a future in which humans will massively increase productivity.  If one squints, we can imagine machines embedded with intelligence that allow humans to enjoy a substantial surplus of necessary goods/service/comforts without requiring traditional physical/mental labor.  While some technologies are speculative and others will take decades to diffuse, the potential is real and probable.  If you need convincing, try a Waymo during your next visit to Los Angeles.  

Rapid adoption of large language models (LLMs), facilitated by ubiquitous access to smart devices and fear of being left behind, has led to an epic boom for the suppliers of picks and shovels to the AI (datacenter) buildout.  

To date, the boom has exceeded expectations.  At year-end 2024 investors were generally bullish on Nvidia, ChatGPT had been introduced two years earlier. Nvidia had seen its revenue increase by 4x, and stock increase 6x since the introduction of ChatGPT.  Despite being a well-recognized beneficiary of AI-led chip demand, analysts materially underestimated Nvidia’s near-term performance.  At year-end 2024, analysts expected 2026 revenue of $130-150bn. Run-rate will be closer to $400bn.  To date, the correct call on fundamentals has been to be more bullish than the bulls.

Nvidia is not alone; semiconductor revenue is through the roof, and the memory market has developed an acute shortage.

Monthly Global Semiconductor Billings ($ billions)7

This spending is primarily supported by the hyperscalers. 

Big Tech Capital Expenditures ($ billions)8

This capital spending comes from cash generated by strong businesses that have seen revenues accelerate over the past few years.  More recently, the baton has been passed to the capital markets through both debt and equity issuance. There is some circularity to this activity, but the vortex of activity is real and well-financed (to date).

As of today, there appears to be no limit to demand for compute/data centers, which is a good thing because supply is coming. 

The US had 40-53 GW of installed capacity at the beginning of 2026 and should double to 95 GW by the end of 2027.9  As shown below, the total pipeline is nearly 300 GW. There are well-known limits around permitting, power supply, semi/memory and other materials, so substantially less than the pipeline will be delivered, but it appears likely that capacity will multiply in the next few years.  

Data Center Capacity by Stage (in GW)10

And that’s just on Earth.  SpaceX went public in the second quarter at a $1.8T valuation.  SpaceX is an awe-inspiring firm and has a credible claim to being the most consequential company on the planet. Still, according to sell-side reports, most of its IPO value is attributable to its speculative promise to build data centers in space.  According to Elon, the company will deliver an additional 1 GW per year of capacity in 2027, 10 GW per year beginning in 2028, 100 GW per year beginning in 2030 and eventually 1 TW per year.  That’s a lot, and the company appears entirely serious about these intentions, given their active building of mega facilities to supply the necessary equipment.

It appears that, on Earth or in space, the market has confidence (based on valuations of the various businesses pursuing/benefiting from the build-out) that significant compute capacity will be delivered over the next few years.  We can’t help but wonder what the returns will be on all this compute spend?  Could compute capacity exceed AI’s ability to diffuse usefully through the economy? Currently, the market for compute is in backwardation, with forward 12-to-24-month prices 30 to 40% below the spot price (on, as best we can tell, a like-for-like unit).  We are also struck by the recent contracts that SpaceX signed with Google and Anthropic for scaled compute capacity immediately available at prices 2 to 5x estimates of the long-term cost, but cancelable with 90-day notice.

We are bullish on AI and compute demand, but not more bullish than the bulls, and don’t feel that the market is offering us any easy question in the AI/hardware space.

 

Given the exciting future and economic importance of AI, we spend time using, reading, and thinking about the technology, its near-term impact, and longer-term potential.  Currently, we have more questions than answers and must admit that, mostly, we just don’t know.

A recent Invest Like the Best podcast, titled “The Cone of Uncertainty” with Krishna Rao, Anthropic’s CFO, neatly captures our feelings.  During the discussion, Rao said, “We think about the world as scenarios. It’s very hard to have a point estimate in this business. And then having a very low bar for updating your current priors, or your current perspective. It could be the case that something a month ago was true that’s just not true today, and that breaks your model, and you have to go back and update it.”

Thus far, all updates – capabilities, revenue, and compute demand — have been positive.  Should we expect all future updates to be as favorable? And, has OpenAI had a somewhat different experience in the past six months? In early 2025, OpenAI appeared to be the clear leader among independent frontier labs; today, not so clear. 

Annualized Revenue (USD)11

Will frontier lab leadership prove persistent or fleeting?  The following chart suggests fleeting.

Selected Frontier LLMs by Aggregate Benchmark Score12

If fleeting, what is the long-term economic return to frontier lab innovation? Where will the economics stick?

Perhaps leadership will be persistent, and if it is, will economic value accrue to the leader, or to the low-cost, slight laggard? Or to some other level of the stack? The following suggests that lagging models (open-source and low-cost) catch up quickly and cost a fraction of the frontier. 

Language Model Inference Price ($/million tokens), by Model Intelligence13

Various reports suggest that companies are moving away from frontier models as costs become uneconomic relative to the value delivered. What activities require genuine frontier or super intelligence? At maturity, how much of the ‘intelligence layer’ can be served by good enough, low-cost commodity models?

Perhaps the most important near-term questions relate to the spend and sustainability of the data center buildout.  Direct (semiconductors/memory) and indirect (power, building materials, commodities, engineering firms, and cooling business) beneficiaries are all experiencing massive tailwinds and driving both the economy and markets.  The data center build is becoming large relative to GDP and relative to previous infrastructure buildouts.14,15

Some analysis proposes that the data center buildout is like nothing we have ever seen – multiples larger in inflation-adjusted terms than even the US railroad and interstate highway expansions of the 19th and 20th centuries.

Data Centers vs. Megaprojects Inflation-Adjusted Costs16

On the other hand, a slightly different lens supports the argument that the buildout is reasonably consistent with past large-scale fundamental infrastructure builds.

Infrastructure Buildout – Cumulative Capex as a Share of World GDP17

Even the data seems confused. If AI (and the robotic revolution it enables) ultimately satisfies all human needs without the need for traditional human labor, ‘Super-Abundance’, then, perhaps, historic precedents are irrelevant, and compute should represent a massive share of economic activity.  But will there be a cycle in what is now a massively capital-intensive industry? And how do intelligence cycles work?

Together, these (and other) questions posed to long-term investors by the AI boom are difficult to answer. We are left uncertain about the sustainability of the current and projected level of AI capital spending and the prospect of it generating reasonable returns on the aggregate investment. Global spending influenced by AI is projected to skyrocket to a record $2.59 trillion in just this year.  To achieve even a relatively low unlevered return on investment suggests (an improbable?) $207 billion in after-tax income, per year, starting immediately.18

In some important ways, AI is unique, and we have long invested in businesses that are well positioned for the digital future.  Many of the businesses we own (Semiconductors/Hyper-scalers/Connectors) benefit from robust tailwinds from data center expansion and compute utilization.  On average, we have owned these companies for over a decade and currently believe that they will each be larger and more profitable 5 and 10 years from now.  These businesses are diverse, serve markets beyond just datacenters, and are growing faster than we expected a few years ago.  Given expanded valuations and the stage of their capital cycle (far from depressed), we are scaling back exposure because we believe the risk-adjusted returns are less attractive – the margin of safety is shrinking, if not disappearing. We have been more aggressive in exiting the names that have benefited most directly from compute shortages.

Given our excitement about the technology, strength of the companies’ positions and belief that compute demand will be greater over time, we think it is advisable to own a bit of the ‘future’ at a supportable (but potentially stretched valuation) should the capital cycle peter out.

AI is not the first theme markets have focused on that requires long-term investors to answer difficult questions in the affirmative.  Since we have been managing the Fund, we recall the early web companies in the Dot-com era, Commodity Super-Cycle, EM Premium (remember the BRICs), MLP enthusiasm, SAAS, and Healthcare/Covid ‘winners’.  In each of those cases, we generally avoided the theme based on what we viewed as unsustainable valuations at stretched points in the capital cycle.

Our strategy is similar in the current environment.  We seek investments that are asking long-term investors easier questions. Today, we have the opportunity to buy businesses we consider relatively AI-agnostic businesses at valuations that support what we see as an attractive expected absolute return.  In the first half of the year, we added 14 new companies and increased our holdings in several companies purchased last year. They fit nicely in the Venn diagram we shared at the end of last year, updated below through June.19

These businesses range from specialty chemical distribution to biotech equipment and were generally purchased at 10-15x our estimate on NTM earnings. While significantly smaller than the index leaders, our recent purchases are industry leaders, with strong balance sheets and attractive growth profiles.

In fact, we believe the generic profile of our portfolio has improved – faster growth and reasonable valuation.

In our view, a portfolio of the companies in the Venn diagram (at our basis) is likely to produce an absolute equity-like return, with little risk of permanent impairment.  This is the type of portfolio we are always striving to create: a diversified mix of high-quality businesses, with good management, purchased at valuation math based on conservative assumptions that should support an equity-like return over the long term.    

These holdings are primarily mid-cap businesses that look nothing like the index.  We are willing to look foolish relative to others and to miss out on the bullish scenarios for AI in order to own a portfolio that we believe is highly likely to result in an acceptable absolute result.

Being out of step is nothing new.  You may remember the healthcare/biotech boom that took place in 2020 and 2021.  At the time, we had taken our exposure to zero – down from our largest exposure at the time of Obamacare concerns in the early 2010s.  We recall being asked repeatedly whether we understood that there was an exciting future in biotech and cell and gene therapy. 

In fact, we were excited, but just not at the prices available at the time. We continued to research the sector and developed a view towards a number of businesses we’d like to own.  In the past year we have bought many of those very same businesses at valuations less than half the prevailing rate in 2020/21, and we submit that the future is just as bright and perhaps enhanced by potential AI innovation.  Willingness to ignore investments with terrific momentum at valuations and long-term fundamentals that ask difficult questions (Will Covid spending be sustainable? Are these new multiples reflective of fundamentally different businesses?), combined with actively deploying capital when fundamentals and valuation are supportive of equity-like returns, will always be our preferred approach.  However, we accept the risk of looking foolish at times and humbly settle for reasonable absolute returns when parts of the market are delivering spectacular returns based on thinking that we don’t quite understand. 

In short, we are seldom more bullish than the bulls.  Though when analysis and consideration support the position, we are happy to be less bearish than the bears, which should continue to help us generate returns that are attractive on an absolute basis and not risk taking the fund (and you) off a cliff.

Respectfully submitted,

FPA Crescent Portfolio Managers

July 30, 2026


Crescent’s top five performers contributed 9.31% to its trailing twelve-month return while its bottom five contributed -2.95%.


Trailing Twelve-Month Top and Bottom Contributors (%) as of June 30, 202620

The following companies impacted portfolio performance but have not been recently discussed.21

Analog Devices (“ADI”), a leading supplier of analog and mixed-signal semiconductors, benefited mostly from strong demand tied to AI infrastructure, industrial automation, and communications equipment. During the quarter, shares rose after ADI reported record quarterly results, including 37% year-over-year revenue growth, expanding margins, and management pointing to record bookings across several end markets. Investors were further encouraged by the company’s announced acquisition of Empower Semiconductor (closed in early July) and guidance for continued growth into the second half of the year.

Samsung C&T’s shares rose on the back of improving sentiment toward the broader Samsung ecosystem, which was supported by surging demand for AI-related semiconductors and memory products. Investors increasingly focused on the value of Samsung C&T’s strategic holdings (namely Samsung Electronics) and exposure to the construction, energy, and life sciences segments. While market volatility emerged late in the quarter following concerns that AI-related capital spending could eventually moderate, the stock still closed the quarter with strong gains.

Nintendo weighed modestly on the portfolio despite continuing operational success. The company entered the period following the highly successful launch and adoption of the Nintendo Switch 2 platform, which drove hardware and software sales. However, the potential for increasing memory prices negatively impacting Nintendo’s hardware margins weighed on shares. We continue to view Nintendo as a strong franchise with valuable intellectual property and a long runway for monetization across gaming.

Amsterdam-listed Prosus owns a collection of global technology and e-commerce businesses and remains among the largest shareholders of China’s Tencent. While the company reported strong results, including accelerating profitability across its e-commerce network and improving cash generation, the share price declined as investors weighed execution risks associated with several large acquisitions and investments.



FPA Crescent Fund Portfolio Highlights