Hero Background Image
Quarterly Letter

Baron Technology ETF | Q3 2026

Portfolio managers Michael Lippert and Ashim Mehra

Dear Baron Technology ETF Shareholder,

During the third quarter, Baron Technology ETF (the Fund) declined 4.50% (NAV), compared to the MSCI ACWI Information Technology Index (the Benchmark), which gained 4.37%, the Nasdaq-100 Index (the NDX), which gained 0.58%, and the S&P 500 Index, which rose 2.30%.*

On a trailing three-year basis, the Fund generated an annualized return of 36.59%, compared to the Benchmark and NDX, which were up 38.33% and 28.31%, respectively.*

Annualized performance (%) for periods ended September 30, 2026†
 ETF Market
Price1,2
ETF 
NAV1,2
MSCI ACWI Information
Technology Index1
S&P 500
Index1
MSCI ACWI
Index1
QTD3(4.71) (4.50) 4.37 2.30 1.60 
YTD323.22  23.52  35.40 12.75 13.03 
1 Year22.58  22.57  39.30 15.74 16.75 
3 Years36.26  36.59  38.33 22.89 21.73 
Since Inception
(12/31/2021)
14.91  15.09  19.64 12.07 10.70 

Performance listed in the above table is net of annual operating expenses. The total annual fund operating expense ratio as of April 30, 2026 was 0.75%. The performance data quoted represents past performance. Past performance is no guarantee of future results. The investment return and principal value of an investment will fluctuate; an investor’s shares, when redeemed, may be worth more or less than their original cost. Total returns assume the reinvestment of all distributions and the deduction of all fund expenses. Current performance may be lower or higher than the performance data quoted. For performance information current to the most recent month-end, visit BaronCapitalGroup.com or call 1-800-99-BARON. 

Pursuant to the Fund’s unitary management fee structure, BAMCO, Inc. (“BAMCO” or the “Adviser”) will pay all expenses of the Fund, except for portfolio transaction costs, interest and dividend expense, acquired fund fees and expenses, fees and expenses related to filing foreign tax reclaims, line of credit expenses, and extraordinary expenses.

NAV and Market Price returns include returns of the Institutional Shares of the predecessor mutual fund prior to the ETF’s commencement of operations. Prior to the ETFs listing on 12/15/2025 the NAV returns of the Institutional Shares of the predecessor mutual fund are used as proxy market price returns. If the predecessor mutual fund had been structured as an ETF, its performance may have differed.

Review & Outlook


Market Backdrop

U.S. equities delivered a mixed but ultimately positive third quarter. A sharp rotation away from AI and technology stocks in July gave way to a two-month recovery, and the S&P 500 and NASDAQ Composite reached new all-time highs despite rising interest rates, persistent inflation concerns, and geopolitical uncertainty. The Federal Reserve raised rates in September for the first time since 2023, and the 30-year Treasury yield climbed to its highest level since 2002. Gains were uneven: the S&P 500 and NASDAQ Indexes rose, while the Dow Jones Industrial Average and the small-company Russell 2000 Index declined, and market leadership narrowed toward the largest technology and AI-related companies as the quarter progressed.

In July, investors shifted their focus from the strength of AI demand to the timing of returns on massive capital spending. Semiconductor stocks fell by double digits on concerns about spending by the large cloud providers, open-source competition, and expanding memory supply, while money flowed into overlooked areas such as software, where valuations had fallen to multi-year lows. Over the final two months, strong corporate earnings, capped by blowout results from NVIDIA Corporation, restored confidence that demand for AI computing remains durable and exceeds supply. The Fed’s rate hike weighed on stocks most sensitive to the economy and interest rates, while technology proved largely insulated, lifting the Magnificent Seven to record highs by quarter end.

Energy was the top-performing sector by a wide margin, as a supply disruption in the Strait of Hormuz, which the International Energy Agency called the largest in the history of the global oil market, pushed oil back above $100 per barrel. Health Care, Information Technology (IT), and Communication Services also outperformed. Within IT, cybersecurity stocks benefited as companies stepped up spending in response to geopolitical threats and AI-enabled attacks, and application software rebounded sharply as solid results eased fears that AI would erode incumbents’ pricing power and displace their platforms.
 

Performance

We manage the Fund with an unwavering focus on powerful technology trends that are disrupting industries and creating sustained, profitable growth opportunities, rather than on short-term sentiment swings, which were pronounced again this quarter. Transformative secular trends such as AI, space exploration and technology, autonomous transportation, robotics, digital commerce, media, and finance, and advanced therapeutics and minimally invasive surgery will shape the future and drive long-term investment returns.

The third quarter was a more difficult backdrop for the Fund. Our exposure to semiconductor and AI infrastructure companies, which drove much of our strong performance in the first half, left the Fund more exposed than the Benchmark to the July reversal in AI sentiment. When technology stocks recovered over the final two months, the gains were led by a small group of mega-cap companies where our positions are smaller than the Benchmark’s, so the Fund did not fully participate in the rebound. The Fund trailed its Benchmark during the quarter. We believe the shortfall reflects the market’s shifting sentiment toward AI spending rather than any deterioration in the fundamentals of our holdings.
 

Versus the Benchmark

The Fund underperformed the Benchmark during the quarter.

Within the IT sector, our non-ownership of several of the Benchmark's largest constituents was the principal source of relative underperformance. Not owning Microsoft Corporation in systems software and Apple Inc. in technology hardware, storage & peripherals acounted for about half of the relative shortfall as both stocks outperformed. Separately, our overweight in semiconductor materials & equipment, particularly our large position in semiconductor equipment leader Lam Research Corporation, weighed on results. Coherent Corp. in electronic components also detracted. Within semiconductors, Taiwan Semiconductor Manufacturing Company Limited (TSMC) was a drag on performance, and our underweight relative to the Benchmark in NVIDIA was also detrimental. The pullback in Lam and TSMC followed their outsized contributions in the second quarter, when both were among the Fund's top performers, and our long-term conviction in each is unchanged.

Outside of IT, our position in Space Exploration Technologies Corp. (SpaceX) in Communication Services detracted, as did our exposure in Industrials, where both public safety technology leader Axon Enterprise, Inc. and power solutions and components leader Forgent Power Solutions, Inc. lagged.
 

Versus the NDX

The Fund underperformed the NDX for the quarter, with the largest relative shortfalls in IT, Industrials, and Communication Services.

Within the IT sector, relative underperformance was driven by weakness in several of our holdings, as well as by our non-ownership of certain NDX constituents. Our overweight in semiconductor materials & equipment weighed on performance, led by semiconductor equipment leader Lam, the largest detractor from performance during the quarter. Coherent in electronic components and TSMC, a non-NDX position in semiconductors, also detracted. The pullback in Lam and TSMC followed their outsized contributions in the second quarter, when both were among the Fund's top performers, and our long-term conviction in each remains unchanged. In addition, not owning Microsoft in systems software, Apple in technology hardware, storage & peripherals, and Palantir Technologies Inc. in application software weighed on relative returns, as all three stocks outperformed.

Outside of IT, our overweight position in SpaceX in Communication Services detracted, as did our underweight in Meta Platforms, Inc., which outperformed. Our overweight in Industrials also weighed on results, driven by weakness in shares of Axon and Forgent. In Utilities, our unique exposure to Fervo Energy Company in renewable electricity also detracted.
 

AI Update

Over nearly four years, we have experienced the first waves of the most significant disruption to the global economy since the Industrial Revolution. With the “ChatGPT moment” of late November 2022, the world took its first toddling steps into the age of machine intelligence. Until then, virtually all work and economic activity depended upon human, biological intelligence. That monopoly is ending. For the first time in history, we can manufacture intelligence. In this section, we will attempt to address the inflection in AI’s usefulness in 2026, the scale of the opportunity it opens, and the debates behind the recent market volatility.
 

The Inflections of 2026
A Really Useful Engine

On the main stage of the Baron Investment Conference last November, we emphasized that our research focus was on the utility and value of AI. Mike recalled a podcast with prominent AI researcher Andrej Karpathy, whose argument that developers should build “useful things,” not “AI animals,” struck him. He chuckled because it captured what we believed would drive AI’s adoption across the economy—and reminded him of Thomas the Tank Engine, a childhood favorite of his oldest son. In Thomas’s world, the highest praise for a locomotive was to be called “a really useful engine.” Our research asked whether AI could earn that distinction by improving productivity, cutting costs, accelerating innovation, and generating incremental revenue—and whether the curve of that value over time would be steep enough to give the AI cycle a long runway. In 2026, AI became a really useful engine.

As NVIDIA CEO Jensen Huang has said repeatedly since the company’s August earnings call, “AI has become useful.” In his telling, AI has gone from unprofitable and not useful to doing real work that is productive and profitable. We have moved from the age of computing, when software sat passively waiting for a person to engage it, to the age of agentic machine intelligence, when AI actively and autonomously plans, acts, and completes work on our behalf.

The catalyst was Anthropic’s release of Claude Opus 4.5 last November—the first model that could handle jobs lasting hours rather than seconds. It could break tasks into steps, work across software and the web, check its output, and deliver finished work at one-third of its predecessor’s price. OpenAI and Google followed within months, making 2026 the year agents became products. Anthropic’s Claude Code, designed for software engineers, was joined in January by Cowork, its counterpart for office workers. In August, SpaceX’s Grok Bot introduced agents that log into your applications and keep working after you close your laptop. Today’s frontier models—Anthropic’s Fable, OpenAI’s Astra, and Google’s Argon—were built to plan and execute work alongside people, rather than wait for questions like a chatbot.
 

Token Use Explosion

The adoption of agentic AI has set off an explosion in token consumption and the demand for compute. Agents consume compute very differently than people do. A person types a few hundred words into a chat box; an agent reads, writes, calls tools, and checks its work thousands of times per task, around the clock. Jensen Huang estimates that an agent can require 15 to 100 times the compute of a person using AI, depending on the problem. Recent data from OpenRouter supports this view: agentic token usage has grown roughly 14-fold over the past seven months, far outpacing human usage.

 

Source: OpenRouter data via Evercore ISI (Mark Lipacis, Semis & Semi Cap Research

 

Meanwhile, the price of intelligence keeps falling, and what matters most is not just the price of a token but the cost of completing a task. Opus 4.5 launched at a price two-thirds below its predecessor’s, and Anthropic says each new model generation completes a given task 25% to 75% more cheaply than the last. Some see deflation. We see Jevons’ paradox: make something useful cheaper, and people use far more of it. Model providers have seen the pattern repeat after each price cut: revenue dips briefly, then climbs past its prior level within a few months as usage outruns the lower price.

 

Monetization at Unprecedented Speed

The monetization of AI is unprecedented. Jensen Huang calls Anthropic and OpenAI the two fastest-growing technology companies in history. OpenAI began 2026 with about $20 billion of annualized revenue and told investors in October that its September run rate was approaching $50 billion. Anthropic’s third quarter revenue was widely reported to be $16.8 billion, which annualizes to roughly $67 billion, more than 10 times its level a year ago. Together, the two labs are running at roughly $120 billion annualized, up about six-fold in a year. For perspective, Salesforce needed about 20 years to reach $30 billion of annual revenue.

The driver is the enterprise. Customers that started with chatbot pilots are now running agents on long, multi-step work, and both labs report their largest customers spending multiples of what they spent a year ago. Still, AI remains a sliver of corporate spending, which is why we believe the runway is measured in years. Anthropic recently disclosed that more than half of the Fortune 500 spend over $100,000 a year on Claude, yet Claude represents just 0.1% of those companies’ combined expenses, and less than 1% even for its top 10 Fortune 500 customers. Ramp’s AI Index shows that 55.9% of businesses now pay for AI models, subscriptions, or APIs, far above the 23.8% reported by the Census Bureau’s Business Trends and Outlook Survey, yet the median monthly AI spend per employee is only $15.70, versus $813 for the top 10% of companies.
 

Source: Ramp AI Index, business spend data from Ramp includes corporate card and invoice / ACH payments. Census estimate from Census Business Trends and Outlook Survey.

 

Source: Ramp AI Index, business spend data from Ramp includes corporate card and invoice / ACH payments. AI spend includes LLM subscriptions, coding agent subscriptions, API tokens, and GPU cloud and infrastructure spend

 

Insatiable Demand for Compute

The surge in demand for agentic intelligence has become a surge in demand for compute. As Jensen explained: “AI agents…use an enormous amount of compute.” NVIDIA’s data center compute business grew 117% in its fiscal second quarter to $89 billion, and the company guided growth of at least 70% in 2027, a figure Jensen says is limited by supply, not demand. Broadcom Inc.’s AI semiconductor revenue rose more than 221% in its fiscal third quarter to $16.7 billion, and management guided the business to triple again in the fiscal fourth quarter, and to double in 2027, to $115 billion, and again in 2028, to $230 billion. The cloud providers are seeing it too. In the June quarter, Amazon Web Services (AWS), Microsoft Azure, Google Cloud, and Oracle together grew revenue about 50% year-over-year, the fifth consecutive quarter of acceleration and an extraordinary feat at roughly $420 billion of annualized revenue, and their combined contracted backlog reached about $2.3 trillion. SpaceX, the newest cloud provider and the Fund’s largest holding, reported third quarter AI segment revenue of $7.6 billion, up more than eight-fold from a year ago. On its recent earnings call, SpaceX confirmed that it expects to exit 2026 at $100 billion of annual recurring revenue (ARR),4 that it was confident in bringing an incremental 10 gigawatts of compute online in 2027, and that this expansion “implie[d]” that SpaceX’s ARR could end 2027 “somewhere between $500 billion and $1 trillion.” On a recent podcast, Amazon.com, Inc.’s CEO Andy Jassy noted that AWS’s AI business reached a run rate of more than $25 billion in the first three years of the generative AI wave, versus roughly $300 million at the same stage of AWS itself, several hundred times faster.

Every one of these management teams said demand exceeds supply and expects that to persist into 2027, and beyond. On a recent podcast, Jensen Huang explained that NVIDIA allocates customers far less than they want, and that every customer would buy more if NVIDIA could deliver more, “by far.” Andy Jassy declared: “2026 is obviously at this point sold out. 2027 is largely sold out ... A good chunk of 2028 is sold out.” Even at roughly $220 billion of capital spending this year, he said, Amazon is “leaving meaningful demand on the table…There’s just incredible demand right now.” SpaceX’s CFO Bret Johnsen announced on the recent earnings call that “[a]t current pricing, new compute deployments are generating a payback of less than a year.” Elon Musk added that “even old computers like [NVIDIA’s] H100s [are] renting for more than they rented when they were new,” that “we’re going to face a shortage of super intelligence compute, and that shortage is going to remain for a long time,” and that “it’s not impossible that we will have a six-month payback period for compute next year.” Every graphics processing unit (GPU) that can be built, shipped, and plugged in is being used. Unlike the dark fiber of the dot-com era, there are no dark GPUs.
 

Returns on AI Investments

During the internet and cloud eras, leading technology companies were viewed as “capital light,” and investors nearly forgot how access returns on invested capital (ROIC). Search, software, and social networks scaled on code and network effects. Even pre-AI cloud computing, technology’s most capital-intensive subsector in technology, generated demonstrably strong returns: we estimate AWS’s average ROIC exceeded 30% from 2017 through 2021. AI flipped the technology space on its head. AI is capital intensive, requiring huge capital investments to build AI factories to manufacture intelligence and generate tokens. Amazon, Google, Meta and Microsoft plan combined AI capital expenditures of about $725 billion in 2026, up 77% from $410 billion in 2025. We now project total hyperscaler capex will exceed $1 trillion in 2027. Deploying capital on this scale ahead of revenue understandably worries investors. A key market question is whether AI revenue will justify this unprecedented buildout.

We are asking this question too and are tracking it from the ground up. We analyze the cost of a gigawatt of compute and the revenue each watt generates. We measure the incremental revenue, gross profit, operating income, and cash flow produced by each dollar of capex.

On Amazon’s second-quarter call, CEO Andy Jassy laid out the economics clearly. An AI data center involves two capital cycles. The first, land, power, and the building shell, costs roughly $15 billion per gigawatt, is spent about two years before revenue, and lasts 30 years or more. The second, IT hardware (compute, memory, and networking), costs $35 billion to $45 billion per gigawatt and is bought only months before it goes into service, against visible demand. Servers break even in a little under three years and last five to six, and most AI capacity is contracted for five years or more. The capital most at risk of obsolescence, in other words, is committed last. Jassy expects AI returns to track those of the original cloud business. SpaceX, which rents compute on short-term contracts, reports paybacks of less than a year and sees a path to six months next year, because, as Elon Musk puts it, better models yield “more useful intelligence per watt,” so the value of compute should rise over time, not fall.

In studies published recently, Morgan Stanley (MS) showed that after-tax ROIC were quite attractive across the AI industry. For cloud service providers renting compute capacity, MS calculated ROIC at around 31%; for model builders serving intelligence through owned infrastructure, MS calculated ROIC of around 46%; and for model builders serving intelligence on rented infrastructure, MS calculated ROIC of around 25%.

Source: Morgan Stanley Equity Research note titled "We Present 3 Frameworks Detailing GenAI ROIC of 25-50%", published on July 27, 2026.

 

Our own work suggests these payback-frameworks are credible. Because capital spending comes before revenue, trailing returns will look depressed while capacity is being built. That is why we focus on unit economics: cost and revenue per watt, server payback, and contract mix. On those measures, we believe the best operators are building assets that will earn attractive returns for years to come. Jensen Huang stresses that tokens are now profitable, and Anthropic’s recent disclosure supports him: the company spent about $2.51 on compute and infrastructure for every dollar of revenue in early 2025, but only about 48 cents by mid-2026.
 

Usefulness Showing Up in Results

AI is also making existing businesses earn more. At Meta, better AI-driven ad ranking helped North American advertising growth accelerate for a second straight quarter, to 33%, with the average price per ad up 14%. At Spotify Technology S.A., AI coding tools roughly doubled engineering output, which translated into what co-CEO Gustav Söderström called a “very significant acceleration” in shipping new features, with headcount flat. Those features are paying off: Spotify’s AI-powered taste model is driving double-digit growth in music discovery and is on track to add billions of listening hours a year.
 

Our Take on the Debates: History Rhymes and Econ 101

AI skeptics concede that demand today is unprecedented; they ask what comes next. Predicting the future often depends on drawing the right lessons from the past, and investors frequently face a tension between past precedents and the claim that “this time is different.” Our take, frankly, includes a little of both. The debates are complex, spanning AI monetization, the risk of over-building, ROIC, capital availability, supply constraints, but in our view they reduce to Econ 101: the laws of supply and demand, and particularly demand. That is why a comment by Jensen Huang at GTC Berlin resonated with us: “Without demand, nothing is resolvable; with demand, everything will be resolvable.”

The crux, then, is the sustainability of demand, which turns on the slope of the curve we described at the outset: the value of AI’s usefulness over time. As long as AI grows more useful, it should generate economic value that exceeds the cost to provide it. Positive returns drive adoption and diffusion through the economy; adoption drives AI revenue; and returns that outweigh the risks keep the investment funded by business cash flows and the capital markets. If that curve is steep enough, we believe the AI cycle has a long runway. The bear case rests on a single link in that chain: what if demand slows? It is the right link to test, because with exponential growth, even a slight change in the rate produces a very different outcome. To answer it, we turn to history, mindful of the saying often attributed to Mark Twain: “History doesn’t repeat itself, but it often rhymes.”
 

Sustainability of Demand: This Time It Is Different

As detailed above, there is little debate about today’s demand. Every GPU that can be built, shipped, and plugged in is being used; unlike the dark fiber of the dot-com era, there are no dark GPUs. Even older generations, such as NVIDIA’s H100, rent for higher hourly prices than a year ago. Yet every great infrastructure build, the skeptics argue, from the railroads to the fiber networks of the internet boom, has ended in oversupply and a crash, so this one will too.

We believe the mechanics of this build are different. In the 1800s, railroad track had to be laid long before passengers and cargo filled the trains. In the 1990s, dark fiber was laid years before the applications, devices, last-mile bandwidth, and users existed to light it up. Most people did not yet own a computer, 14.4 kbps dial-up modems were painfully slow, and the internet had no installed base. Generative AI faced almost none of that friction. It arrived instantly through smartphones and web browsers already in billions of hands. Anyone who could type a text message could use it, and adopting it was often a single click. That is why AI has been adopted faster than any technology in history: ChatGPT reached 100 million users within about two months of its launch, and more than a billion people now use AI applications every week.

Nor are builders deploying capital years ahead of demand. As we noted above, Amazon’s CEO explained that cloud providers buy data center IT hardware only months before it goes into service, and SpaceX’s short-term contracts are renewing at higher prices and its new compute is paying back in under a year. In this buildout, the capital most at risk of obsolescence is committed last, against visible and contracted demand.5

Skeptics will say the executives we quoted earlier are talking their books, and the telecom executives of the late 1990s were bullish, too. The difference is that today’s leaders are reporting indisputable facts: sold-out capacity, signed contracts, and rising prices. And they are backing their words with hundreds of billions of dollars of their shareholders’ capital and with their own credibility, in the form of public guidance. The bears’ expectation of an inevitable reversal is just a forecast. So, whom should investors believe: the skeptics, or the people writing the checks? We believe the people writing the checks see something the skeptics underestimate: the size of the prize.
 

The Size of the Demand Prize: A Precedent from the Industrial Revolution

The mechanics of the AI build differ from those of the railroads and the internet, but its scale rhymes with something far larger: the Industrial Revolution. Before it, nearly all work depended upon muscle. By most estimates, human and animal muscle provided more than 90% of the world’s mechanical work in 1800; by 1950, machines provided more than 80%. 6 The shift took roughly 150 years, with the steep part of the curve lasting about a century. 7 Today, virtually all knowledge work, and the direction of much physical work, depends on human, biological intelligence. As we noted at the outset, that monopoly is ending: for the first time, we can manufacture useful intelligence.

That is why we believe AI’s ultimate addressable market is the work done by human intelligence itself. 8 As the table below shows, labor income represents roughly half of the global economy, and knowledge work alone, an estimated $42 trillion to $50 trillion a year by 2030, is about five times what the world spends on all information technology.
 

The Prize: The Global Economy in 2030 (estimated)
CategoryShare of Global GDP

$ Amount

Global GDP100%

$150 trillion

Labor income52%

$78 trillion

Knowledge work28–34%

$42–50 trillion

Physical work18–24%

$27–35 trillion

Global IT spending*6%

$9 trillion

* IT spending is shown for comparison and is not additive to the categories above.
Sources: Baron Capital estimates based on International Monetary Fund, International Labour Organization, and other sources. Figures are approximate.

 

Skeptics have criticized the addressable markets cited by SpaceX, $26.5 trillion for AI in its IPO filing, and by Anthropic, who values global knowledge work at more than $40 trillion, because they are derived from the wages paid to people doing jobs that AI might someday replace. We think that is exactly the right way to size the opportunity, because it has happened before: machines took share from organic muscle, and machine intelligence will take share from organic intelligence. The steep part of the Industrial Revolution took roughly a century. AI is diffusing at a breakneck pace. We believe it is highly likely that AI will capture or augment a significant share of the work done by human intelligence over the next 10 to 20 years.
 

Capital Supply: The Markets Are Open

Financing is a concern we watch closely. Credit has become more expensive this year, as larger U.S. government deficits compete with ever-larger AI investments for capital and the Federal Reserve has raised rates. But the markets are open, and high-quality companies are getting deals done at scale. Meta raised about $37.5 billion of common stock and mandatory convertible preferred, upsizing the common offering on strong demand. SpaceX committed to a roughly $40 billion lease of next-generation NVIDIA GPUs with investors led by Apollo and Goldman Sachs, the largest committed GPU financing to date. NVIDIA has partnered with six of the world’s largest infrastructure investors on platforms expected to raise more than $500 billion of long-term capital for the AI labs. Major banks and Blackstone have committed $63 billion to fund Broadcom-designed compute for Anthropic, which filed publicly for what may be the largest IPO in history. Notably, SpaceX’s lenders are underwriting GPUs over roughly five years, hardly the behavior of investors who view them as rapidly depreciating assets.

Spreads have widened, but mostly for weaker borrowers. As MS Research describes it, the market is showing a quality divide, not a closure: most AI-related issuance is tied to high-quality hyperscalers and semiconductor companies, while lower-quality borrowers, including some neoclouds, pay more and may need support from partners. Even assuming no growth in operating income and another $400 billion of new bonds at 7%, MS estimates the strongest hyperscalers would still cover their interest expense about 12 times over in 2027.

There is a great deal of cash in the system. On FactSet consensus estimates, 12 leading AI semiconductor, memory, and hyperscale companies are expected to generate roughly $5.5 trillion of operating cash flow in 2027 and 2028 combined, more than one prominent estimate of $5.3 trillion of global data center spending over the same two years.9
 

Consensus Cash from Operations, 2027–2028*
Company TypeCY2027ECY2028E2027–28 Total
AI semiconductors and memory$1,293$1,663$2,956
Hyperscalers and SpaceX$1,116$1,450$2,567
Total, 12 companies$2,409$3,113$5,522

* Whole-company figure; FactSet does not publish cash-from-operations estimates at the division level.
Source: FactSet consensus mean estimates. Non-December fiscal years (NVIDIA, Broadcom, Micron, Microsoft, and Oracle) calendarized by weighting overlapping fiscal-year estimates by months. Totals may not sum due to rounding.

 

Of course, not all that cash will go to fund AI capacity expansion, and supplier support for some financings, such as Broadcom’s residual value guarantees, bears watching. But this build is anchored by highly profitable companies, a very different foundation from the debt-fueled telecom build of the late 1990s, and we are comfortable that capital will be available to fund it.
 

Power Supply: A Key Bottleneck

Power is harder to come by. The leading AI builders agree that the gating factor is no longer chips but electricity and the physical buildout of data centers. Jensen Huang cites land, permitting, power generation, skilled construction labor, and local moratoriums as among the biggest obstacles to the gigawatts now planned, and Elon Musk says the limit on deploying AI is “more in power and cooling and infrastructure than it is in chips.”

As with capital, we expect the highest-quality players to secure the power they need, and increasingly they are doing so “behind the meter”: generating electricity on site, with turbines, reciprocating engines, fuel cells, and batteries, rather than waiting years for a grid connection. Behind-the-meter power is faster, and it also eases a growing political concern that data centers drive up electricity bills for households and other businesses.

The leaders are moving quickly. SpaceX is building its own power generation, casting its own turbine blades to bypass a 12- to 18-month industry backlog, and has set an ambitious goal of adding roughly 20 gigawatts of power and cooling infrastructure in 2027. Bloom Energy Corporation, a Fund holding, shows how fast onsite power can move: for Nebius in New Jersey, approvals came within three months of signing, and the first block of power two months after that. Bloom Energy’s fuel cells are clean, quiet, and use little water, and its customers now extend from hyperscalers to utilities, the U.S. Army, and international markets. Bloom Energy grew revenue 138% last quarter. And the search for power is global: as Jensen Huang puts it, if data centers aren’t built in the U.S., “every country in the world will love for it to be built out there.” Smaller developers will face longer queues and higher costs. The leaders, we believe, will get their power, in the U.S. and around the world.
 

A Really Useful Engine

We remain disciplined. We separate the signal from the noise, take the debates seriously, and keep our eyes on the slope of the curve. The evidence says demand is outrunning supply, and capital and power are available for those best positioned to use them. The prize–the work done by human intelligence–is the largest in economic history. History rhymes. This time, it is not dark fiber. It is the Industrial Revolution, and AI has become a really useful engine.

Below is a partial list of the secular megatrends we focus on:

  • AI
  • Semiconductors and semiconductor equipment
  • Cloud computing
  • Software-as-a-service
  • Digital media/entertainment/advertising
  • E-commerce
  • Minimally invasive surgical procedures
  • Cybersecurity
  • Electric vehicles/autonomous driving
  • Electronic payments
  • Robotics
  • Space technology

We continue to run a high-conviction portfolio with an emphasis on the secular trends cited and listed. Among others, during the third quarter we initiated or actively added to the following positions:

  • Digital content, commerce, and AI services: Meta Platforms, Inc.
  • Semiconductors: NVIDIA Corporation, Intel Corporation, Advanced Micro Devices, Inc., Micron Technology, Inc., Monolithic Power Systems, Inc., SiTime Corporation, and SiMa Technologies Inc.
  • Power and energy infrastructure: Bloom Energy Corporation and Forgent Power Solutions, Inc.
  • Networking and optical equipment: Lumentum Holdings Inc. and Arista Networks, Inc.
  • Data storage: Seagate Technology Holdings plc
  • Autonomous transportation: Tesla, Inc.

Top Contributors & Detractors

Top contributors to performance for the quarter
 Contribution to Return
(%)
Rubrik, Inc.0.96 
NVIDIA Corporation0.87 
Shopify Inc.0.81 
CrowdStrike Holdings, Inc.0.49 
Cloudflare, Inc.0.40 

Rubrik, Inc., a cybersecurity and data resilience platform, contributed to performance amid increased cybersecurity spending. A series of high-profile agentic AI security incidents has elevated cyber resilience and data recovery among enterprise security priorities. Rubrik is a direct beneficiary of this shift, as its platform unifies backup, identity, and recovery to help organizations rapidly restore operations after breaches. Fiscal second quarter results reinforced the trend, with net new subscription ARR growing 35% year-over-year, roughly 27% above consensus, while subscription ARR contribution margin expanded to 14% from 9.4% a year ago. Management raised full-year guidance across all metrics based on the growing pipeline and expanding portfolio of security products they can cross-sell into the installed base. We retain long-term conviction in Rubrik as enterprises increasingly prioritize resilience alongside prevention.

Shares of NVIDIA Corporation, the leading provider of accelerated computing platforms for AI, contributed to performance. As discussed in our AI update, demand is shifting toward inference and agentic workloads, and NVIDIA’s full-stack platform, spanning accelerators, CPUs, networking, and software, is well suited to that shift. Its scale is also an advantage in the supply chain, where its purchasing power with TSMC, memory suppliers, and packaging partners secures priority access to the industry’s most constrained components. With its Vera Rubin platform now in full production, NVIDIA introduced a containerized “AI factory” design that can bring pre-integrated racks online in about a month, rather than four to five months, once power is available. With each gigawatt generating roughly $40 billion of annual revenue, by management’s estimate, that speed is worth billions of dollars to customers. NVIDIA remains the Fund’s second-largest holding, and we retain strong conviction in its role as the leading enabler of the AI buildout.

Shares of Shopify Inc., a leading commerce platform powering millions of merchants, contributed to performance after the company delivered another quarter of acceleration at scale. Gross merchandise volume grew more than 31% and revenue rose nearly 34%, with expanding margins, and management’s third-quarter guidance came in well ahead of Street expectations. We believe Shopify is also emerging as a central beneficiary of agentic commerce. AI shopping agents depend on clean, structured product data, and Shopify’s catalog spanning millions of merchants provides exactly that, with its storefronts now powering shopping experiences within ChatGPT, Microsoft Copilot, and Google. Early results are encouraging, as 75% of AI-attributed orders came from outside the top 100 product categories. This suggests that agent-driven discovery rewards product fit over advertising spend and opens demand to the long tail of merchants. We believe Shopify is becoming the infrastructure layer for AI-driven commerce and remain confident in its long runway for growth.

Top detractors from performance for the quarter
 Contribution to Return
(%)
Lam Research Corporation(2.35) 
Space Exploration Technologies Corp.(1.54) 
Coherent Corp.(1.00) 
Tesla, Inc.(0.62) 
Axon Enterprise, Inc.(0.54) 

Despite strong operating results, shares of Lam Research Corporation declined during the quarter, giving back a portion of the gains that made it the Fund’s largest contributor in the prior period. Lam, a leading supplier of etch and deposition equipment to the semiconductor industry, reported revenue and earnings ahead of guidance, raised its outlook for calendar 2026 wafer fabrication equipment (WFE) spending to the low $150 billion range, and guided to another step-up in September quarter revenue as memory investment accelerated. Weighing on sentiment were profit-taking across the semiconductor equipment group following first half performance run and periodic concerns about export controls on China, which represents roughly a third of Lam’s revenue. We remain focused on the long term. Recent industry data point to WFE growing 30% or more annually for at least the next two years, anchored by Dynamic Random-Access Memory (DRAM) capacity plans that extend through the end of the decade, and October might mark the first month of this cycle in which equipment suppliers have been unable to meet planned shipment slots. Because high-bandwidth memory and high-layer-count NOT AND (NAND)require far more etch and deposition steps per wafer than the technologies they replace, Lam captures a growing share of every dollar of memory spending with each node transition. We believe Lam remains of the higher quality way to participate in this cycle, where we continue to expect strong earnings growth, and have maintained our position.

Space Exploration Technologies Corp. develops and launches advanced rockets, satellites, and spacecraft, and builds AI software and infrastructure, with the long-term goal of making humanity multiplanetary. The stock was a detractor as shares pulled back from their post-IPO highs, despite very strong underlying fundamentals. Its first earnings report as a public company underscored that strong momentum. Revenue nearly doubled year-over-year, driven by rapidly scaling compute-hosting deals with AI leaders including Anthropic and Google, alongside continued Starlink growth. Losses, too, narrowed sharply, aided by high incremental margins. Moreover, management guided to an astonishing $100 billion annualized revenue run-rate by year-end 2026, roughly triple the latest quarter’s annualized run-rate, and is targeting up to 5 times more compute capacity exiting 2027, well ahead of expectations. SpaceX also closed its acquisition of Cursor, which is helping to accelerate product innovation, with viral hits like Grok Bot and faster releases of advanced Grok models. Since the quarterly report, the company has expanded its compute pipeline with additional multibillion-dollar hosting deals, lifting annual contracted revenue to over $50 billion from effectively zero just months ago. With additional test flights, the team also advanced Starship toward rapid, full reusability, including its first orbital mission and its first revenue-generating flight, launching Starlink V3 satellites into operational orbit. We believe these developments position SpaceX at the center of two defining secular trends, the commercialization of space and the build-out of AI infrastructure, and we remain long-term holders, convinced SpaceX is in the early days of a multi-decade growth opportunity.

Shares of Coherent Corp., a leading photonics company serving the data center and industrial markets, detracted from performance during the quarter. Part of the decline reflected a broad sell-off in optical stocks amid concerns about the durability of AI spending, but company-specific factors also weighed on the shares. Coherent is still early in ramping optical circuit switches and does not expect scale-up optics revenue until the second half of 2027, while Lumentum Holdings Inc., which we also own, is already shipping meaningful volumes into both. Near-term gross margin guidance also called for only modest improvement, as a mix shift toward lower-margin transceivers offset early benefits from the six-inch indium phosphide ramp. We continue to view Coherent as a leader in photonics with one of the industry’s broadest portfolios and a differentiated vertically integrated model. It is a leading Western supplier of indium phosphide and gallium arsenide lasers, which are critical for high-speed transmission and in short supply, and its global manufacturing footprint is an advantage in the current geopolitical environment. We expect gross margins to expand as six-inch yields improve and higher-value products reach volume. We trimmed the position given the stock’s strong run and more competitive near-term setup, but Coherent remains a meaningful holding.

Portfolio Structure

We invest in companies of any market capitalization that we believe will deliver durable growth from the development, advancement, and/or use of technology. We invest principally in U.S. securities but may invest up to 35% in non-U.S. securities.

At the end of the third quarter, the largest market cap holding in the Fund was $5.5 trillion and the smallest was $0.7 billion. The median market cap of the Fund was $112.0 billion and the weighted average market cap was $1.4 trillion.

We had investments in 40 unique companies at the quarter-end. Our top 10 positions accounted for 55.8% of net assets.

To end the quarter, the Fund had $257.5 million in net assets. Flows were positive in the second quarter.

Top 10 holdings
 Quarter End Market Cap
($B)
Quarter End Investment Value
($M)
Percent of Net Assets
(%)
NVIDIA Corporation5,504.0 19.0 7.4 
Lam Research Corporation411.1 18.4 7.1 
Taiwan Semiconductor Manufacturing Company Limited2,366.0 17.7 6.9 
Space Exploration Technologies Corp.2,047.6 16.5 6.4 
Alphabet Inc.4,188.2 15.9 6.2 
Micron Technology, Inc.1,202.9 13.0 5.0 
Spotify Technology S.A.100.3 12.1 4.7 
Advanced Micro Devices, Inc.998.7 11.2 4.3 
Amazon.com, Inc.2,687.4 11.0 4.3 
Shopify Inc.190.6 8.9 3.5 
Fund investments in GICS industries
 Percent of Net Assets
(%)
Semiconductors & Semiconductor Equipment41.6    
Software8.9    
Interactive Media & Services8.8    
Diversified Telecommunication Services6.4    
IT Services4.9    
Entertainment4.7    
Broadline Retail4.3    
Electrical Equipment3.9    
Communications Equipment3.7    
Automobiles3.1    
Health Care Providers & Services2.2    
Electronic Equipment Instruments & Components2.0    
Aerospace & Defense1.4    
Construction & Engineering1.4    
Technology Hardware Storage & Peripherals1.3    
Health Care REITs0.7    
Independent Power And Renewable Electricity Producers0.5    
Cash and Cash Equivalents0.1    
Total100.0* 

* Individual weights may not sum to the displayed total due to rounding.

Recent Activity

Top net purchases for the quarter
 Quarter End Market Cap
($B)
Net Amount Purchased
($M)
NVIDIA Corporation5,504.0 5.7 
Intel Corporation635.4 5.3 
Meta Platforms, Inc.1,847.4 4.5 
Lam Research Corporation411.1 4.3 
Alphabet Inc.4,188.2 4.2 

During the quarter, we initiated a position in Intel Corporation, the only U.S.-headquartered company that both designs and manufactures leading-edge semiconductors. Our thesis centers on the growing need for a second source of advanced chip manufacturing. The AI buildout has created acute shortages of leading-edge manufacturing capacity, nearly all of it supplied by TSMC in Taiwan, and chip designers increasingly want an alternative, one that only Intel can offer on U.S. soil. After years of lost manufacturing leadership and market share, we believe Intel has reached a turning point under CEO Lip-Bu Tan. Its newest manufacturing process is now in high-volume production, with output running ahead of plan, and its next process, the first Intel has designed from the outset for outside customers, is tracking ahead of schedule, with management expecting the first customer commitments by early 2027. Intel also stands to benefit as AI agents expand the role of CPUs in the data center, where it remains the largest supplier. Its outside manufacturing business is still nascent, and winning customers will determine whether the turnaround lasts.

We initiated a position in Seagate Technology Holdings plc this quarter. Seagate is one of the largest makers of high-capacity hard disk drives, which remain the lowest cost way for cloud providers to store the growing volumes of data created by AI applications. Over many years, the company developed heat-assisted magnetic recording (HAMR), a technology that stores more data on each disk without adding disks or heads to the drive. HAMR products now make up roughly 40% of Seagate’s data center capacity shipments, and its newest 44-terabyte drive is ramping with the two largest cloud providers. Because these capacity gains come with largely flat material costs, we believe the transition supports better pricing and profitability than in past cycles

During the quarter, we added to our position in Meta Platforms, Inc., as its AI investments increasingly translate into financial results and its new personal AI agent opens a potential new platform. In September, Meta launched Muse, which acts on a user’s behalf rather than simply answering questions. Muse reached more than 10 million setups in the U.S. and Canada in about six weeks, and more than a quarter of active users are already shopping through it, with average purchases above $200. A global launch, limited for now by available compute, would put Muse in front of the more than 3.6 billion people who use Meta’s apps daily, and CEO Mark Zuckerberg expects to monetize it sooner than Meta’s usual build-then-monetize playbook, through commerce fees and subscriptions. Third quarter results, reported in October, reinforced our view: revenue grew 26% to $64.6 billion, and U.S. and Canada advertising growth accelerated for a second straight quarter, to 33%, as AI-driven improvements to recommendations and ad ranking lifted the average price per ad 14%. We believe Muse could become Meta’s next major platform.

We also increased positions in NVIDIA Corporation, Lam Research Corporation, and Alphabet Inc., primarily by putting new inflows to work across the portfolio in proportion to existing position sizes, which meant these top holdings received a correspondingly large share of capital.

Top net sales for the quarter
 Quarter End Market Cap or
Market Cap When Sold 
($B)
Net Amount Sold 
($M)
Broadcom Inc.1,676.5 5.1 
Applied Materials, Inc.408.2 1.7 
Axon Enterprise, Inc.34.3 1.4 
Coherent Corp.56.4 0.2 
INNIO N.V.14.0 0.1 

We reduced our position in Broadcom Inc. The reduction reflects position sizing rather than a meaningful change in our view. Broadcom’s AI semiconductor revenue rose more than 221% in its fiscal third quarter to $16.7 billion, and management guided the business to triple again in the fiscal fourth quarter, and to double in 2027, to $115 billion, and again in 2028, to $230 billion, anchored by multi-year commitments from Google, Anthropic, and OpenAI. We believe Broadcom remains the leading custom accelerator and AI networking franchise.

During the quarter, we exited our position in Applied Materials, Inc. to concentrate our semiconductor equipment investments in Lam Research Corporation and ASML Holding N.V., both of which are more directly exposed to the expansion of memory manufacturing. We reduced Axon Enterprise, Inc. following strong performance, and Coherent Corp. to help fund the networking and optical purchases noted above. Finally, we sold our small position in INNIO N.V. after receiving only a minimal allocation in its IPO, preferring to deploy the capital elsewhere.
 

Looking Ahead

The Fund trailed its Benchmark in the third quarter, but we enter the final quarter of the year confident in the fundamental trajectory of our holdings. On the portfolio construction side, we added to our highest-conviction AI compute and digital platform names and concentrated our semiconductor equipment exposure in the names most leveraged to memory. We believe these changes leave the Fund well positioned going forward.

We remain focused on owning category-defining technology businesses that sit at the heart of durable, secular growth trends. Led by visionary, execution-driven management teams, these companies convert breakthrough innovation into expanding free cash flow. We believe this combination uniquely positions the Fund to compound our investors’ capital over the long term.

Sincerely,

Portfolio Manager Michael Lippert signature
Michael A. LippertHead of Technology Research, Portfolio Manager
Portfolio Manager Ashim Mehra signature
Ashim MehraPortfolio Manager

Featured ETF

Learn more about Baron Technology ETF.