Rebalancing the WisdomTree Artificial Intelligence and Innovation Index
Every technology buildout eventually changes the question investors ask. In the early innings, it is who is spending. Later, it becomes who is actually getting paid, and who is borrowing to keep up. We believe AI has reached that second stage, and our ad hoc rebalance of the WisdomTree Artificial Intelligence and Innovation Index1 is built around it.
Three ideas run through the changes.
- First, we want more exposure to companies that turn AI into revenue and can fund their own ambitions, and less to businesses whose results depend on capital markets staying generous.
- Second, we want to own the bottlenecks, which is to say the scarce chips, tools, components and power that AI cannot scale without.
- Third, we want discipline on price, taking profits where success is already assumed and refusing to mistake ‘cheap’ for ‘good.’
The rebalance turns over roughly 11% of the portfolio, which is meaningful but deliberately focused. The core is intact. NVIDIA (5.4%), Micron Technology (4.9%), Meta Platforms (4.9%) and Amazon (4.4%) remain the largest positions, and each already reflects these ideas, for example:
- NVIDIA’s leadership in AI system economics
- Micron’s tight supply of high-bandwidth memory
- The advertising cash flow that funds Meta’s AI investment
- The AI demand running through Amazon’s cloud business
Why does the line between getting paid and borrowing matter for an AI strategy? Spending announcements are easy to count, but they say little about who keeps the profit. A company can be central to the buildout and still be a poor investment if it must borrow to participate, if its customers hold the pricing power or if the market already assumes everything goes right. The rebalance is an attempt to sort the portfolio along those lines, rather than simply by who is most exposed to AI.
Follow the Cash, Not the Capital Call
The biggest single change is a new position.
Microsoft (3.0%)
Microsoft sits where AI demand converts into recurring revenue, through Azure AI workloads and the adoption of Copilot and its E5 enterprise suite, with its own Maia chips offering inference efficiency. The tests ahead are real, since paid adoption and the returns on infrastructure spending still have to show up, but Microsoft can fund those tests from its own earnings.
We also added to Alphabet (4.2%), where cloud growth, TPU integration and Gemini-powered enterprise agents are paired with a valuation that looks appealing relative to peers. The risk we are watching is that agentic search could disrupt advertising economics before new revenue scales. Smaller additions to Oracle (3.3%) and SoftBank (1.8%) extend our reach on the demand side, through Oracle’s compute backlog and SoftBank’s OpenAI and Arm holdings. Both lean on leverage, so we sized them accordingly.
Even the strongest balance sheets face a timing problem. Capital spending and compute costs arrive before the revenue they are meant to create, which is why we weigh evidence of monetization and not just spending plans. The difference with companies like Microsoft and Alphabet is that they can absorb that gap without needing anyone in the lending markets’ permission.
The proceeds came largely from the part of the AI complex where financing, rather than demand, is the binding constraint. We exited CoreWeave, Nebius, DigitalOcean and Cerebras, along with server maker Wiwynn, together roughly 4.7% of the portfolio. Demand for GPU capacity is real, and in places rental pricing is firming. But in these businesses, equity holders sit behind lenders, and delays are expensive. When a build slips, revenue is deferred while interest keeps accruing. Cerebras leans on a single anchor customer in OpenAI and still has yields and a share lock-up overhang to work through. Wiwynn sits in the same supply chain, with thin server-assembly margins, customer concentration and exposure to any reset in its customers’ capital spending. We do not need to be bearish on the category to prefer owning the balance sheets over the borrowers.
Own the Bottlenecks
AI’s growth is limited less by ideas than by physical constraints. These include leading-edge chips, advanced packaging, lasers, manufacturing tools and electricity. Where those are scarce, suppliers hold pricing power, which is why most of the remaining additions sit here.
Taiwan Semiconductor Manufacturing (4.2%) combines leading-edge foundry capability with scarce advanced-packaging capacity and a strong balance sheet. Lumentum (2.8%) makes the high-speed lasers that move data between chips in AI clusters, and qualified capacity for them is in short supply. Behind every advanced chip stand the equipment makers, whose workloads grow with each step up in complexity. Some of the critical names include ASML (2.5%) in extreme ultraviolet lithography, Applied Materials (2.0%) across logic and memory, and Tokyo Electron (1.7%) in coating, etching and deposition. Complexity is the key word. Each new generation of logic and memory chips requires more process steps, and those steps run through a small number of suppliers. That ties their fortunes to the overall pace of leading-edge manufacturing rather than to any one chip design winning share. Their risks are equally concrete: customer budgets must confirm the growth, and ASML in particular faces production limits and China restrictions.
Then there is power. Bloom Energy (2.2%) provides on-site fuel cells that address the time-to-power problem facing data centers that cannot wait for the grid. Vertiv (1.8%) benefits as denser racks and higher-voltage architectures raise the power and cooling content of every data hall. Each carries execution risk. Bloom must convert backlog into delivered systems, and Vertiv’s near-term earnings depend on data-center timelines. Most of these are additions of 0.5 to 1.0 percentage points to themes already in the portfolio, not new concepts being added.
Price Discipline, in Both Directions
Scarcity does not excuse any price. Strong performance is not, by itself, a reason to sell, and weak performance is not a reason to buy. What matters is what has to go right from here.
Marvell Technology (1.3%) is the clearest example. The stock has performed strongly over much of 2026, and we trimmed the position while keeping a stake in its optics ramp and custom-chip wins, against a backdrop of intense competition.2 Broadcom (4.0%), meanwhile, has fallen about 17% since roughly the midpoint of 2026, even though its TPU backlog, custom-accelerator wins and Tomahawk switching products continue to drive AI growth.3 We added to Broadcom and to Credo Technology (2.0%), whose cables and optical chips benefit as AI clusters grow larger, and whose valuations have reset. In effect, we swapped a stock that has run for stocks with better starting points, within the same theme of AI connectivity and custom silicon.
The same logic applied to software and security. Palo Alto Networks (3.2%) has risen about 42% since roughly the middle of 2026, and Snowflake (2.0%) and CrowdStrike (0.9%) have also run.4 Their stories are strong, from platform consolidation in security to enterprise AI products at Snowflake. But premium valuations demand sustained, faster growth, so we trimmed each. Palo Alto remains a top-ten position. Cisco (1.1%) was trimmed in the same spirit.
Adobe shows the other side of the discipline. Its valuation is low, yet we exited, because AI could erode its design franchise before Firefly and GenStudio monetization can offset the damage. Cheap alone is not a thesis, and this one has been taking some time to play out.
What Could Go Wrong
No rebalance removes risk, and this one carries its own. Custom silicon is the key threat to NVIDIA even as it benefits Broadcom, so the portfolio sometimes sits on both sides of the same trade. Taiwan Semiconductor depends on customer spending and carries geopolitical exposure. Premium valuations at Lumentum and Bloom leave little room for delivery delays. Memory prices, supported today by tight supply and long-term agreements, could normalize faster than expected as Chinese supply grows. And the platforms we favor could slow their own spending if monetization disappoints. A portfolio built around who gets paid is only as sound as the assumption that the payments continue.
What Ties It Together
Taken together, the changes tilt the WisdomTree Artificial Intelligence and Innovation Fund (WTAI) toward companies that are paid for AI today or that hold something AI cannot do without, and away from those that depend on borrowed money or a flawless future. Not everything moved. Memory, including Samsung Electronics (4.1%), SanDisk (2.6%), Kioxia (2.4%) and SK hynix (2.0%) alongside Micron, remains a large scarcity position and was left unchanged.
For investors, the practical takeaway is a portfolio with a clearer shape. One group of holdings is made up of platforms that are monetizing AI and funding their own growth. A second group owns the scarce inputs, from foundry and packaging capacity to lasers, tools and power. A third group, in connectivity and custom silicon, was repositioned toward better entry points. The names we trimmed or exited share a common feature: their outcomes depend more on external funding or on expectations staying high than on the underlying demand for AI itself.
The position changes are modest by design. None is a call that the AI cycle ends or accelerates. Each is a view on which parts of it compound, and which parts have to borrow, or be perfect, to keep up.
1 The WisdomTree Artificial Intelligence and Innovation Fund (WTAI) is designed to track the total return performance, before fees and expenses, of the WisdomTree Artificial Intelligence and Innovation Index. Each time we mention a company that is held within the WisdomTree Artificial Intelligence and Innovation Index as of the ad hoc rebalance, we note the weight in parentheses. This is the targeted weight as of the October 21, 2026 rebalancing.
2 Sources: Bloomberg, FactSet.
3 Sources: Bloomberg, FactSet.
4 Sources: Bloomberg, FactSet.
For more news, information, and strategy, visit the Modern Alpha Content Hub.
