The AI Trade After the Obvious Part

The easy AI thesis is no longer scarce. Everyone knows the large models matter. Everyone knows @Nvidia matters. Everyone knows hyperscaler capex matters. Everyone can say “agents,” “compute,” “power bottleneck,” and “data centers” without adding much information.

That does not make the next three months uninteresting. It changes what is worth looking for.

The Q3 2026 AI trade is less about discovering that AI is real and more about finding the public tickers that become overflow valves for demand that cannot yet buy the private labs directly. The obvious objects of desire are OpenAI, Anthropic, $SPCX /xAI, and the companies getting paid by them. Most investors still cannot buy those labs. So they buy wrappers, proxies, suppliers, and companies that can show AI is already changing their cost structure.

1. The access trade: public wrappers for private AI scarcity

The cleanest near-term frenzy is likely to be around any public security that offers even partial exposure to the big private AI names.

That means things like $DXYZ, $RVI, and weird proxy tickers like $SKM. These are not simple value investments. They are access trades. Their appeal comes from the fact that the thing people want is hard to buy.

Q3 has the right setup for this. Reuters has reported that @OpenAI is preparing a confidential IPO filing and could go public as early as September 2026, while @Anthropic has hired @wilson-sonsini for possible IPO preparation, though Anthropic says no decision has been made about when or whether to list. Reuters has also reported a @SpaceX listing process that could begin even earlier, with SpaceX/xAI sitting in the same private-mega-tech scarcity complex. (Reuters)

That creates a simple behavioral trade: every IPO rumor, filing headline, secondary mark, tender offer, valuation update, lockup detail, index-inclusion discussion, or fund-holding disclosure can send money searching for a public thing to buy.

$DXYZ is the purest public-market expression of this behavior. It is a closed-end fund listed on the NYSE, holding a portfolio of venture-backed private technology companies. As of its latest public materials, its largest AI exposures included @Anthropic, $SPCX, and @OpenAI. A recent portfolio page listed Anthropic at 18.1%, $SPCX at 14.5%, and OpenAI at 5.8% of holdings. (Destiny)

The catch is the price. $DXYZ reported NAV of $24.56 per share as of March 31, 2026. The stock recently traded around $60.21, which is roughly a 145% premium to that NAV by simple arithmetic. That is absurd if treated as a normal closed-end fund valuation. It is also exactly why the ticker matters for Q3: people are not only paying for assets; they are paying for access, immediacy, and a public-market button labeled “private AI.” (Stock Titan)

$RVI, the RobinHood ventures fund I, is a newer version of the same idea. It priced its IPO at $25 per share, began trading on the NYSE under $RVI in March 2026, and was designed to give public investors exposure to private companies without accreditation requirements or large minimums. Robinhood later said $RVI purchased about $75 million of @OpenAI common stock, making OpenAI one of the fund’s largest investments. The disclosed portfolio also included names such as @Databricks, @ElevenLabs, @Ramp, @Revolut, @Stripe, and others. (Robinhood)

$RVI also has the same structural danger as $DXYZ. It is a closed-end fund, so shares can trade at a premium or discount to NAV. Robinhood’s own disclosures emphasize that the fund is speculative, concentrated in private companies, has valuation uncertainty, offers no redemption right, and has no guarantee that portfolio companies will ever have liquidity events. That is not a reason the trade cannot work. It is the reason the trade has to be sized and understood as a frenzy instrument rather than a clean look-through claim. (Robinhood)

$SKM is messier, which is part of the appeal. SK Telecom disclosed a $100 million investment in Anthropic and a partnership to build a telco-customized large language model. The exact current ownership after later Anthropic funding rounds is not the clean public fact. The clean public fact is simpler: $SKM is a tradeable ADR with an Anthropic story attached, and its market cap is small enough that narrative attention can matter more than it would for a megacap. (SK Telecom)

The Q3 view on these tickers is deliberately narrow: they can be bad long-term values and good short-term trading vehicles. The driver is not NAV purity. The driver is public-market desperation for exposure to private AI assets before the normal #IPO window fully opens.

2. The operating trade: companies using AI to change the shop

The second bucket is companies that are using AI well inside the business, not merely selling an AI story to investors.

Shopify is the example.

$SHOP is interesting because the AI claim is attached to operating behavior. Tobi Lütke’s internal memo made AI usage a baseline expectation, and reporting around the memo said teams would need to show why AI could not do the work before asking for more headcount or resources. That is the right test. The company is not asking investors to believe in a demo. It is pushing AI into the daily production function. (The Verge)

The numbers give the story room to breathe. In Q1 2026, $SHOP reported GMV above $100 billion, revenue of $3.17 billion, revenue growth of 34%, and free cash flow of $476 million, equal to a 15% free-cash-flow margin. For Q2, $SHOP guided to high-twenties revenue growth and a mid-teens free-cash-flow margin. (Shopify)

That is why $SHOP belongs in this basket. The Q3 question is not “does Shopify have AI features?” The question is whether AI lets $SHOP keep growing fast while bending the cost curve. Watch operating expenses as a percentage of revenue, support costs, product velocity, merchant acquisition, merchant retention, and attach rates for higher-value services.

This is the higher-quality version of the AI trade. It does not need an #IPO rumor. It needs proof that AI lets a company do more work with fewer incremental people.

3. The supplier trade: public companies getting paid by the private labs

The third bucket is suppliers to the companies that are either private, preparing to go public, or about to become more visible.

$GOOG is the more interesting supplier expression because it is both competitor and vendor. Anthropic announced an expansion with Google Cloud involving up to one million TPUs, worth tens of billions of dollars, with well over one gigawatt of capacity expected to come online in 2026. Reuters also reported that @OpenAI planned to add Google Cloud as a provider for training and running models. (Anthropic)

That creates a nice asymmetry in the story. If the market worries that #google-search is threatened by AI, Alphabt can still benefit from #AI-lab compute demand through Google Cloud and TPUs. The supplier angle does not require Google to be the only model winner. It requires Google to sell scarce infrastructure to model companies with enormous capital needs.

$AMZN fits too, especially throughAnthropic. @Anthropic announced a new agreement with $AMZN securing up to 5 gigawatts of capacity, representing more than $100 billion of AWS demand over 10 years, alongside Amazon’s expanded investment commitments. (Anthropic)

For Q3, the supplier trade should be judged by contracted demand, backlog, capacity availability, and whether lab IPO excitement improves the perceived durability of that spend. If @OpenAI or @Anthropic come to market at huge valuations, the read-through to suppliers is simple: these customers are better financed, more visible, and more likely to keep buying compute.

The actual Q3 shape

The clean expression is a three-part book.

First, a scarcity/access basket: $DXYZ, $RVI, $SKM. This is the most reflexive piece. It is also the most dangerous. These tickers can move on headlines that have little to do with current NAV. The right mental model is event volatility around private AI exposure.

Second, an AI-operator basket: $Shopify as the lead example. The question is whether AI adoption shows up in operating discipline. Revenue growth is nice. Revenue growth with restrained headcount and durable free cash flow is the thing that makes the market believe.

Third, a lab-supplier basket: $GOOG at the center, with $AMZN also fitting through @Anthropic. This bucket is less about finding an undiscovered AI winner and more about owning the public cash registers that private labs are already using.

The key distinction: this memo is not bullish on “AI” in the abstract. It assumes the abstraction has been bid up. It is bullish on places where Q3 can still produce mechanical demand, visible operating proof, or supplier receipts.

What would break the thesis

The access trade breaks if the IPO window slips, if @OpenAI or @Anthropic filings disappoint, if SpaceX/xAI enthusiasm fades after listing mechanics become clearer, or if $DXYZ /$RVI premiums become too stretched even for scarcity buyers.

The $SHOP trade breaks if we see ordinary opex creep, weakening GMV quality, or AI rhetoric without measurable productivity.

The supplier trade breaks if investors decide the AI buildout is becoming circular financing rather than durable demand. It also weakens if labs shift faster than expected toward alternative chips, internal silicon, or lower-cost inference architectures that reduce the take rate for the obvious suppliers.

Bottom line

The Q3 2026 AI memo is: the obvious trade is priced, but the access trade is not exhausted.

People will still pay up for anything that looks like public exposure to @OpenAI, @Anthropic or the checks those companies are writing. $DXYZ and $RVI are the pure frenzy vehicles. $SKM is the odd public proxy. $SHOP is the real-operator case. $GOOG and $AMZN remain supplier expressions because private-lab ambition keeps turning into public-company revenue.

The market already believes in AI. Q3 is about who gets to sell shovels, who gets to sell tickets, and who can prove they are using the machinery better than everyone else.