App Economy Insights — Wall Street's Top Stocks in Q2: the AI trade broadened beyond NVIDIA

Cerebras builds AI computers the hard way: instead of cutting a silicon wafer into hundreds of separate chips, it uses the entire wafer as one enormous processor. That removes most of the wiring between chips, which is why its systems are pitched as unusually fast at inference — running an already-trained AI model to answer questions, as opposed to training it in the first place.

It appears here because it was the standout new name in Q2's 13F filings: the #1 new position for both Altimeter and Tiger, with large stakes also disclosed by Coatue and Atreides, and a top-five buy at four funds overall — level with Taiwan Semiconductor. One honest caveat App Economy attaches: all four were already private-market shareholders before the May IPO, so part of what the filing shows is old exposure becoming visible rather than fresh buying.

The reason it matters beyond the name itself is what it represents. Four sophisticated AI investors putting a non-GPU compute company at the top of their buy lists is the sharpest single piece of evidence for the article's thesis that the AI trade is broadening past NVIDIA.

Astera Labs makes the unglamorous parts that let the glamorous parts talk to each other — connectivity chips that shuttle data between GPUs, CPUs, memory and the network inside an AI server rack. Think of it as the on-ramps and interchanges of a highway system: you can add lanes forever, but if the interchanges jam, nothing moves faster.

Two separate funds ranked it at the very top of their quarter: Light Street made it their #1 buy, Whale Rock their #2. The thesis App Economy gives is structural rather than about the quarter's earnings — "as AI clusters become larger and more complex, connectivity is becoming an increasingly important bottleneck."

That is the useful generalisation. When a system scales, the constraint moves. Compute was the constraint; as clusters grow to tens of thousands of accelerators, moving data between them becomes the limit — and whoever sells the fix gets a bigger share of the budget.

Ferrari is classified as an automaker but does not behave like one. It deliberately builds fewer cars than people want, sells heavy personalization on top, and prices with the confidence of a luxury house — which is why its profit margins look more like a handbag brand's than a car company's.

Viking made it their #1 buy of the quarter, the most prominent non-AI conviction bet in the entire round-up. App Economy's read on the logic: "Viking may be betting that this scarcity-driven model can keep compounding regardless of broader pressure in the auto industry."

In plain terms, this is a bet on the business model insulating the stock from its sector. If tariffs, EV price wars and weak consumer demand hurt car companies, a manufacturer with a multi-year waiting list simply doesn't face the same problem — it is supply-constrained by choice.

Kilroy is a real-estate investment trust — a listed landlord — that owns high-end office buildings clustered in West Coast tech markets. Offices have been the most disliked corner of commercial property since remote work emptied them, and Kilroy's buildings sit in exactly the cities where that hit hardest.

Route One made it their #1 buy. The contrarian argument App Economy relays is that "improving tech and AI leasing could help fill still-elevated vacancies, creating meaningful upside if the office recovery continues."

Why the upside could be outsized: a landlord's costs are largely fixed and much of its debt is fixed too, so the rent from newly-filled space drops through to profit with little offsetting cost. Add a depressed starting price, and a partial recovery in occupancy can move the equity a long way. It is also the round-up's most indirect AI trade — owning the buildings the AI hiring boom would have to fill.

TSMC is the foundry that actually manufactures the chips other companies design — NVIDIA's accelerators, Apple's processors, AMD's and Broadcom's silicon all come off its lines. It sells to every side of the AI race at once, which makes it the closest thing to a toll booth on AI compute.

In Q2 it became, jointly with Amazon, the most widely held stock across the 20 funds — a top-five position at nine of them — and was also a top-five buy at four. Being simultaneously the most-held and among the most-bought is rare; it means the funds that already own it are still adding.

This is the cleanest expression of the article's headline. If you believe AI demand is real but can't be certain which chip designer wins, you buy the factory they all have to use.

Amazon is two businesses in one: the retail marketplace everyone sees, and AWS, the cloud-computing arm that rents out servers and now AI capacity — where most of the profit comes from.

It is tied with TSMC as the most widely held name across the 20 funds, a top-five position at nine of them. The move worth studying is Druckenmiller's: he had nearly eliminated Amazon in Q1 and this quarter rebuilt it to 542,000 shares, alongside reopening Alphabet.

App Economy's own gloss on that reversal is the lesson rather than the trade — it shows "just how quickly his positioning can change." A quarterly filing is a photograph of a portfolio that may already look completely different by the time you read it.

Alphabet is Google — search, YouTube, Android, the Gemini models and Google Cloud. It spent much of the last two years as the mega-cap investors argued about most, on the theory that AI chatbots would erode search.

Q2's filings show that argument resolving in one direction. It is the third-most widely held name (a top-five position at eight of the 20 funds); Berkshire increased its stake by 83%, making Alphabet its third-largest holding behind only Apple and American Express; and Druckenmiller reopened a 336,000-share position after having exited entirely one quarter earlier.

The dissenting vote is worth keeping in view: Ackman finished selling out of Alphabet, completing a rotation he began the previous quarter. Two respected managers took opposite sides of the same name in the same three months — which is the practical argument against treating any one 13F as a signal.

NVIDIA sells the accelerator chips that train and run most large AI models, and has been the default way to own the AI theme since it began.

The Q2 filings show it moving from the trade to a holding. It is still a top-five position at five of the 20 funds — but it was a top buy for only one, while Taiwan Semiconductor and Cerebras were top buys at four each, and Applied Materials, Micron and AMD at three each.

That gap between "widely held" and "actively bought" is the whole headline: the funds have not turned against NVIDIA, they have decided the next incremental dollar is better spent on the suppliers, the memory and the alternatives. Nothing here says the business is deteriorating; it says the crowd has already arrived.

Micron makes memory chips — including the high-bandwidth memory that is stacked next to AI accelerators and feeds them data. An AI chip without enough fast memory is a fast engine with a narrow fuel line.

It was a top-five buy for Coatue, Altimeter and Sands, part of what App Economy calls memory and storage "moving into focus": "as AI infrastructure scales, funds are increasingly looking beyond compute to the components that feed and store all that data."

One dissent to hold alongside it — Druckenmiller exited Micron in the same quarter. Memory has historically been a brutally cyclical business, so reasonable investors disagree about whether the AI build-out has genuinely changed that. Three buyers against one seller is a lean, not a verdict.

Seagate makes hard drives — the high-capacity mass storage where data centres keep the enormous volumes of data that AI systems are trained on and that their outputs generate. Unglamorous, cyclical, and until recently not thought of as an AI stock at all.

It appeared as a top-five buy for Lone Pine, Sands and Tiger — three funds independently reaching the same conclusion in the same quarter.

Taken with Micron's three buyers, six of the twenty funds bought into memory or storage in Q2. That kind of clustering across separate managers is a stronger piece of evidence than any single fund's top pick, because it is much harder to explain as one manager's idiosyncratic view.

AMD is the main listed alternative to NVIDIA in AI accelerators, as well as a serious competitor in the ordinary server processors that sit beside them.

It was a top-five buy at three funds — more buying attention than NVIDIA itself received in the quarter, which is the point rather than a claim that AMD is the better business.

The pattern is what to take away: when a market leader's position is fully priced, institutional money often looks for the credible second source. That is a positioning decision about valuation and crowding, not a technology call.

SpaceX went public in June 2026, so this quarter is the first in which funds holding it had to disclose the position publicly. On the surface that produced a dramatic result: it appeared among the top new positions at five of the 20 funds.

App Economy immediately defuses it. "Several were already private-market investors before the June IPO, so the filings largely reveal existing SpaceX exposure becoming publicly reportable, rather than necessarily new Q2 buying."

This is the single most useful trap in the whole article to internalise. A 13F reports what a fund holds at quarter end, not what it did during the quarter. When a big private company lists, everyone who already owned it suddenly appears as a "new buyer" — an artefact of disclosure rules, not a stampede.

Netflix is the streaming business that graduated from growth-at-all-costs to profitability, ads and password-sharing crackdowns.

It is the headline of Bill Ackman's six new positions — the biggest overhaul of Pershing Square's portfolio in years — and notable for a specific human reason: Ackman exited Netflix in 2022 at a roughly $400 million loss, and has now bought it back publicly.

Re-entering a stock that cost you a very public $400 million requires overcoming the discomfort of admitting the exit was wrong, so it is a louder signal of changed conviction than an ordinary new position. It says nothing about whether he is right this time.

Microsoft is Azure, Office, Windows and the deepest commercial relationship with OpenAI — for years an automatic core holding for almost any large fund.

In Q2 it nearly vanished from this list: "Microsoft, once a fixture on this list, appeared only once at the end of June after falling more than 20% YTD." One appearance out of twenty funds' top-five holdings.

Two readings, and the article doesn't choose between them. Either institutions rotated the money into Amazon, TSMC and Alphabet on a view about relative AI positioning — or a 20% drawdown mechanically pushed Microsoft out of "top five by value" lists without anyone selling a share. Both are consistent with the data, which is exactly why a top-five-holdings table is a weak instrument for detecting a change of heart.

Apple sells iPhones, Macs and an enormous services business, and has been conspicuously slower than its peers to show a convincing AI product story.

The observation here is a blank space: "Apple and Tesla were entirely absent from the top-five holdings" of all 20 funds. Not trimmed — absent.

Read it carefully, though. These 20 funds are a hand-picked sample of aggressive, alpha-seeking managers; Apple remains Berkshire's largest stock holding, mentioned in this very article. Absence from a momentum-tilted sample means Apple is not where those managers are hunting for return, not that it has been abandoned.

Tesla is the electric-vehicle maker whose valuation increasingly rests on autonomy, robotaxis and robotics rather than on selling cars.

Like Apple, it was entirely absent from every fund's top-five holdings — and, unlike Apple, it appears nowhere else in the article either: not among top buys, not in the curveball section, not in any fund's notable moves.

In a quarter where money spread outward into memory, storage, connectivity, power, materials, rail and even offices in search of AI exposure, a total absence from a group of twenty AI-hungry funds is a conspicuous piece of negative space.

Berkshire appears in this article as one of the filers rather than as a stock to buy — but its filing was the quarter's most-discussed. It became a net buyer of stocks for the first time in 14 quarters: Alphabet increased 83% to its third-largest holding, adds to Delta and Lennar, a small new D.R. Horton stake, and Constellation Brands sold out entirely.

Three and a half years of net selling ending is a policy change, not a trade. It says the people running the largest cash pile in corporate America finally found things worth owning — concentrated in one mega-cap platform and in housing.

The neutral framing carries over from App Economy's own coverage of the Q2 report three days earlier: the money moving replaces the question "when will they deploy?" with the harder one — at what return, under a new chief executive.

GURU is an exchange-traded fund built to do automatically what readers of a 13F article are tempted to do manually: read the filings and buy the hedge funds' biggest positions.

Its record is the article's central piece of evidence: it "has underperformed the S&P 500 since its inception in 2012" — and, as App Economy notes, that comparison flatters the strategy, because it excludes the "2 and 20" fees (2% of assets plus 20% of gains) that the underlying funds actually charge their own investors.

The reason it fails is built into the data. Filings arrive 45 days late, cover only US long equity, and exclude the shorts and hedges that shape a fund's real risk. You end up copying a stale, partial photograph of someone else's portfolio without the conviction, the sizing rules or the exit discipline that made it work — which is why the article's conclusion is to use 13Fs "as a starting point," never as a portfolio.