ARK Invest Doubles Down on AI Infrastructure with CoreWeave and OpenAI Stake

ARK Invest added major new positions in AI and adjacent sectors, buying 83,764 shares of CoreWeave worth about $6.9 million and taking a first direct stake in OpenAI via 348,995 shares/units across ARKF, ARKK, and ARKW. The firm also added Kodiak AI, Oklo, DoorDash, Arcturus Therapeutics, and GeneDx while trimming Strata Critical Medical by 745,000 shares and reducing Veracyte, Teradyne, Pinterest, Discovery, and LY Corp. The moves reinforce Cathie Wood’s bullish AI thesis, though ARKK remains down about 12% year to date and has seen $1.2 billion in net outflows over the past 12 months.

Analysis

This reads less like a stock-picking update and more like a deliberate factor rotation into the “picks-and-shovels” layer of the AI capex cycle. The important second-order signal is that capital is moving from visible public AI beneficiaries into scarce private-market access, which can tighten float, raise implied scarcity premia, and pull public comps higher even when near-term operating guides disappoint. CoreWeave is the clearest transmission mechanism: if hyperscalers keep outsourcing incremental GPU capacity, the public market may start pricing infrastructure owners more like strategic toll roads than cyclical cloud vendors.

The bigger setup is that AI is now colliding with energy and compute constraints, and that makes OKLO and CRWV complementary expressions of the same thesis. If inference costs truly compress as fast as Wood argues, the bottleneck shifts from model training to power, grid interconnects, and datacenter reliability; that is a tailwind for nuclear and a headwind for smaller, less-capitalized infrastructure providers that cannot finance capacity fast enough. On the flip side, the willingness to buy into a crowded AI trade after a run-up raises near-term drawdown risk if the market starts demanding proof of monetization rather than narrative.

The contrarian miss is that private participation in OpenAI may be signaling late-cycle enthusiasm rather than an early-edge advantage. When a strategy becomes broad enough to span public infra, private AI, autonomy, and nuclear, factor crowding increases and correlation risk rises; the portfolio can look diversified while actually being one bet on a single macro regime shift. If rates back up or AI spending pauses for even one quarter, the high-duration names in this basket should re-rate first.

Near term, the most tradable expression is relative rather than outright long. The trims in diagnostics, testing equipment, and consumer internet suggest capital is being pulled from lower-beta, slower-growth cash generators into a narrower set of “AI optionality” names, but that pivot may be premature if positioning unwinds. Expect the next catalyst to be either another hyperscaler capex print or a funding-market shock in private AI; both would likely separate true infrastructure beneficiaries from story stocks within weeks, not years.

Key Decisions for Investors

• Long CRWV / short TER on a 4-8 week horizon: expresses the AI infrastructure vs. semicap test-spend divergence. Target 15-20% relative upside if capex remains strong; stop if CoreWeave loses momentum on any financing or demand reset.
• Buy OKLO on pullbacks, but size as a venture-style equity proxy, not a utility: 3-6 month horizon with asymmetric upside if power scarcity becomes the binding AI constraint. Use tight risk controls because the name will trade with sentiment around capital intensity.
• Fade crowded AI beta with a tactical short in ARKK or via ARKK puts into strength over the next 1-2 months. The risk/reward is attractive if the market starts penalizing unprofitable duration and private-market enthusiasm cools.
• Pair long GOOGL/MSFT against a basket of AI infrastructure pure-plays for 1-3 months if hyperscaler capex disappoints: the platforms have balance-sheet resilience and are better able to absorb a pause in GPU spend.
• Avoid chasing CRWV outright after the recent move; prefer waiting for a post-event retrace or using call spreads. The trade works best if the market re-prices the stock as scarce infrastructure, but financing and guidance risk can cut both ways quickly.