Ask any portfolio manager what separates a decade-winning stock from a flash-in-the-pan, and the answers usually circle back to two things: durable demand and an ecosystem that makes switching costly. That rule explains why the market is handing serious attention to companies as different as GPU farms, compact nuclear reactors, voice-AI platforms and rare-earth miners. They sit at intersections—AI compute, power supply, mobility and critical materials—where growth can compound if the pieces fall into place.
Three clusters worth watching
AI infrastructure
If the last few years taught investors anything, it's that AI isn’t just a software story—it’s a hardware and power story, too. Companies that sell the compute and the data-center plumbing (and those that promise cheaper, faster access to GPUs) are suddenly in the growth conversation.
CoreWeave, which pivoted from crypto mining to become a dedicated GPU cloud, has expanded rapidly; analysts forecast revenue and adjusted EBITDA to grow at triple‑digit CAGRs through 2027 as customers flock to GPU-optimized clouds. Nebius positions itself as a "full stack" AI infrastructure provider, and some forecasts expect stratospheric revenue growth as it signs multibillion-dollar contracts with major cloud customers. SoundHound sits a layer above—voice and conversational AI that enterprises can bolt into products without handing data to hyperscalers.
These are complementary plays: GPUs and data-center capacity (CoreWeave, Nebius), plus the AI applications that consume that capacity (SoundHound). For investors, that means exposure to the sector can be tuned toward hardware, services or software depending on whether you prefer capital-intensive scale or higher-margin recurring revenue. The broader trend also ties into moonshot ideas such as putting data centers in unusual locales—everything from remote colocation hubs to more exotic concepts—illustrating how infrastructure is being reimagined to serve AI's appetite: see Google's longer-term experiments around novel data-center concepts in space and other frontier locations (Project Suncatcher).
Power and the new grid
Not all infrastructure is silicon. Oklo, a compact nuclear start-up, is pitching small, fast-build reactors that could sit adjacent to power-hungry sites like AI data centers. The company’s commercial case rests on two tailwinds: surging electricity demand from AI and a push for cleaner baseload power. It’s still pre-revenue and navigating heavy regulatory hurdles, but if regulators and offtakers align, the payoff could be outsized for patient investors.
There’s a neat narrative overlap here: GPUs need power. More GPUs demand more reliable, low-carbon electricity—or new ways to site and finance power close to compute. Oklo is betting on that proximity.
Mobility and materials
Two other long-horizon themes deserve a seat at the table. Joby Aviation is trying to commercialize eVTOL air taxis—ambitious, capital-intensive, and dependent on regulatory sign-offs and urban infrastructure build-outs. It’s the kind of speculative growth story that could remake short-haul travel if safety, economics and vertiport networks all line up.
On the materials side, MP Materials owns and operates one of the few scalable rare-earth sites in the U.S. High-performance magnets—used in EV motors, defense systems and countless consumer devices—are a strategic bottleneck. Similarly, more speculative miners like The Metals Company (focused on deep-sea nodules) could offer parabolic upside if permits and commercialization milestones clear, albeit with higher environmental and execution risk.
How to think about valuations and timing
Some of these names trade at lofty multiples that bake in rapid execution. CoreWeave and Nebius are priced for massive expansion; SoundHound is betting on a future where voice interfaces become ubiquitous across cars, restaurants and financial services. Oklo and Joby are earlier-stage and hinge on regulatory approvals and capital deployment. MP Materials already generates revenue but must scale production facilities to meet demand.
Three practical filters to apply before picking a position:
- Time horizon: These are decade bets. Expect volatility and be ready to hold through execution cycles. Short-term traders will be tested.
- Cash runway and capital intensity: Nuclear reactors and eVTOL fleets need sustained capital; GPU clouds require rapid scaling of physical infrastructure. Watch balance sheets and financing conditions closely.
- Customer and contract anchors: Suppliers that already have multiyear contracts with hyperscalers or major OEMs reduce execution risk. That matters for infrastructure companies in particular.
A few things investors miss at first glance
1) AI growth isn’t uniform. Demand for specialized GPU cycles (training vs. inference) can move differently; companies that optimize for one may miss the boat on the other.
2) Power and location economics are strategic assets. Firms that secure large, cheap power allocations or control unique colocation footprints can extract outsized margins. This ties back to broader innovations in data-center siting and the rising conversation about where compute should live to serve real‑time AI—an angle that shows up in companies experimenting with novel deployment models and in advances in AI tooling that link models to user data (Gemini’s deeper search integrations).
3) Geopolitics and supply chains matter for rare-earth and materials plays. Diversifying away from single-country dependencies is a national priority and a growth opportunity for domestic producers.
A note on risk and portfolio construction
Don’t treat these names as a single theme. They are components of an emerging ecosystem: compute, power, software layers and materials. A balanced way to express conviction might combine exposure across the stack—say, a GPU/cloud play, one software layer, and one materials or energy hedge—rather than go all-in on a single company.
If you’re interested in the mechanics—how GPUs are provisioned, where data centers get sited, and why power contracts can make or break margins—follow developments across infrastructure and regulatory news as closely as you would earnings reports. This is an era where technical details (chip generations, watt-per-flop economics, reactor licensing timelines) move markets as much as user growth numbers.
Markets love stories with a path to scale. These companies each tell one, from Oklo’s reactors to CoreWeave’s GPU farms to MP Materials’ magnets. The hard part is judging which paths are most likely to clear the regulatory, financing and execution hurdles ahead. If you can live with uncertainty, and you frame positions as long-term experiments rather than guaranteed winners, there’s a rich set of opportunities to explore.