Three Long-Term Stocks Analysts Still Like After Volatility

Have you noticed how every week the market finds a new reason to sound exhausted? Oil spikes, rates shift, geopolitics flare up, and then someone on a screen starts arguing about whether artificial intelligence is a miracle or a liability. I get it. The noise is loud. Still, I keep coming back to a simpler question: which businesses can compound through that mess rather than live or die by next quarter’s mood?

Why Patient Investors Keep Circling The Same Few Names

Short-term tapes punish hesitation. Long-term ownership rewards a different skill set. You need businesses that can sell more of something the world actually wants, at better margins, without needing a perfect macro script. That is not a slogan. It is the filter a handful of well-ranked analysts still apply when they put real price targets on paper.

Three names keep showing up in that kind of work: an enterprise software and cloud giant racing to add power-hungry capacity, a space company trying to become a full-stack industrial platform, and a consumer-tech titan rebuilding its model stack while already sitting on a distribution machine most startups would sell a kidney for. I am not saying you should mortgage the house. I am saying the research notes are unusually aligned on duration.

In my experience, the useful part of analyst research is rarely the exact target. It is the operating story underneath. Capacity timing. Backlog quality. Customer concentration. Gross-margin shape. Those details survive a bad week. Headlines do not.

Oracle And The Decade-Long Cloud Bet

Oracle just posted a first quarter that beat expectations, and the cleanest driver was cloud infrastructure. That is the part of the company that used to feel like a late arrival to a party Amazon and Microsoft already owned. The late arrival is now booking work at a pace that forces people to rewrite old assumptions.

After a run of investor meetings with the company’s investor-relations lead, one high-ranked analyst kept a buy rating and a $400 price target. He still calls the name a decade stock. That phrase is doing a lot of work. It means the thesis is not “the next two prints look fine.” It means training and inference demand, the traditional public cloud book, and AI-enabled databases and applications can stack for years.

We continue to believe this is a decade stock, driven by the massive and profitable opportunity ahead for AI training and inferencing.

That is the kind of sentence you either dismiss as marketing or treat as a map. I lean toward the map, with conditions. Capacity coming online is not the same thing as revenue recognized in a given quarter. Management delivered about 850 MW of data-center capacity in the quarter, which sounds like a trophy number until you hear the caveat. Infrastructure-as-a-service revenue depends more on when that capacity actually goes live for customers than on the headline megawatts. Timing still matters. It always will.

Even so, the same analyst expects IaaS growth to stay fierce, with full-year growth running above the quarter’s already striking 120% constant-currency clip. Management also guided toward sequential acceleration in total revenue through the fiscal year as the faster infrastructure slice becomes a larger share of the mix. If that mix shift happens, the old software story starts looking like a platform story with better operating leverage.

Customer Concentration Is The Question Everyone Asks

Let’s be honest. When one famous AI lab shows up in the backlog conversation, investors get twitchy. Concentration risk is not a vibes issue. It is a cash-flow issue. The useful update is that OpenAI’s share of remaining performance obligations has fallen from about 80% last June to roughly 50% now, and the company expects that share to keep sliding as new contracts land with both new and existing customers.

Is 50% still high? Yes. Is the direction healthier? Also yes. I would rather own a book that is broadening than a book that is secretly one handshake. Diversification of RPO is not glamorous. It is how a “decade stock” stops being a single-customer story wearing a cloud costume.

• Cloud infrastructure is doing the heavy lifting in the growth mix.
• Megawatts installed are not a perfect proxy for quarterly IaaS revenue.
• Management is signaling sequential acceleration as infrastructure becomes a bigger slice.
• Customer concentration is improving, even if it is not gone.

Perhaps the most interesting aspect is how ordinary the rest of the software franchise starts to look if infrastructure keeps compounding. Databases and applications that ride the same cloud rails can turn AI from a demo into a billing line. That is the unglamorous path to profit acceleration, and it is the path the bull case actually needs.

Rocket Lab Wants To Be More Than A Launch Vendor

Space stocks have a habit of sounding like science fiction until the backlog shows up. Rocket Lab is trying to assemble launch, spacecraft, components, payloads, and optical communications into one industrial stack. Neutron and a major constellation-related transaction are supposed to complete the medium-lift and applications-spectrum picture around 2027. That is a lot of moving parts. It is also the only way this stops being a “cool rockets” ticker.

A five-star analyst recently initiated coverage with a buy rating and an $80 target. The argument is blunt. With the Iridium deal completed, Rocket Lab would become the only vertically integrated space company with a path to positive free cash flow that does not feel like a prayer. The acquired constellation is described as a deployed asset valued at more than $3 billion, plus globally coordinated L-band spectrum, about $500 million of EBITDA, and roughly $300 million of free cash flow.

Those are not hobby numbers. They are the difference between a company that spends to exist and a company that can fund the next vehicle without constantly tapping equity markets. I’ve found that investors underestimate how much narrative risk disappears once FCF turns durable. Space still looks speculative on a slide. Cash generation makes it look like an industrial.

Backlog, Margins, And The End Of The Build Years

Revenue growth is projected around 59% in 2026 on the analyst’s numbers, with a backlog that has more than doubled to almost $2.4 billion from $1.1 billion in the third quarter of 2025. That doubling is the kind of detail I actually circle. Backlog is not revenue. It is a queue. A thicker queue, if execution holds, is how you get from investment mode to monetization mode.

Gross margins are modeled to trough near 35% in 2027 and expand toward about 47% by 2030. Research and development, currently a heavy tax on the model, is expected to fall from around 35% of sales in 2026 toward about 10% in 2028. That is operating leverage in slow motion. Ugly first. Then, if the platform works, much less ugly.

Will Neutron slip? Possibly. Will integration of a constellation business be messier than a slide deck? Almost certainly. That is the honest risk. The counterweight is vertical integration. If you design the vehicle, fly the mission, build the spacecraft, and own applications plus spectrum, you stop living on launch-day lottery tickets. You start selling a system.

The company is moving from a heavy investment phase into monetization, and that shift is the whole point of owning it for more than a news cycle.

Meta Is Trying To Turn Models Into A Second Engine

Meta has had a messy year in the tape even while remaining one of the most cash-generative consumer platforms on earth. That combination creates a strange emotional market. People know the ads machine. They are less sure about the next act. One well-followed analyst recently moved the stock from hold to buy and lifted the target to $820 from $640. That is not a timid revision.

The upgrade rests on a simple claim. Meta is early in launching frontier models and AI products that live outside the advertising box, especially an agent effort and API access to its model family. Frontier models, in this view, sit at the center of product and monetization for years. Superintelligence language gets thrown around in these notes. I treat that word carefully. The practical version is more grounded: better models, better products, better ways to charge for intelligence on top of a two-sided network that already reaches billions of people.

Last summer the company set a goal to rebuild its superintelligence lab effort and ship frontier-class models within a year. The same analyst argues the team has largely hit that clock, moving from an early Muse Spark release in July to a later revision that looks competitive with the best-known closed models. Competitive does not mean permanently ahead. It means Meta is no longer a spectator in a race it cannot afford to watch from the stands.

Distribution Still Matters More Than Demo Videos

Here is where I get opinionated. Model quality is necessary. Distribution is decisive. A product that can sit inside apps already used by something like 4 billion people does not need to win every benchmark screenshot. It needs to be good enough, useful enough, and close enough to the daily habit. That is a different contest from a research lab posting leaderboard scores.

A consumer-facing layer called Watermelon, in the analyst’s framing, could open lanes in consumer products, business intelligence, and engagement across the family of apps. Internal operations may get a lift too. Efficiency is not a headline people click. It is how a company funds the next wave of training runs without looking reckless.

1. Ship models that are close enough to the frontier to be taken seriously.
2. Put those models inside products people already open every day.
3. Sell access, agents, and tools without starving the ads engine.
4. Keep building infrastructure so compute is not the bottleneck later.

Infrastructure is the quiet half of the story. If you believe the product roadmap, you also have to believe the company will keep standing up capacity for its own models and, eventually, for others. Compute is not a side quest. It is the factory floor.

What These Three Ideas Actually Have In Common

On the surface they look unrelated. One sells enterprise cloud. One launches rockets and wants to own the stack around them. One runs the social graph and is trying to wrap intelligence around it. Dig one layer down and the pattern is almost boring. Each company is spending now so a scarcer resource can be rented later: power and racks, launch and spectrum, models and distribution.

That is why the notes talk in multi-year language. Oracle’s mix is tilting toward infrastructure. Rocket Lab’s backlog and a constellation deal are meant to pull the company out of perpetual investment mode. Meta is treating models as a product pipeline, not a press-cycle. None of those plots resolve in a month.

A simple ownership filter I keep using:
  Can the company sell more of a scarce input?
  Does the mix get better as that input scales?
  Is concentration risk falling, not rising?
  Will cash generation catch the story before fatigue does?

If you cannot answer those four with at least a shrug toward yes, the target price is decoration. If you can, volatility becomes a scheduling problem rather than a verdict.

How To Think About Risk Without Turning Into A Pessimist

Oracle can miss a quarter because capacity comes online later than the slide implied. That does not automatically kill the decade thesis. It does mean you should watch IaaS timing like a hawk. Rocket Lab can slip on Neutron or spend longer digesting a constellation business than anyone wants. Space execution has a long history of teaching humility. Meta can spend enormous sums on models that users treat as a novelty. Advertising can also wobble if the consumer tightens.

Those are real risks. They are not reasons to pretend the opportunity set is empty. I would rather own a company with a messy but expanding demand curve than a tidy company whose demand curve is already fully priced and fully mature. That bias will not fit every portfolio. It fits a long-term sleeve.

Position size is the adult part of this conversation. A space name with 2027 and 2028 milestones is not the same animal as a cash-machine platform with a model-upgrade story. Treat them that way. Concentration in a single theme, even a fashionable one like AI, is how a good idea becomes an uncomfortable night.

A Practical Way To Follow The Stories From Here

You do not need a dozen dashboards. You need a short list of tells. For Oracle, watch sequential revenue acceleration, IaaS growth versus installed megawatts, and the slide in any single customer’s share of remaining performance obligations. For Rocket Lab, watch backlog conversion, margin trough timing, and whether FCF commentary starts sounding scheduled rather than aspirational. For Meta, watch whether new AI surfaces show up in engagement and whether management talks about monetization in complete sentences instead of slogans.

• Oracle: mix shift toward infrastructure and broader RPO.
• Rocket Lab: backlog quality plus the path to positive cash flow.
• Meta: model cadence plus distribution into existing apps.

That is enough to stay honest. If those tells break, the thesis is breaking. If they hold while the tape panics about next week’s headline, you at least know what you own.

The Unfashionable Case For Looking Past The Noise

Markets love a villain. This month it might be rates. Next month it might be energy. After that, someone will decide AI spending is either too little or too much. Meanwhile, companies still sign multi-year cloud contracts, still book launch and spacecraft work, still put software in front of billions of people. The calendar does not pause because the comment section is tired.

I am not asking anyone to feel brave. Bravery is overrated in investing. Patience with a checklist is underrated. The three names in this piece are not magic. They are simply businesses where serious analysts still see a multi-year slope instead of a one-quarter bounce. That slope can flatten. Until it does, it is worth more attention than the latest reason to stay glued to the blinking numbers.

So here is the uncomfortable close. If you only buy when the tape feels calm, you will keep waiting for a mood that rarely lasts. If you buy only stories with no visible risk, you will own very little that can still surprise to the upside. The better habit is smaller, slower, and more specific. Know the capacity timing. Know the backlog. Know whether the second engine is a product or a press release. Then let time do the part screens cannot.

That will not make the next selloff pleasant. It might make the next few years make sense. And in a market this noisy, sense is already a scarce asset.