Best AI Stocks for December 2025: Beyond the Obvious Mega-Cap Plays

Everyone’s buying NVIDIA and Microsoft. You know what they’re not buying? The smaller AI companies actually building the infrastructure that makes all this shit work.

I’m not talking about penny stocks or speculative garbage. I’m talking about profitable companies with real revenue, real customers, and real moats that happen to trade under $50 billion market cap. You know, the kind of stocks that can actually 10x instead of just keeping up with the S&P.

December 2025 Update: I’m refreshing this list because Q3 earnings just separated the real AI companies from the pretenders. While September was all about “AI exposure,” the current data shows which companies are actually monetizing AI versus just mentioning it in press releases. I’ve adjusted my conviction scores to reflect who’s executing and who’s full of shit.

The mega-caps are great if you want stable 15-20% annual returns. But if you want asymmetric upside, you need to look at companies where AI revenue is becoming material to their overall business, not just a rounding error on a $3 trillion balance sheet.

Here’s what actually matters when you’re evaluating smaller AI plays:

Revenue growth that’s accelerating, not decelerating. Any idiot can show 50% growth off a tiny base. What matters is whether that growth rate is speeding up as the company scales. That’s the signal that product-market fit is real.

Customer retention metrics. In B2B AI, net dollar retention above 120% means customers are not just staying, they’re spending more every quarter. Below 100% means you’re on a treadmill replacing churned revenue.

Gross margins above 70%. Software companies with AI infrastructure should be printing money at scale. If margins are compressing as they grow, something’s broken in the business model.

Path to profitability that’s visible. I don’t need them to be profitable today, but I need to see the math on how they get there without raising another $500 million.

The smaller AI stocks that matter aren’t trying to compete with NVIDIA on chips or Microsoft on cloud infrastructure. They’re solving specific problems in ways that create genuine switching costs. That’s where the opportunity is.

Before finalizing my December picks, I ran these tickers through the Chaikin Power Gauge to make sure my fundamental thesis wasn’t being contradicted by the technical “Big Money” flow. When institutional money is rotating out while you’re rotating in, that’s usually a bad sign.

The Infrastructure Layer (Not Named NVIDIA)

Arista Networks (ANET)

Conviction Score: 9/10

Everyone’s obsessing over GPUs while ignoring the networking gear that actually connects all those data centers. Arista makes the switches and routers that handle the insane bandwidth requirements of AI training clusters.

They reported 20% revenue growth in Q3 with operating margins at 44%. Not some speculative AI startup burning cash, a profitable networking company that’s become essential AI infrastructure.

Cloud titans are their customers. When Microsoft, Meta, and Oracle are building out AI data centers, they’re buying Arista gear because it’s the only thing that can handle 400G and 800G ethernet speeds at scale. The switching costs here are massive because you can’t just rip out your entire network architecture.

What changed since September: Stock’s up 28% since my last update, but the valuation still makes sense given the growth trajectory. AI-driven networking revenue is now 55% of their total business, up from 35% last year.

The risk I’m watching: They trade at 45X forward earnings, which isn’t cheap. Any sign that hyperscaler capex is slowing would crush this stock. But right now, capex is accelerating, not slowing.

Snowflake (SNOW)

Conviction Score: 5/10

I’ve been skeptical of Snowflake’s valuation for years, and I’m still skeptical. But something changed in Q3 that’s worth paying attention to.

Their AI/ML workloads grew 120% year-over-year and now represent 30% of total consumption. That’s not a rounding error anymore. Companies are using Snowflake to store and process the training data for AI models, and once that data is in Snowflake’s platform, moving it is expensive and painful.

Product revenue hit $900 million with 29% growth. Not the 50%+ growth from the hype days, but stabilizing around 30% while improving unit economics is actually a good sign in this market.

What changed since September: They announced partnerships with NVIDIA and AWS that make it easier to run AI workloads directly on Snowflake without moving data around. That reduces friction, which should accelerate adoption.

The risk I’m watching: Still trading at 11X sales with decelerating growth. If that 29% drops to 20%, this gets cut in half. This is a small position for me because the valuation leaves zero room for error.

Datadog (DDOG)

Conviction Score: 8/10

Observability and monitoring for cloud infrastructure. Not sexy, but absolutely critical when you’re running AI models in production.

Every company deploying AI needs to monitor performance, track costs, and debug issues in real-time. Datadog’s platform does all of that, and their AI-native monitoring tools are becoming essential infrastructure.

Q3 revenue hit $690 million with 26% growth and 80%+ gross margins. They’re profitable with $200 million in free cash flow. This is what a mature SaaS business with AI tailwinds looks like.

What changed since September: They launched AI-specific monitoring products that track GPU utilization, model inference costs, and performance metrics that didn’t exist two years ago. Early adoption is strong, and it’s incremental revenue on top of their core business.

The risk I’m watching: Competition from the hyperscalers. AWS, Azure, and GCP all have native monitoring tools. Datadog’s advantage is being platform-agnostic, but if customers consolidate on one cloud provider, that advantage weakens.

The Application Layer (Where the Real Money Gets Made)

Palantir (PLTR)

Conviction Score: 7/10

Yeah, I know this is becoming a mega-cap. Market cap just crossed $150 billion. But it’s still small enough to have explosive growth ahead.

I’ve been wrong about Palantir’s valuation for two years. Kept thinking “this can’t sustain 300X earnings multiples” and kept watching it climb. Q3 finally made me stop fighting it.

U.S. commercial revenue grew 54% to $179 million. Government revenue grew 40% to $320 million. Their AI Platform (AIP) is now driving customer acquisition in ways that their legacy products never did.

What Palantir figured out is that enterprises don’t want to build AI infrastructure from scratch. They want someone to show up, integrate with their existing data, and make AI actually work in their specific workflows. That’s what AIP does.

What changed since September: Commercial customer count grew 51% year-over-year to 321 customers. They’re not just growing with existing customers, they’re winning new logos at an accelerating rate. That’s rare in enterprise software.

The risk I’m watching: Valuation is absolutely insane at 175X forward earnings. Any stumble in growth and this drops 40% overnight. But sometimes you pay up for companies that are legitimately changing how enterprises adopt AI.

CrowdStrike (CRWD)

Conviction Score: 6/10

Cybersecurity company that’s using AI to detect threats in real-time. The more AI gets deployed across enterprises, the bigger the attack surface becomes. CrowdStrike is positioning itself as the security layer for AI adoption.

They reported $786 million in Q3 revenue with 29% growth and $223 million in free cash flow. Profitable, growing, and operating in a market where spending never goes down.

Their Falcon platform now includes AI-powered threat hunting that identifies attacks traditional signature-based security misses. Net new annual recurring revenue was $223 million in Q3, showing strong demand.

What changed since September: Nothing dramatic. This is a steady execution story. But that’s exactly why it’s in the portfolio. Sometimes boring consistency beats wild swings.

The risk I’m watching: Remember the July outage that took down airlines and hospitals? That was CrowdStrike. They recovered quickly, but enterprise customers have long memories. If there’s another major incident, this gets punished severely.

UiPath (PATH)

Conviction Score: 4/10

Robotic process automation (RPA) company that’s pivoting hard into AI agents. The original RPA business was solid but commoditizing. Their bet is that AI agents will be the next evolution of workflow automation.

Q3 revenue was $354 million with 10% growth. That’s the problem. Growth has slowed significantly from the 30%+ rates they were posting two years ago. But ARR is stable at $1.57 billion, and they’re guiding for improvement in 2026.

The thesis here is simple: if AI agents become the dominant way enterprises automate work, UiPath’s existing customer base and workflow knowledge gives them a massive head start. But that’s a big if.

What changed since September: They announced an AI agent platform that competes directly with Microsoft’s Copilot Studio. Early customer feedback is positive, but adoption is slow because enterprises are cautious about agentic AI.

The risk I’m watching: This is a turnaround play betting on a product pivot. If the AI agent platform doesn’t gain traction by mid-2026, the stock probably drifts lower. This is my smallest AI position.

The Wild Cards (High Risk, High Reward)

SoundHound AI (SOUN)

Conviction Score: 5/10

Voice AI company that powers conversational interfaces for automotive, restaurants, and customer service. Think of them as the picks-and-shovels play for voice AI adoption.

Market cap under $6 billion, which is tiny compared to the others on this list. Q3 revenue was $25 million with 89% growth, and they just raised guidance for full-year 2025 to $85-87 million.

The interesting part is their customer base: Stellantis, Hyundai, and major restaurant chains are using SoundHound to handle voice ordering and customer interactions. This isn’t experimental, it’s in production at scale.

What changed since September: They acquired Amelia AI for $80 million, adding enterprise conversational AI capabilities. This expands their addressable market beyond automotive into customer service and contact centers.

The risk I’m watching: They’re still burning cash and need to raise more money in 2026. If the market turns risk-off, small-cap AI companies with negative cash flow get destroyed first. Don’t bet more than you can afford to lose.

C3.ai (AI)

Conviction Score: 3/10

Enterprise AI software company that’s been struggling to prove its business model works. Revenue growth has been inconsistent, customer wins have been lumpy, and the market is skeptical.

But Q2 (their fiscal quarters are weird) showed 29% revenue growth to $94 million with subscription revenue growing 30%. Pilot programs are converting to production deployments at higher rates than they were a year ago.

Their partnership with Google Cloud is starting to gain traction. Joint pipeline is now $1.8 billion, up 150% from last year. Whether that translates to actual bookings is the question.

What changed since September: They reported their first operating profit in company history. Small ($4 million), but directionally important. Proves the unit economics can work if they execute.

The risk I’m watching: This has been a show-me story for years. Lots of promises, inconsistent delivery. I’ve got a tiny position because the optionality is interesting, but I need to see at least two more quarters of consistent execution before I add more.

The One Stock I’m Avoiding: BigBear.ai (BBAI)

Government-focused AI company that reports lumpy revenue and has constant cash flow issues. They keep winning contracts that sound impressive in press releases but don’t translate to consistent financial performance.

Q3 revenue was $41 million, down 21% year-over-year. They blame “contract timing” but that’s been the excuse for eight quarters straight. At some point, contract timing becomes a business model problem.

The government AI market is real, but Palantir is eating everyone’s lunch. I don’t see how BigBear competes when customers can just go with Palantir’s proven platform instead of taking a risk on a struggling vendor.

What Actually Matters for Small AI Stocks

The metrics I watch aren’t the same as mega-caps. With NVIDIA or Microsoft, you care about total addressable market and competitive moat. With smaller AI plays, you care about:

Rule of 40: Revenue growth rate plus profit margin should exceed 40%. If you’re growing 30% and have 15% margins, you hit 45%. That’s healthy. Below 40% means you’re not efficient enough or not growing fast enough.

Magic Number: Sales efficiency metric calculated as net new ARR divided by sales and marketing spend. Above 0.75 is good, above 1.0 is excellent. Below 0.5 means you’re spending too much to acquire revenue.

Net Dollar Retention: For B2B SaaS, this measures how much revenue you keep from existing customers. 120%+ means they’re spending 20% more each year. Below 100% means you’re shrinking with existing customers, which is a death spiral.

Burn Multiple: For unprofitable companies, how much cash are you burning to generate each dollar of new ARR? Under 1.5X is acceptable, over 3X is concerning. Shows capital efficiency.

These metrics separate real businesses from hype stocks. NVIDIA doesn’t need to pass these tests because they’re printing $30 billion in quarterly profit. But smaller AI companies absolutely do.

My Actual Position Sizing in December

I’m at about 12% of my portfolio in smaller AI stocks (under $200B market cap), up from 8% in September. Split like this:

• 30% ANET (was 20%) – increased because hyperscaler capex keeps accelerating
• 25% DDOG (was 25%) – unchanged, steady execution story
• 20% PLTR (was 15%) – increased after Q3 proved commercial growth is real
• 10% CRWD (was 15%) – trimmed slightly after the July outage
• 8% SNOW (was 10%) – trimmed because valuation still concerns me
• 5% SOUN (was 5%) – lottery ticket, staying small
• 2% PATH and AI combined (was 10%) – cut these significantly, show-me stories

This isn’t “diversification for diversification’s sake.” Each position represents a specific thesis about how AI infrastructure and applications mature over the next 2-3 years.

The mega-caps (NVIDIA, Microsoft, Google) are in a different bucket. Those are 28% of my portfolio and covered in my tech stocks article. The smaller AI plays get their own allocation because the risk/reward profile is completely different.

The Reality Check

Will all of these stocks work out? Absolutely not. PATH and AI could easily go to zero if their turnarounds fail. SOUN could run out of cash before reaching profitability. Even the higher-conviction plays like ANET and DDOG could get cut in half if we hit a recession and enterprise spending freezes.

But that’s why position sizing matters. I’m not betting the farm on any single name. I’m taking calculated risks on companies where the upside is 3-5X if they execute, and the downside is losing my investment if they don’t.

The key is that these companies are solving real problems with real revenue from real customers. They’re not science projects hoping to monetize in 2028. They’re businesses that exist today, with AI tailwinds that could accelerate growth for years.

The market’s paying 60X earnings for companies growing 30% annually. That pricing only works if the 30% growth continues for years. As soon as it doesn’t, multiples compress violently.

That’s why the metrics matter. That’s why customer retention matters. That’s why I’m watching burn rates and sales efficiency instead of just growth rates.

Small AI stocks can make you rich or destroy your portfolio. Usually both, depending on when you buy and whether you have the discipline to hold through volatility.

Standard disclaimers: Not a financial advisor, this isn’t advice, you could lose money, do your own research, consult professionals, etc. Small-cap AI stocks are volatile and many will fail. But sometimes calculated risks on real businesses create real wealth.