Are AI Stocks Still Worth Buying for the Long Term?

In January 2023, you could have bought NVIDIA shares for around $15 (split-adjusted). By early 2025, they topped $140. That is a nearly 10x return in just two years — the kind of gain most investors wait a lifetime to see. But here is the question that keeps portfolio managers up at night in 2026: was that the beginning of something much bigger, or the kind of parabolic run that ends with investors holding the bag?

The artificial intelligence trade has been the defining investment theme of the mid-2020s. Hundreds of billions of dollars have poured into AI infrastructure, AI software, and AI-adjacent companies. The “Magnificent 7” mega-cap tech stocks have driven an outsized share of market returns, and the word “AI” has become a magic incantation on earnings calls — mention it enough times, and your stock goes up.

But seasoned investors have seen this movie before. The internet was going to change everything in 1999, too. And it did — just not before the Nasdaq lost 78% of its value and companies like Pets.com and Webvan became cautionary tales. So where does AI stand today? Is this the internet circa 1996 — still early, with the real gains ahead — or is it 1999, with euphoria masking unsustainable valuations?

This article dives deep into that question. We will examine current AI stock valuations against historical tech bubbles, lay out the strongest arguments on both sides, identify which companies have real AI revenue versus hype, and build a framework for constructing a diversified AI portfolio that can weather whatever comes next.

The AI Gold Rush — Where Are We Now?

To understand where AI stocks are headed, we first need an honest assessment of where they stand right now. The numbers are staggering by any historical measure.

Global spending on AI infrastructure — data centers, GPUs, networking equipment, and cooling systems — surpassed $300 billion in 2025 and is projected to exceed $400 billion in 2026. The hyperscalers (Microsoft, Google, Amazon, Meta) alone have committed over $250 billion in combined capital expenditure for AI-related infrastructure. NVIDIA’s data center revenue went from $15 billion in fiscal year 2023 to over $115 billion in fiscal year 2025. That is not a typo — it is a nearly 8x increase in two years.

The AI software market is growing rapidly too, though from a smaller base. Enterprise AI spending on software and services is estimated to reach $150 billion by 2027, up from roughly $50 billion in 2024. Companies like Palantir, ServiceNow, and Salesforce have seen their AI-related revenue lines grow at double or triple-digit rates.

The stock market has responded accordingly. The Philadelphia Semiconductor Index (SOX) more than doubled from its 2022 lows. AI-related ETFs have attracted tens of billions in inflows. And the concentration of market gains in AI-exposed mega-cap stocks has reached levels not seen since the dot-com era.

But beneath the headline numbers, a more nuanced picture is emerging. Not every company claiming to be an “AI play” is actually generating meaningful AI revenue. And even among those that are, the question of valuation — how much you are paying for each dollar of earnings — remains critical.

Bubble or Revolution? Comparing AI to Past Tech Manias

Every generation of investors encounters at least one transformative technology that creates both enormous wealth and enormous losses. The railroad boom of the 1840s, the radio craze of the 1920s, the PC revolution of the 1980s, the internet bubble of the late 1990s — each followed a remarkably similar pattern. A genuinely important technology emerges, early investors make fortunes, enthusiasm becomes euphoria, valuations detach from fundamentals, and then reality reasserts itself.

The dot-com bubble is the most relevant comparison for today’s AI boom, so let us examine it carefully.

The Dot-Com Parallel

In March 2000, the Nasdaq Composite peaked at 5,048. It would not reach that level again until April 2015 — fifteen years later. Companies with no revenue, no business model, and no path to profitability traded at billion-dollar valuations. The poster children of the era — Pets.com, Webvan, eToys — are now business school case studies in what happens when speculation overwhelms reason.

But here is the part that bubble comparisons often miss: the internet actually was a transformative technology. Amazon survived the crash and became one of the most valuable companies in history. Google was founded during the bubble and went public after it burst. The winners of the internet era generated trillions of dollars in value — you just had to survive the crash and pick the right companies.

The comparison reveals important differences. Unlike the dot-com era, today’s AI leaders are generating massive real revenue and profits. NVIDIA earned over $70 billion in net income in fiscal 2025. Microsoft’s AI-related cloud revenue is growing at over 50% annually. These are not companies burning through cash with a dream — they are profit machines that happen to be at the center of a technological revolution.

That said, there are genuine parallels that should give investors pause. The sheer scale of capital expenditure being deployed is unprecedented, and the return on that capex is still uncertain for many companies. When Microsoft spends $80 billion on data centers in a single year, the implicit assumption is that AI will generate enough revenue to justify that investment many times over. That may prove true — but it is still an assumption.

The Bull Case — Why AI Is Different This Time

The optimists have a compelling argument, and it goes well beyond “this time is different” hand-waving. Here are the strongest pillars of the bull case for AI stocks.

Real Revenue, Real Profits

The single most important difference between the AI boom and the dot-com bubble is that AI companies are generating enormous, rapidly growing revenue from actual products that real customers are paying for. This is not vaporware. NVIDIA’s data center GPUs have multi-quarter backlogs. Microsoft’s Azure AI services are being adopted by Fortune 500 companies. Amazon’s Bedrock platform is seeing exponential growth in API calls.

Consider the trajectory: OpenAI’s annualized revenue reportedly exceeded $10 billion by late 2025, up from essentially zero three years earlier. Palantir’s AI Platform (AIP) drove revenue growth from 17% to over 30% year-over-year. ServiceNow’s AI-related annual contract value reached several billion dollars. These are not projections — they are reported numbers.

Measurable Productivity Gains

Unlike many dot-com era technologies, AI is already delivering measurable productivity improvements. McKinsey estimates that generative AI could add $2.6 to $4.4 trillion annually to the global economy. A Harvard Business School study found that consultants using AI completed tasks 25% faster and produced 40% higher quality results. Software developers using AI coding assistants report 30-55% productivity gains on certain tasks.

These productivity gains create a self-reinforcing cycle: companies that adopt AI gain competitive advantages, which pressures their competitors to adopt AI as well, which drives more spending on AI infrastructure and software. This is not speculative — it is happening right now across every major industry.

We Are Still in the Early Innings

Perhaps the most powerful bull argument is that AI adoption is still in its very early stages. According to various surveys, only 5-15% of enterprises have deployed AI at scale. Most companies are still in the experimentation or pilot phase. If you believe that AI will eventually be as ubiquitous as the internet or cloud computing, then the current level of spending and adoption represents just a fraction of the ultimate market size.

The Agentic AI Wave Is Just Beginning

The current AI boom has been driven primarily by large language models and generative AI. But the next wave — agentic AI, where AI systems can autonomously perform complex multi-step tasks — is just beginning. This could unlock trillions of dollars in value as AI moves from being a tool that assists humans to one that independently executes workflows, manages processes, and makes decisions.

Companies building agentic AI platforms (Salesforce with Agentforce, Microsoft with Copilot Studio, ServiceNow with AI Agents) are positioning for what could be the largest software market ever created. If agentic AI delivers on its promise, the current spending wave is not the peak — it is the foundation.

The Bear Case — Priced for Perfection

The bears are not arguing that AI is fake. The smartest AI skeptics freely acknowledge that artificial intelligence is a genuinely transformative technology. Their argument is more subtle and, in many ways, more dangerous for investors who dismiss it: AI stocks are priced for a future that may take much longer to arrive than current valuations assume, and the path from here to there may be far rockier than consensus expects.

The Valuation Problem

Let us be blunt about the numbers. In early 2026, NVIDIA trades at roughly 30-35 times forward earnings — which sounds reasonable until you consider that those “forward earnings” already assume continued exponential growth. If you normalize for a more sustainable growth rate, the multiple expands significantly. Palantir trades at over 60 times forward revenue (not earnings — revenue). Broadcom trades at 30+ times forward earnings with substantial AI-revenue growth already priced in.

The issue is not that these are bad companies. The issue is that the market is pricing in years of perfect execution with no stumbles, no competition, and no slowdown. History shows that even the best companies rarely deliver on that kind of expectation without at least a few disappointing quarters along the way.

The Capex Question — Will It Pay Off?

This might be the single most important question in the AI investment thesis. The hyperscalers are spending a combined $250+ billion per year on AI infrastructure. For that investment to generate acceptable returns, AI needs to produce revenue that justifies a massive acceleration in capital intensity.

Consider this simple math: if the hyperscalers spend $250 billion on capex and expect a 20% return on invested capital (ROIC), that infrastructure needs to generate $50 billion in annual incremental profit. That requires hundreds of billions in new AI-related revenue across the ecosystem. Is that achievable? Possibly. Is it guaranteed? Absolutely not.

Competition and Commoditization

Another bear concern is the risk of commoditization. In the early days of any technology wave, first movers enjoy fat margins because they are selling a scarce resource. But as more players enter the market and the technology matures, margins compress. We are already seeing signs of this:

• AMD and Intel are ramping AI GPU production, potentially challenging NVIDIA’s dominance
• Custom AI chips from Google (TPUs), Amazon (Trainium/Inferentia), and Microsoft (Maia) reduce dependence on NVIDIA
• Open-source AI models are closing the gap with proprietary ones, potentially commoditizing the model layer
• Chinese AI companies like DeepSeek have demonstrated that competitive AI models can be built at a fraction of the expected cost

If AI hardware and software become commoditized faster than expected, the margins that justify current valuations could erode significantly.

Regulatory and Geopolitical Risk

AI faces increasing regulatory scrutiny worldwide. The EU AI Act is already in effect, with compliance costs and restrictions that could slow adoption. The US is tightening export controls on AI chips to China, which directly impacts NVIDIA’s addressable market. And the broader geopolitical tension between the US and China creates supply chain risks, particularly for companies like TSMC that manufacture the most advanced AI chips.

A severe escalation in US-China tensions — or a crisis involving Taiwan — could devastate the AI supply chain and send shockwaves through every AI-related stock.

Picks and Shovels — Infrastructure Layer Investing

During the California Gold Rush, the people who most reliably made money were not the miners — they were the merchants selling pickaxes, shovels, and denim pants. The same principle applies to AI. Regardless of which AI applications ultimately win, the companies providing the underlying infrastructure are generating real revenue today.

NVIDIA (NVDA) — The Undisputed AI Infrastructure King

There is simply no way to discuss AI investing without starting with NVIDIA. The company controls an estimated 80-90% of the AI training GPU market and a dominant share of AI inference as well. Its CUDA software ecosystem creates a powerful moat — developers have spent years building on CUDA, and switching costs are enormous.

NVIDIA’s financial trajectory has been nothing short of extraordinary. Data center revenue grew from $15 billion in FY2023 to over $115 billion in FY2025. Gross margins have expanded to over 70%. Free cash flow has exploded. The company is not just selling chips — it is selling an entire AI computing platform that includes hardware, networking (Mellanox/InfiniBand), software (CUDA, cuDNN, TensorRT), and increasingly, full data center solutions (DGX, MGX).

The bull case: AI training and inference demand continues to grow exponentially, NVIDIA’s Blackwell and subsequent architectures maintain technological leadership, and the company’s software moat keeps competitors at bay. The bear case: custom AI chips erode NVIDIA’s market share, the capex cycle eventually slows, and competition from AMD intensifies.

Broadcom (AVGO) — The Custom Chip and Networking Giant

Broadcom has emerged as a stealth AI winner through two key businesses: custom AI accelerators (ASICs) and networking. The company designs custom AI chips for hyperscalers — Google’s TPU is partially designed by Broadcom, and multiple other tech giants have engaged Broadcom for similar projects. Additionally, Broadcom’s networking solutions (including Tomahawk switches and Jericho routers) are critical infrastructure for AI data centers.

Broadcom’s AI revenue is growing at triple-digit rates and is expected to reach $15-20 billion annually. The company’s diversified revenue base (networking, broadband, storage, wireless) provides downside protection that pure-play AI companies lack. CEO Hock Tan has proven to be one of the most astute capital allocators in the semiconductor industry.

TSMC (TSM) — The Foundry That Makes It All Possible

Taiwan Semiconductor Manufacturing Company is the indispensable company in the AI supply chain. Every advanced AI chip — from NVIDIA’s H100/H200/B100 to Apple’s M-series to AMD’s MI300 — is manufactured by TSMC. The company’s technological lead in advanced node manufacturing (3nm, 2nm) is measured in years, not months.

TSMC’s AI-related revenue has surged, with the company reporting that AI accelerators now represent a substantial and rapidly growing portion of its total revenue. The bull case for TSMC is straightforward: more AI = more chips = more TSMC revenue. The bear case centers on geopolitical risk (Taiwan’s proximity to China) and the possibility that rising competition from Samsung or Intel Foundry could eventually erode TSMC’s dominance.

The Software Layer — AI Application Plays

If the infrastructure layer is the “picks and shovels” of the AI gold rush, the software layer is where the gold itself is being mined. These are the companies building AI-powered applications that enterprises actually use — and pay for. The opportunity is enormous, but so is the uncertainty about which companies will capture the most value.

Palantir (PLTR) — The AI Platform for Enterprises and Governments

Palantir has been one of the most controversial AI stocks, and for good reason. The company’s stock has delivered extraordinary returns — rising several hundred percent from its 2023 lows — driven by explosive growth in its Artificial Intelligence Platform (AIP). Revenue growth has accelerated from the mid-teens to over 30% year-over-year, with US commercial revenue growing even faster.

Palantir’s strength lies in its ability to integrate AI into complex, real-world decision-making environments. Its “boot camp” approach — intensive workshops where customers build AI solutions on the Palantir platform in days rather than months — has proven remarkably effective at driving adoption and converting pilots into long-term contracts.

The catch? Valuation. Palantir trades at an eye-watering premium — over 60 times forward revenue and well over 100 times forward earnings. At these levels, the stock is priced for years of 30%+ revenue growth with significant margin expansion. That is achievable, but there is very little room for disappointment.

Salesforce (CRM) — The CRM Giant Pivots to AI

Salesforce has bet its future on Agentforce, its agentic AI platform that enables businesses to deploy autonomous AI agents for customer service, sales, marketing, and more. The company’s massive installed base of CRM users gives it a natural distribution advantage — it does not need to convince companies to adopt a new platform, just to upgrade their existing one.

Early results have been encouraging, with Agentforce signing up thousands of customers and showing strong engagement metrics. Salesforce CEO Marc Benioff has called Agentforce the “most important technology shift” in the company’s history. The stock trades at a more reasonable valuation than Palantir — roughly 25-30 times forward earnings — making it a less speculative way to play the enterprise AI adoption trend.

ServiceNow (NOW) — The Enterprise Workflow AI Play

ServiceNow is one of the most quietly impressive AI stories in the market. The company’s Now Platform, which automates enterprise workflows across IT, HR, customer service, and other functions, is a natural fit for AI augmentation. ServiceNow has embedded AI capabilities across its platform, including AI-powered search, virtual agents, and predictive analytics.

The numbers speak for themselves: ServiceNow’s annual recurring revenue has consistently grown at 20%+ rates, with AI-related bookings accelerating. The company generates strong free cash flow and has a reputation for disciplined execution. At roughly 50-60 times forward earnings, it is not cheap — but it is arguably one of the highest-quality enterprise software companies in the world.

Separating Real AI Revenue from Hype

One of the most critical skills for AI investors is distinguishing between companies with genuine, growing AI revenue and companies that are merely sprinkling “AI” into their earnings calls. Here is a framework for doing so:

Signs of real AI revenue:

• The company discloses specific AI-related revenue figures or growth rates
• AI features are driving measurable improvements in customer retention, upsell, or net new bookings
• Customers are paying premium prices for AI-enhanced products
• The company’s AI products solve specific, measurable problems (not vague “intelligence” promises)
• Management can articulate a clear AI monetization roadmap with milestones

Signs of AI washing (hype without substance):

• AI is mentioned frequently on earnings calls but never quantified
• “AI-powered” features are rebrands of existing analytics or automation
• The company has no disclosed AI-specific revenue or bookings data
• AI initiatives are described in vague, future-tense language with no timeline
• The company’s core business has no natural AI advantage but is pivoting anyway

The Magnificent 7 Problem — Concentration Risk

Perhaps no single data point better illustrates the state of the AI trade than this: in 2024 and into 2025, the Magnificent 7 stocks (Apple, Microsoft, Alphabet, Amazon, NVIDIA, Meta, Tesla) accounted for roughly 30-35% of the entire S&P 500’s market capitalization and an even larger share of its total returns. This level of concentration is historically unprecedented and creates risks that even bullish investors need to understand.

The Numbers Are Striking

The Magnificent 7 collectively are worth over $15 trillion — more than the GDP of every country on earth except the United States and China. Their combined weight in the S&P 500 means that a significant decline in these stocks would drag down the entire index, regardless of what the other 493 stocks are doing.

This creates a paradox for index fund investors. If you own an S&P 500 index fund, you are making a massive bet on a handful of AI-exposed mega-cap stocks whether you intended to or not. Your “diversified” index fund is actually a concentrated bet on the continuation of AI-driven mega-cap outperformance.

Why Concentration Matters for Your Portfolio

Concentration risk cuts both ways. When the Magnificent 7 are performing well, portfolios heavy in these stocks outperform. But when sentiment shifts — even temporarily — the damage can be severe. We saw a preview of this in the summer of 2024 when a rotation out of mega-cap tech briefly sent the Magnificent 7 stocks down 10-20% while smaller stocks rallied.

For long-term investors, the key concern is not a temporary pullback but a sustained period of underperformance. If the AI trade disappoints — or if the market simply decides that these stocks are overvalued — a portfolio that is 30%+ concentrated in Mag 7 names could significantly underperform the broader market for years.

Building a Diversified AI Portfolio

So you are convinced that AI is a long-term megatrend — but you also recognize the risks of overpaying, overconcentrating, and picking the wrong stocks. How do you build a portfolio that captures AI’s upside while managing the downside?

The Barbell Strategy

One effective approach is a barbell strategy: combine a core holding of high-quality, profitable AI infrastructure companies with smaller positions in higher-growth, higher-risk AI application companies. This gives you exposure to the most certain part of the AI value chain (infrastructure) while maintaining upside optionality through application-layer bets.

Here is what that might look like in practice:

Core Holdings (60-70% of AI allocation):

• NVIDIA (NVDA) — The dominant AI infrastructure provider with proven revenue
• TSMC (TSM) — The indispensable foundry partner for all AI chips
• Microsoft (MSFT) — Diversified tech giant with the strongest enterprise AI distribution
• Broadcom (AVGO) — Custom chips and networking with diversified revenue base

Growth Holdings (20-30% of AI allocation):

• Palantir (PLTR) — High-growth AI platform with government and enterprise traction
• ServiceNow (NOW) — Enterprise workflow automation with strong AI integration
• CrowdStrike (CRWD) — AI-native cybersecurity, benefiting from rising AI-related threats
• AMD (AMD) — The most credible challenger to NVIDIA in AI GPUs

Speculative/Satellite (5-10% of AI allocation):

• AI-focused ETFs for broader exposure
• Emerging AI companies that go public
• International AI plays (Asian semiconductor companies, European enterprise AI)

Position Sizing and Risk Management

Even within an AI-focused portfolio, position sizing matters enormously. Here are some practical guidelines:

No single stock should exceed 10-15% of your total portfolio — no matter how convicted you are. NVIDIA might be the greatest AI investment of this generation, but even the best stocks can lose 50% from their highs during market panics. Can you stomach a 5-7% portfolio-level loss from a single stock?

Use dollar-cost averaging for AI stocks, especially at current valuations. Investing a fixed amount monthly or quarterly reduces the risk of buying everything at the peak. Given the volatility of AI stocks, this discipline can significantly improve your average entry price.

Rebalance regularly to prevent any single position from becoming too large. If NVIDIA doubles and becomes 20% of your portfolio, trim it back to your target weight. This is psychologically difficult — you are selling your winner — but it is essential risk management.

Beyond Individual Stocks — ETFs and Other Vehicles

For investors who want AI exposure without the risk of picking individual stocks, several ETF options exist:

• VanEck Semiconductor ETF (SMH) — Heavy exposure to NVIDIA, TSMC, Broadcom, and other AI chipmakers
• Global X Artificial Intelligence & Technology ETF (AIQ) — Broader AI exposure across hardware and software
• iShares Expanded Tech-Software Sector ETF (IGV) — Software-focused with significant AI application exposure
• Invesco QQQ (QQQ) — Nasdaq 100 exposure, heavily weighted toward AI beneficiaries

ETFs provide instant diversification and professional management, but they also include companies you might not want to own individually. Read the holdings before buying — an “AI ETF” might have very different composition than you expect.

The Importance of Time Horizon

Perhaps the single most important factor in AI investing is your time horizon. If you are investing for the next 10-20 years, the case for AI stocks is strong despite current valuations. Technology revolutions tend to be underestimated in their long-term impact, even when short-term valuations seem stretched.

But if you need this money in 2-3 years, the risk calculus changes dramatically. AI stocks could easily decline 30-50% from current levels during a recession, a capex slowdown, or simply a change in market sentiment. The underlying technology would still be transformative, but your portfolio would not care about the long-term thesis while it is down 40%.

This is the fundamental tension of AI investing in 2026: the technology is real, the opportunity is enormous, but the prices reflect a lot of optimism. The investors who will do best are those who combine conviction in the thesis with discipline in execution — buying quality, managing position sizes, and maintaining a long enough time horizon to ride out the inevitable volatility.

Conclusion — The Long View on AI Investing

So, are AI stocks still worth buying for the long term? The honest answer is: it depends on what you buy, how much you pay, and how long you are willing to hold.

The AI revolution is real. Unlike many previous technology hype cycles, AI is generating hundreds of billions in actual revenue, delivering measurable productivity gains, and being adopted by enterprises at an accelerating pace. The companies at the center of this revolution — NVIDIA, TSMC, Microsoft, Broadcom — are not speculative startups. They are some of the most profitable, well-managed businesses in the world.

But “the revolution is real” does not automatically mean “the stocks are cheap.” Many AI stocks are priced for years of perfect execution, leaving little margin of safety for investors. The capex cycle could slow. Competition could compress margins. Regulation could create headwinds. And the concentration of AI-related gains in a handful of mega-cap stocks creates systemic risk that even diversified index fund investors cannot fully escape.

The path forward for most investors is a balanced one:

1. Stay invested in AI — The long-term opportunity is too significant to ignore. Sitting on the sidelines while AI transforms every industry is a risk in itself.
2. Focus on quality — Buy companies with real AI revenue, strong competitive moats, and proven management teams. Avoid companies where “AI” is a marketing slogan rather than a revenue driver.
3. Manage concentration — Do not let AI stocks become your entire portfolio. Diversify across the AI value chain (infrastructure, software, applications) and beyond AI entirely.
4. Use time to your advantage — Dollar-cost average into positions rather than making large lump-sum bets at potentially elevated prices.
5. Have a plan for volatility — AI stocks will have 20-30% drawdowns. Knowing this in advance and having a plan (buy more, hold steady, or trim) prevents emotional decision-making.

The investors who made the most money from the internet did not buy every dot-com stock at the peak and pray. They bought Amazon at $50, held through an 85% decline, and came out the other side with a 100x return. The AI era will likely follow a similar pattern — enormous long-term wealth creation for patient, disciplined investors, and painful losses for those who chase hype without a plan.

The AI revolution is happening. Your job as an investor is not to predict every twist and turn, but to position yourself to benefit from the long-term trend while surviving whatever volatility comes along the way. Buy quality, diversify, maintain your time horizon, and let compounding do its work.

References

1. NVIDIA Corporation — Fiscal Year 2025 Annual Report and Earnings Releases (investor.nvidia.com)
2. McKinsey Global Institute — “The Economic Potential of Generative AI,” June 2023, updated 2025 (mckinsey.com)
3. Harvard Business School — “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity,” 2023 (hbs.edu)
4. Gartner — “Forecast: AI Semiconductors, Worldwide, 2022-2028,” 2025 (gartner.com)
5. S&P Global Market Intelligence — S&P 500 Concentration and Magnificent 7 Weighting Data, 2025-2026 (spglobal.com)
6. Bloomberg Intelligence — “AI Capex Tracker: Hyperscaler Spending Trends,” 2025-2026 (bloomberg.com)
7. Palantir Technologies — Quarterly Earnings Reports, FY2024-2025 (palantir.com)
8. Broadcom Inc. — Quarterly Earnings Reports, FY2024-2025 (broadcom.com)
9. ServiceNow Inc. — Quarterly Earnings Reports, FY2024-2025 (servicenow.com)
10. Salesforce Inc. — Agentforce Launch and Earnings Commentary, FY2025 (salesforce.com)
11. TSMC — Monthly Revenue Reports and Annual Investor Conference, 2024-2025 (tsmc.com)
12. U.S. Securities and Exchange Commission — Company Filings and Financial Disclosures (sec.gov)