AI Bubble vs. Dot-com Bubble: A Data-Driven Comparison

The rapid emergence of generative AI and the corresponding surge in tech investments has sparked debate: Are we witnessing an AI bubble comparable to the dot-com craze of 1999–2000? Proponents of the bubble thesis point to skyrocketing stock prices (e.g. NVIDIA’s share price up ~1300% since late 2022 (^[25])), huge private valuations (OpenAI at $730B, Anthropic at $380B, Databricks at $134B as of early 2026), and stories of hyped products lasting only until “the next hot AI announcement.” Skeptics counter that AI is an enduring foundational technology, with broad corporate usage (dozens of enterprise AI deployments daily) and strong near-term revenue growth. In practice, reality is complex: some sectors and stocks are undeniably riding a speculative wave, while others are grounded in real economic value.

This report systematically examines the “bubble question” through data and history. We begin by reviewing the dot-com bubble’s key features: how internet excitement drove tech valuations in the late 1990s and the conditions of its crash. We then profile the current AI boom: venture funding, stock market trends, and adoption statistics. Comparative tables highlight where numerical indicators align or diverge between the two periods. We survey expert analyses and survey data, presenting both bubble-alarms and bubble-doubters, and include case studies of representative companies (e.g. Cisco Systems in 2000 vs NVIDIA in 2024, Pets.com vs OpenAI, etc.). Finally, we discuss possible futures: if this ends in a bust akin to 2000–02, or if the apparent excesses will give way to sustained tech-driven growth in the 2020s.

Methodologically, we incorporate extensive quantitative data (market caps, P/E ratios, funding volumes, adoption statistics) and qualitative insights (surveys of investors, executive warnings, historical analogies). All claims are supported with credible sources, often with inline citations to news reports, academic studies, or industry data. By juxtaposing historical and contemporary data, this report seeks to clarify “where the numbers make sense and where they don’t.”

Historical Context: The Dot-Com Bubble (Late 1990s)

To assess the parallels, we first briefly summarize the dot-com era. From the mid-1990s onward, the commercialization of the Internet spurred a speculative boom. New web companies proliferated, often with business models centered on rapid user growth rather than immediate profits. Venture capital and public markets poured money into anything “.com,” and traditional valuation metrics like earnings and cash flow were frequently dismissed as irrelevant (^[26]). A typical storyline: an Internet startup goes public at a massive valuation despite little or no revenue, raising huge capital to burn on “customer acquisition” and expansion. Employees with stock options became overnight paper millionaires (^[26]), reinforcing the craze.

By March 2000, this frenzy peaked. The Nasdaq Composite index (heavy with tech stocks) hit roughly 5,048 (^[27]), three times its level in early 1998. Valuation metrics soared: for example, the forward P/E ratio of the Nasdaq-100 reached ~60× at the bubble’s peak (^[15]). Many large-cap tech firms (Cisco, Intel, Yahoo, Sun Microsystems, etc.) traded at sky-high price/sales ratios. Notably, only a small fraction of dot-com companies were profitable; one analysis puts it around 14% (^[16]). The common narrative was summed up by investor Rob Arnott decades later: “the narrative was correct, but the market bet that narrative would play out a lot faster than it ultimately did.” (^[21])

In early 2000, when the Federal Reserve raised interest rates, the bubble began to pop (^[28]). Investors panicked; between March 2000 and October 2002 the Nasdaq plunged from 5,048 to 1,139 (^[27]), erasing virtually all of its gains. Thousands of dot-com ventures failed or were acquired for pennies on the dollar. The collapse inflicted huge losses on many investors, teaching the financial world that “traditional factors” (assets, profits, cash flow) still mattered (^[26]). In retrospective analyses, the dot-com bubble was characterized by extreme over-optimism, highly concentrated stock ownership, and reckless valuations.

We summarize key dot-com era data below:

Indicator	Dot-Com Bubble Peak (~2000)	Source
Nasdaq Composite Index peak	5,048 (March 2000) ([27])	Britannica
Nasdaq-100 Forward P/E	60.1× (March 2000) ([15])	VisualCapitalist
% of dot-com firms profitable at peak (one estimate)	~14% ([16])	LinkedIn Pulse (citing historical data)
Top tech company by market cap	Cisco Systems: ~$371B (Mar 2000) ([29])	StatMuse*
Cisco price/sales at the March 2000 reference point	About 20× FY2000 sales, using a ~$371 billion market-cap estimate and Cisco’s $18.928 billion FY2000 sales	Calculated from Cisco’s annual report and the stated market-cap estimate
VC funding (US, peak year)	~$112.3B (2000)	NVCA data
Internet/Web adoption	~50% of US households had Internet (2000)	Pew Research

*StatMuse indicates Cisco’s market cap ~$371B on March 31, 2000 (^[29]); other estimates put it higher at times.

These figures paint 1999–2000 as a time of overheated tech valuations and irrational exuberance, culminating in a crash. The question today is whether the AI-driven markets of 2022–25 match that profile.

The Modern AI Boom: Data and Trends

Since late 2022, generative AI (spurred by tools like ChatGPT, Anthropic’s Claude, Google Gemini, etc.) has ignited massive interest across industries. Investors and companies are rushing to “get on the AI train.” Below we analyze key metrics for the AI era.

AI Market Growth and Spending

Multiple research firms forecast tremendous AI market expansion. For example, IDC estimated that global annual spending on AI (hardware, software, services) will more than double to about $632 billion by 2028 (^[30]). Data center deployments for AI accelerators are being planned worldwide. Bain & Co. notes that to feed the generative-AI revolution, companies will need combined revenues of about $2 trillion by 2030 to pay for computing power, yet likely will come up nearly $800B short (^[22]) (Bain’s analysis suggests a substantial "AI compute funding gap").

Venture investment echoes this growth: AI startups have been attracting historically large funding. For full-year 2025, OECD data shows AI firms captured $258.7 billion in VC globally (61% of all global VC investment), up from 30% in 2022 (Crunchbase tallies $211B, an 85% increase from $114B in 2024). In the U.S. alone, AI attracted 75% of all VC ($194B). The concentration at the top is extreme: the top 5 deals accounted for ~$63 billion (25% of total), and mega-deals comprised 73% of total AI VC value. By comparison, no single sector dominated VC to this extent during 1999–2000. This extraordinary concentration reflects both excitement about AI’s potential and fear of missing out (FOMO). The scale of individual rounds has reached jaw-dropping levels: OpenAI announced a $110 billion round in February 2026 at a $730B pre-money valuation (it closed at $122 billion in March), while Anthropic raised $30 billion at a $380B valuation the same month. Databricks CEO Ali Ghodsi explicitly commented in late 2024 that AI fundraising seemed “peak”; his company raised $10B at a $62B valuation amid this fervor (^[23]), yet by early 2026 Databricks had soared to a $134 billion valuation with $5.4B in annualized revenue.

The table below compares some scale metrics of the two eras:

Metric/Indicator	Dot-Com Era (Peak ~2000)	Current AI Era (2024–2025)	Source(s)
Index Valuation: Forward P/E (Nasdaq)	~60.1× (Mar 2000) ([15])	~26× (Nov 2023); S&P 500 at ~23× (Feb 2026) and 19.0× (end of Sep 2026) ([10])	VisualCapitalist, FactSet
Major Tech Market Cap (1st rank)	Cisco ~$370–400B (Mar 2000) ([29])	NVIDIA ~$4.3T (Feb 2026), just under $5.7T (early Oct 2026) ([2]), world's most valuable company	AP, Reuters, StatMuse, CNBC
Top 3 Tech firms share of index	Very high (e.g. Microsoft, Cisco, Intel dominated Nasdaq)	Also very high (Nvidia, Apple, Microsoft lead S&P tech) ([31])	Reuters/Zalwora analysis
VC funding in tech sector	~$112B (2000 total US venture)	FY2025: AI startups $258.7B globally (61% of all VC, per OECD); US alone: $194B (75% of US VC)	OECD, PitchBook, Crunchbase
% startups unprofitable	Extremely high (est. 80–90% of dot-com IPOs)	High for AI unicorns: e.g. 70% of some AI-funded start-ups have no revenue ([19])	Business Insider, Tekta.ai (market reports)
Corporate Adoption by firms	No directly comparable measure cited	88% of McKinsey survey respondents reported regular AI use in at least one business function in 2025; 7% reported fully scaled use	McKinsey Global Survey 2025
Per-Employee Funding	N/A (less service-intensive tech)	$0.4–$1.2 billion per AI employee for some startups ([18])	Reuters Q.

Several patterns stand out. First, valuations of major AI firms are huge but not unprecedented. NVIDIA’s ~$4.3T market cap (February 2026) is larger in nominal terms than any dot-com-era company, but its profit margins are also enormous: NVIDIA posted $215.9 billion in FY2026 revenue with gross margins of 71% and net margins of ~53%, far above historical averages. By contrast, Cisco’s market cap was hugely inflated relative to its slower growth. Second, some new pressures appear: surveys show roughly 54% of fund managers now call AI stocks a bubble (^[20]), implying caution. Third, concentration is at historic extremes: by late 2025, the 5 largest companies held 30% of the S&P 500 and 20% of MSCI World, the greatest concentration in half a century (^[31]).

Many numbers of concern arise. For instance, Bryan Yeo of Singapore’s GIC warns that investors are paying astronomical multiples for minimal revenues (^[32]). Indeed, Reuters documents AI startups valued hundreds of millions per employee (^[32]), whereas in 2000 dot-com startups were often valued e.g. at tens of times forward sales. The Bain projection that AI companies will need $2 trillion in annual revenue by 2030 to pay for compute, and may fall about $800 billion short (^[22]) suggests an enormous mismatch between hype and actual business outcomes. Gartner, for its part, has placed generative AI in the “trough of disillusionment” of its hype cycle even as it expects spending to keep rising (^[33]).

In summary, investment scale is larger than ever, but many indicators caution that valuations and financing operate under bubble-like dynamics (largest stakes from small companies, funding disproportionate to revenues, etc.). We next turn to deeper comparisons of underlying fundamentals.

Adoption and Revenue

Beyond investing money, what concrete revenues and users back this boom?

• Enterprise adoption: By 2025, 71% of organizations regularly use generative AI in at least one business function (up from 65% in early 2024), and 87% of large enterprises have implemented AI. OECD data shows firm-level AI adoption doubled from 8.7% (2023) to 20.2% (2025) across member countries. Enterprise GenAI spending hit $37 billion in 2025, up from $11.5 billion in 2024, a 3.2× increase. Major corporations like Walmart, Goldman Sachs and many others are rolling out AI across business units. This suggests genuinely broad engagement with AI tech, far beyond hype projects.

• Consumer usage: Tools like ChatGPT accumulated 100 million monthly users within two months of launch, unprecedented for any consumer app. Billions now use voice assistants, recommendation engines, or image-generation. In contrast, by 2000 the Internet (web/email) had neither such scale of instant global adoption (just over half of US households were online) nor such visible consumer “killer apps” (social media was nascent, mobile was infancy).

• Revenue Growth: Companies central to AI are reporting hefty sales growth. NVIDIA posted $130.5B in FY2025 revenue (up 114% YoY), then $215.9B in FY2026 (up 65% YoY), with Q4 FY2026 alone at $68.1B. Meta Platforms reported Q4 2025 revenue of ~$59.9B, with AI-driven ad targeting fueling continued growth even as it plans $115–135 billion in AI capex for 2026. Cloud providers (AWS, Azure, Google Cloud) report strong AI services uptake. In contrast, most pure-play dot-com firms of 2000 had negligible revenues and heavy losses.

• Profitability: Among large firms: NVIDIA’s ~53% net margin and 71% gross margin far exceed industry averages, reflecting the unique economics of AI chips. But many AI-specialist startups still burn enormous amounts of cash. OpenAI generated ~$13 billion in 2025 revenue but does not expect to turn a profit until 2030, with projected cash burn of $17B (2026), $35B (2027), and $47B (2028). Anthropic, despite 10× revenue growth in 2025, remains deeply unprofitable. This mirrors dot-com’s pattern of a few big survivors versus many money-losing startups, though the revenue scale of today’s AI startups far exceeds what dot-coms achieved.

In summary, product-market traction is clearer in AI today than in 2000’s Internet boom. High adoption and revenue growth lend credibility to the AI revolution. Yet the extraordinary funding (a significant fraction of future worldwide IT spend) is far ahead of actual profits generated. The 2T/1.2T Bain gap (^[22]) highlights that fundamental returns will need to materialize to justify the hype.

Market Valuations and Metrics

Next, we examine market valuation metrics, drawing direct parallels.

• Stock Index P/E: A 2023 VisualCapitalist chart shows the Nasdaq’s forward P/E at ~60× in Mar 2000 vs ~26× in Nov 2023 (^[15]). This suggests that, on aggregate, tech stocks today are cheaper relative to earnings than during the dot-com apex. (Of course, tech today includes substantial profits, whereas the 2000 Nasdaq included many lossmakers).

• Market Concentration: Both eras saw tech mega-caps dominating. In 1999 the top 7 NASDAQ stocks (Cisco, Intel, etc.) comprised a large share. Today, the 5 largest companies hold 30% of the S&P 500 and 20% of MSCI World, the greatest concentration in half a century (^[31]). The similarity is stark: major indices now hinge on AI and big tech winners, just as the old NASDAQ hinged on dot-com incumbents.

• Top Company Values: Dot-com peak saw triple-digit billions. Today’s leaders are in the trillions. NVIDIA reached ~$4.3 trillion market cap by February 2026, making it the world’s most valuable company, supported by $215.9B in FY2026 revenue. Apple and Microsoft were worth about $4.9 trillion and $4.0 trillion in October 2026 (^[34]) (^[35]). These numbers dwarf any dot-com market caps. The vulnerability was starkly illustrated on January 27, 2025, when the DeepSeek shock wiped $588.8 billion from NVIDIA’s market cap in a single day (the largest single-day loss for any stock in history) after Chinese startup DeepSeek released a competitive AI model reportedly trained for just $5.6 million. Markets largely recovered, but the event highlighted how concentrated and fragile these valuations can be. Today’s tech bubble, if it is one, is concentrated in a handful of ultra-giants.

• P/S Ratios and Burn Rates: Many AI startups have astronomical price-to-sales ratios. Reuters notes cases where startups are valued $400M–$1.2B per employee (^[18]). By comparison, the approximately $371 billion March 2000 market-cap estimate and Cisco’s reported FY2000 sales of $18.928 billion give Cisco a price/sales ratio of about 20× (^[36]). What is new is the per employee framing. This is an extreme signaling of how much capital is flowing into few hands.

The following table summarizes some valuation comparisons:

Valuation Metric	Dot-Com Bubble	Current AI Boom	Source/Notes
Nasdaq Composite (peak)	5,048 (Mar 2000) ([27])	~22,668 (Feb 2026); S&P 500 ~6,879. Oct 9, 2026: Nasdaq 27,366, S&P 500 7,812 ([12])	Both indices set record closes on Oct 6, 2026; Nasdaq about 5.4× dot-com peak
Forward P/E (Nasdaq-100)	60.1× (Mar 2000) ([15])	S&P 500 ~23× (Feb 2026), most stretched since dot-com era; 19.0× (end of Sep 2026)	VisualCapt. chart; analyst data; FactSet
Top U.S. Tech Co. market cap	Cisco: ~$370B (Mar 2000) ([29])	NVIDIA: ~$4.3T (Feb 2026), just under $5.7T (early Oct 2026); Apple ~$3.5T (Feb 2026)	NVIDIA is world’s most valuable company
US VC to tech/AI as % of total	Not specialized (many sectors)	AI = 61% of all global VC (2025, $258.7B per OECD)	OECD, Crunchbase data
% of Tech IPOs unprofitable	~75–90%	AI unicorns: ~70% (per one report) ([19])	TechCrunch, Tekta.ai
Largest single deal (funding)	Google IPO (2004, after the bubble, $1.7B)	OpenAI $122B round (announced Feb, closed Mar 2026) at $852B post-money; Anthropic $65B at $965B (May 2026)	Largest private funding rounds in history

These data indicate that on traditional financial metrics, today’s valuations are high but not as extreme as 2000’s. Yet the rate of change is breathtaking, and the capital intensity (compute costs, hiring) is unprecedented. It also shows a key difference: whereas dot-com mania was sector-blind (anything Internet), AI mania is mainly in a narrow tech niche. Industry observers see this as risk concentration: if AI disappoints, a huge portion of market cap could evaporate.

Investor and Expert Sentiment

To gauge perceptions, we note surveys and expert statements:

• Fund Manager Surveys: Bank of America’s Oct 2025 survey found 54% of managers said AI equities are “in a bubble” or overvalued (^[20]), and 60% labeled global equities overpriced (^[20]). A Quartz summary reported similarly that 54% thought tech stocks (driven by AI hype) were too high (^[37]). Importantly, BofA’s analysts noted AI as the top “tail risk” for investors, surpassing inflation and geopolitical threats (^[38]).

• Executive Warnings: Several AI luminaries have publicly warned against overheating. OpenAI’s Sam Altman, after raising new funds, cautioned that investor enthusiasm may exceed what AI has proven (^[39]). Notably, Databricks CEO Ali Ghodsi, while raising a record round, explicitly dubbed it “peak AI bubble” (^[23]). These warnings from insiders contrast with earlier AI hype-phase bravado (e.g. “AGI is near” stories), indicating more sobering sentiment at the top of the cycle.

• Economists and Analysts: Industry economist Torsten Sløk (Apollo Global) argued that “the top 10 companies in the S&P 500 today are more overvalued than the top 10 companies were during the tech bubble in the mid-1990s” (^[40]). Rob Arnott, an investing pioneer, told the FT that AI hype is “a classic example of a big market delusion… just like the dot-com era” (^[24]). Meanwhile, other experts (e.g. from McKinsey, Bain) caution that we might have a “trough of disillusionment” yet to come in the AI hype cycle, or at least some consolidation.

• Venture Community: Surveys show a split: some believe AI valuations are unsustainable, others remain bullish on the long-term. There is no consensus: many VCs are riding the wave, but a sizable faction is nervous.

In summary, the subjective judgment of market participants is that AI is at least at risk of bubble-like euphoria. Many codify it as a bubble in progress (or imminent). Equally, all sides stress that AI itself, like the Internet before it, is real and transformative. The quantitative fallout may hinge on one’s perspective of timing and expectations.

Comparing Internet Mania and AI Hype: Key Similarities and Differences

Having laid out the facts, we now directly compare the two booms in several dimensions. This helps clarify where “the numbers make sense” and where they don’t.

Valuation Multiples and Index Levels

• Forward P/E and Price/Sales: As noted, dot-com stocks traded at far higher multiples (Nasdaq-100 P/E ~60× vs ~26× today (^[15])). Using the ~$371 billion March 2000 Cisco market-cap estimate and Cisco’s reported FY2000 sales of $18.928 billion yields a price/sales ratio of about 20×. Comparisons with current companies should use the same date basis and clearly identify whether they use trailing or forward sales.

• Index performance: The shape of the rallies differ. The Nasdaq’s rise to 5,048 by Mar 2000 was an exponential blow-off. The current S&P 500 and Nasdaq runs were strong in late 2023/early 2024, and both indices now stand far above their dot-com highs in nominal terms: the Nasdaq Composite closed at 27,366 on October 9, 2026, about 5.4 times its March 2000 peak (^[12]). However, drawing a direct index comparison is tricky (the economy, index composition, and Fed policy differ). For reference, after its 2000 peak the Nasdaq fell to 1,139 by October 2002 (^[27]), and much of today's gain has come from a few names.

Profitability and Business Models

• Maturity of Companies: Dot-com mania was filled with juvenescent startups, many with zero profits and untested business models. Many of those went bust when enthusiasm died. In contrast, most of today’s “big players” are decades-old firms branching into AI (MSFT, GOOGL, AMZN, NVDA). These incumbents had already generated sustaining revenues before AI and often strong cash flows. Therefore, even if AI investments disappoint, these companies have other businesses as a buffer.

• Unicorns and Startups: However, the surge in AI startup valuations is reminiscent of dot-coms. A late-2025 Reuters report noted investors are plowing money into AI startups regardless of modest revenues (^[32]). The per-employee valuations cited earlier (^[18]) are difficult to compare directly with the dot-com era because business models, labor markets, and capital requirements differ. Many AI startups still lack clear revenue models (e.g. generative content companies selling enterprise licenses or ad-supported consumer apps which may or may not scale). So as with dot-coms, the balance of risk may lie in the unsustainable projections of those new entrants, rather than the big incumbents.

• R&D and Infrastructure: Unlike the 1990s, today’s bubble (if it were one) is heavily capex-driven, meaning much money is going into physical infrastructure (data centers, GPUs). This has mixed implications. On one hand, heavy spending on hardware (100,000-GPU clusters, fiber networks, etc.) suggests durable capital assets are being built, not just marketing gimmicks. On the other hand, if demand projections prove too high, this capital could sit idle and impair returns. The Business Times notes that finance structures (debt, complex deals) for AI build-outs are eyebrow-raising (^[41]), indicating strain beyond pure equity.

Stock Market Behavior

• Concentration: As mentioned, both eras exhibit extreme concentration. In 1999, four companies (Microsoft, Cisco, Intel, Dell) made up a large share of the Nasdaq (^[27]). S&P Dow Jones Indices reports that the 10 largest companies represented almost 40% of the S&P 500 by mid-2025, a level not seen since the mid-1960s (^[42]). This means overall market movements can be heavily tied to the relative fortunes of a small number of entities. The danger was illustrated starkly on January 27, 2025, when the DeepSeek shock wiped $588.8 billion from NVIDIA alone, dragging the Nasdaq down 3.1% and the S&P 500 down 1.5% in a single session.

• Volatility and IPO Frenzy: The dot-com era saw hundreds of IPOs and huge year-to-year volatility. C3.ai’s initial public offering occurred in December 2020, while SoundHound became public through its April 2022 business combination with Archimedes Tech SPAC Partners; neither is an example of a 2023–2025 AI IPO (^[43]; SoundHound filing). Many AI unicorns remain private, so a direct public-market comparison with the dot-com IPO wave remains limited.

• Capital Market Responses: In 2000, investor loss in equities triggered a bear market broadly. Today, central banks are warier of deflation, and tech is a larger fraction of wealth portfolios. This could mean an AI sector shakeout might be contained to tech stocks, but the integrated nature of markets means fallout could be widespread.

Macro and Policy Environment

• Interest Rates: A major trigger for the dot-com bust was rising interest rates in 2000 (^[28]). The AI boom has followed a different path so far: the Federal Reserve raised rates sharply in 2022 and 2023, cut its target range six times between September 2024 and December 2025, then raised it by a quarter point to 3.75% to 4.00% on September 16, 2026, saying inflation remains elevated (^[13]) (^[14]). As in 2000, if economic growth or earnings disappoint, a higher cost of capital could prick the froth.

• Regulation and National Strategy: One major difference is that AI is now squarely on national policy agendas (from the U.S. Executive Office to the EU’s AI laws). Governments are planning for “AI safety” and competition. Dot-coms had essentially no such oversight at first. While regulation adds risk (could slow innovation), it also implies a strategic commitment that might prevent a total collapse, at least in large players.

Summary of Comparisons

Taken together, we see a mix of parallels and contrasts. Parallel: Early-stage exuberance, disproportionate capital inflows, high-profile skeptic warnings, and broad expectations of industry upheaval. Differences: Today’s AI landscape has far more tangible utility and revenue backing some of the hype (many AI companies make or plan to make money), and the “bubble” is largely in the valuations (and derivatives marketplace speculation) rather than wasteful consumer spending (as with web startups) or novel securities. In short, one can credibly argue that AI is partly a bubble but also a legitimate tech supercycle.

Data Analysis and Evidence

Here we present detailed data to substantiate the above statements. The Analysis and Implication columns indicate whether a given metric looks rooted in economic fundamentals (“where the numbers make sense”) or appears inflated (“where they don’t”).

Valuation Metrics

Metric	Dot-Com Era	AI Era	Analysis
Index P/E Ratio	Nasdaq 60.1× (Mar 2000) ([15])	S&P 500 ~23× forward (Feb 2026), most stretched since dot-com era; 19.0× (end of Sep 2026)	Dot-com multiples were extreme; the current multiple was elevated in early 2026 but eased below its 10-year average (19.1×) by late September 2026.
Index Level	Nasdaq 5,048 (Mar 2000) ([27])	Nasdaq ~22,668; S&P 500 ~6,879 (Feb 2026); Nasdaq 27,366, S&P 500 7,812 (Oct 9, 2026)	Both indices set record closes on Oct 6, 2026; Nasdaq about 5.4× dot-com peak (nominal).
Market Cap Top Co.	Cisco ~$370B (Mar 2000) ([29])	NVIDIA ~$4.3T (Feb 2026), just under $5.7T (early Oct 2026); Apple ~$3.5T (Feb 2026)	Top AI tech caps 10× larger; stronger earnings support justify some value.
Median Tech P/S Ratio	Very high (often >50×)	High, but mostly <50× (for big players)	Dot-coms often had no sales (→ undefined P/S). Current AI leaders trade on strong growth.
% of Big Tech Profitable	Low: about 14% of dot-com firms, by one estimate ([16])	High: NVIDIA, Apple, MS, Alphabet all profit	Dot-com era had many pure losses. Today’s leaders are profitable.
VC Funding Concentration	Less focused	AI = 61% of all global VC in 2025 ($258.7B per OECD)	AI draws overwhelmingly high VC share, crowding out other sectors (bubble sign).
Valuation per Employee	N/A	$400M–$1.2B per AI head ([18])	Surreal modern metric with no dot-com parallel, indicates intense heat (bubble indicator).

We cite NASDAQ P/E from VisualCapitalist (^[15]), and NVIDIA market-cap figures from AP/Reuters (^[44]) (^[45]). The table shows that purely on multiple grounds, today’s valuations are not as stretched as 2000’s (the P/E data is the clearest evidence). However, standardized comparisons (like P/S or per-employee) show anomalously large values today, hinting at speculative excess.

Investment Flows and Market Share

Metric	Dot-Com Era	AI Era	Implication
Total VC $ per year (global)	~$112B (2000 US)	~$427B global VC (2025); AI alone: $258.7B (OECD)	2025 AI VC alone exceeds 2× the total dot-com era VC.
% VC to Internet/Tech startups	Very high in late 90s (tech craze)	61% of all global VC to AI (2025, per OECD); up from 30% in 2022	Today’s capital allocation is extraordinarily skewed towards AI (bubble-like).
# of Tech IPOs	~527 US IPOs (1999)	Few mega-IPOs (C3.ai, Arm, etc.); many private	Dot-com issued IPOs en masse; AI companies largely raise private mega-rounds.
Avg. Time to profitability (year)	Didn’t exist for many (bust)	Many startups projecting profitability (some not)	Today’s unicorns often have obscured paths; some can survive longer (not easily).
Major M&A blow-ups	AOL-Time Warner merger (announced January 2000), followed by a $99B loss in 2002 after a goodwill write-off ([46])	Unclear so far	Past had large wasted acquisitions; no similarly wasteful spree seen yet in AI M&A.
Investor Positioning (survey)	N/A	54% say AI bubble ([20]), top tail risk ([38])	Majority of fund managers voice concerns now (sign of crowd wariness).

Notably, PitchBook and Reuters (^[32]) (^[47]) emphasize that over half of new venture funding is chasing AI, an asymmetry unparalleled even in 1999. This suggests either that investors now see AI as the only tech growth story (which may be short-sighted), or that the market is indeed unusually narrow. In 1999, tech IPOs were widespread; today, regulators and market conditions have meant the $60B+ funding rounds (e.g. OpenAI’s) happen behind closed doors. If that funding eventually turns into IPO listings, subsequent lock-up expirations may provide a more direct public-market comparison.

Company Case Studies

We examine how examples from each era illustrate “numbers that make sense/don’t make sense.”

Dot-Com Era Case Study: Pets.com (1999–2000)

• The Hype: Pets.com, an online pet-store, became an emblem of dot-com folly. Its sock puppet mascot ran expensive (albeit memorable) ads, and the company raised hundreds of millions of VC based purely on branding.
• The Numbers: Pets.com began operations in February 1999 and completed its initial public offering in February 2000. The offering sold 7.5 million shares at $11 per share for $82.5 million in gross proceeds. In July 2000, it acquired certain Petstore.com assets and partnerships.
• Outcome: On November 4, 2000, Pets.com’s board adopted a plan of complete liquidation and dissolution. The company later dissolved and its stock was delisted from Nasdaq in January 2001. Its failure illustrates the risk of a business whose operating model and financing needs cannot be sustained.

Contrast to today: It is rare today to see a publicly traded company valued in the hundreds of millions with no sales at all. Many of today’s AI startups remain private longer, and while they similarly bet on future growth (e.g. expecting to monetize large language models), the larger public AI companies have proven earnings. Pets.com stands for a case “where numbers don’t make sense”: high valuation, no revenue. Many argue AI has not (yet) reached Pets.com levels of absurdity publicly, though some startups could be in the same category.

AI Era Case Studies

NVIDIA (2022–2026)

• Context: NVIDIA, originally a graphics-chip maker, saw its stock explode due to AI demand for GPUs. From October 2022 to early 2026, its share price soared more than 10× (^[25]), reaching ~$177 per share by February 2026.
• The Numbers: By February 2026, NVIDIA’s market cap reached ~$4.3 trillion, making it the world’s most valuable company. Its FY2026 revenue hit $215.9 billion (up 65% YoY), with Q4 alone at $68.1B. GAAP gross margins reached 75% in Q4, with net profit margins of ~53%. NVIDIA guided Q1 FY2027 revenue to $78B, implying continued massive growth; it went on to report $96.2 billion for Q2 FY2027 (up 106% year over year) and guided to $108 billion for Q3 (^[1]). Its P/E ratio stood at ~47× as of February 2026, having moderated from far higher trailing multiples in 2023.
• Analysis: NVIDIA’s case is unique. It has extremely strong fundamentals (dominance in AI chips, huge backlog orders). The monstrous market cap partly reflects future growth priced in. The DeepSeek shock of January 27, 2025 was a stark reminder of vulnerability: NVIDIA lost $588.8 billion in market value in a single day after Chinese startup DeepSeek demonstrated a competitive AI model trained at a fraction of the cost. Markets recovered, but the event showed that even NVIDIA’s dominance can be questioned by efficiency breakthroughs that undermine the "more GPUs = better AI" thesis.
• Outcome (ongoing): NVIDIA remains hugely profitable and growing, unlike dot-com stocks (which generally died). Yet it accounts for a huge fraction of indices, so any correction in its stock heavily impacts the market. After its Q4 FY2026 earnings beat in February 2026, the stock still fell ~4–5% as investors questioned the sustainability of AI infrastructure spending. The top-heavy nature suggests a fragility akin to 1999 (when Cisco’s blink would sink a broker’s day).

OpenAI (2023–2026)

• Context: OpenAI is the high-profile AI startup, creator of ChatGPT. It remains private but has reached extraordinary valuations that have accelerated sharply through early 2026.
• The Numbers: OpenAI’s valuation trajectory illustrates the pace of AI mania: from ~$80B (2023) to $157B (late 2024) to $730 billion pre-money (February 2026), when it announced a record $110 billion funding round, the largest private funding round in history, with $50B from Amazon and $30B each from NVIDIA and SoftBank. The round closed on March 31, 2026 at $122 billion and an $852 billion post-money valuation (^[3]). OpenAI reported ~$13 billion in 2025 revenue and a ~$20B annualized run rate in early 2026, with 800 million weekly active users and 1 million business customers; by the end of September 2026 it told investors its annualized revenue was about $50 billion (^[4]).
• Analysis: Despite rapidly growing revenue, OpenAI does not expect to turn a profit until 2030, with projected cash burn of $17B (2026), $35B (2027), and $47B (2028). The company estimates it needs $50–80 billion in additional capital. Its $730B pre-money valuation at ~56× its 2025 revenue is extreme by any standard. OpenAI’s technology (ChatGPT) has clearly disrupted entire sectors (writing, coding, search), indicating real potential, but the economics remain deeply speculative. This is a case of “numbers on the hype side”: hundreds of billions of value today versus a company that may not profit this decade. It reflects dot-com mania (parallel: lack of profit, high market cap) but at a scale the dot-com era never approached. OpenAI confidentially filed for an IPO in June 2026 and is expected to list by early 2027, according to media reports (^[6]).

Others

• Meta Platforms (ex-Facebook): Meta has pivoted aggressively to AI, planning $115–135 billion in AI capex for 2026 (nearly doubling from $72B in 2025), driven by its “Meta Superintelligence Labs” efforts. Its market cap stands at ~$1.6 trillion (February 2026), and Q4 2025 revenue reached ~$59.9B. Unlike its earlier metaverse misadventure, Wall Street has broadly endorsed Meta’s AI spending given strong underlying business performance. However, the sheer scale of planned expenditure, comparable to the GDP of some countries, represents a bet on AI infrastructure that could prove excessive if AI revenue growth disappoints.
• Databricks: This data-analytics/AI platform startup raised at a $62B valuation in late 2024, when CEO Ghodsi called it “peak bubble” (^[23]). By early 2026, Databricks reached a $134 billion valuation (more than doubling in just over a year) with $5.4B in annualized revenue, a valuation-to-revenue ratio of ~25×. The company’s AI product revenue alone hit $1.4B annualized, demonstrating real traction, though the valuation remains eye-watering.
• Anthropic: Perhaps the most dramatic AI valuation story of early 2026: Anthropic raised $30 billion in a Series G round at a $380 billion valuation in February 2026, with annualized revenue of $14 billion, representing roughly 10× revenue growth sustained over three years. This valuation-to-revenue ratio (~27×) is more moderate than OpenAI’s, but the absolute numbers are staggering for a company still in its early years. In May 2026 Anthropic raised a further $65 billion at a $965 billion post-money valuation, with run-rate revenue above $47 billion (^[5]), and its IPO prospectus, as reported by Reuters in September 2026, shows a 2025 net loss of about $42 billion, including a roughly $34 billion accounting charge (^[6]).
• AI Startup Failures: Not all AI companies are thriving. Builder.ai, a Microsoft-backed startup that had raised about $445 million, entered insolvency proceedings in May 2025 after lowering its previously reported revenue (^[48]); media reports alleged that much of its “AI-powered” development relied on human engineers. The AI ecosystem is increasingly described as moving from “novelty to selection,” with the market filtering aggressively for companies with proprietary data, real unit economics, and deep enterprise integration.

These cases illustrate that very large valuations with minimal or nascent earnings (classic bubble attributes) are present in key AI ventures (OpenAI, Anthropic, Databricks) as in the dot-com era (Pets.com etc.). However, where dot-com firms often had zero path to profit, many AI companies have rapidly growing revenues and credible business models (e.g. software subscription, cloud services, API access), even if valuations remain far ahead of current earnings.

The juxtaposition of AI and dot-com eras raises important questions for investors, policymakers and businesses.

• Potential for Correction: If history is any guide, overheated markets often experience a correction followed by consolidation. In 2001–02, only a few survivors (Amazon, Google [in 2004], eBay) endured. Today’s tech giants are diversified enough to weather downturns, but smaller AI pure-plays could vanish. A painful reset might clear weaker projects and leave stronger companies better capitalized.

• Innovation vs. Speculation: One crucial difference is that, unlike 2000, we have already seen tangible AI value creation (e.g. massive productivity gains, new products). Thus even after a shakeout, the overall AI sector is likely to rebound to a higher base than pre-2022. McKinsey and other analysts argue that the next 5–10 years will see a productivity supercycle from AI (^[16]). If this plays out, much invested capital (in computing, talent, startups) will yield lasting economic gains rather than pure losses.

• Policy and Regulation: Governments are now acutely aware of AI’s disruptive potential. Regulatory actions (privacy laws, AI safety guidelines) could temper growth but also slow speculative excess (by imposing guardrails). In contrast, dot-coms faced almost no regulation initially (beyond general securities law). The new oversight might blunt irrational exuberance.

• Broad Tech Health: Even if a tech correction occurs, the broader economy may differ. The dot-com bust coincided with a recession. By early 2026, inflation remains persistent (complicating any central bank rescue), and the Nasdaq fell ~3% in February 2026 alone amid AI impact concerns. Big Tech has committed to an estimated $650–700 billion in combined AI capex for 2026, a 60–67% YoY increase. If AI revenue growth disappoints, this massive spending could weigh heavily on earnings and trigger broader market stress.

Given the stakes, various experts advise caution. Investing.com’s survey summary cautions that today’s record-high valuations and strategy suggest “bubble or fine line”: investors still hold tech overweight (^[20]). Barclays and Citi strategists have even labeled equities in general vulnerable, though they often temper that the AI-driven rally could still have legs before topping out (^[38]). On the positive side, PwC expects AI-driven transformation to be multi-year, not a two-year fad (^[16]).

Foresight demands monitoring of key signals: slowing venture flows, widening losses, major product disappointments (e.g. AI models failing to improve or losing public trust), or regulatory clampdowns. Early warning signs have already appeared: the DeepSeek shock demonstrated that efficiency breakthroughs can instantly wipe hundreds of billions in value; MIT research showed 95% of enterprises getting zero return on GenAI investments; and AI-attributed layoffs reached ~55,000 in the US in 2025, though the reality of "AI replacing jobs" has proven complex (Klarna, which had said its AI assistant did the work of about 700 customer-service agents, later said service quality had suffered and began hiring human agents again). Conversely, persistent aggressive expansion (Big Tech planning $650–700B in 2026 AI capex) and continued meteoric earnings (NVIDIA's $215.9B FY2026 revenue, Anthropic's 10× revenue growth) could validate the hype narrative and push the peak further out.

Two tables summarize key comparative indicators:

Comparison	Dot-Com Bubble (1999–2000)	AI Boom (2022–2026)
Peak Index Valuation	Nasdaq ~5,048 (Mar 2000) ([27])	Nasdaq ~22,668; S&P 500 ~6,879 (Feb 2026); Nasdaq 27,366, S&P 500 7,812 (Oct 9, 2026)
Index Forward P/E	Nasdaq-100 ~60× (Mar 2000) ([15])	S&P 500 ~23× forward (Feb 2026), most stretched since dot-com era; 19.0× (end of Sep 2026)
Large-company concentration	No directly comparable measure cited	The 10 largest companies represented almost 40% of the S&P 500 by mid-2025, a level not seen since the mid-1960s ([42])
Investor Sentiment	Broad euphoric belief in Internet as new economy	Mixed: 54% see AI assets as bubble ([20]); DeepSeek shock showed fragility; some analysts see the next few years as the main risk window
Adoption Rate	No directly comparable measure cited	88% of McKinsey survey respondents reported regular AI use in at least one business function in 2025; 7% reported fully scaled use
VC Funding Share	Dot-com specific - late 90s (peak ~$112B US in 2000)	AI = 61% of all global VC in 2025 ($258.7B per OECD)
Corporate Profits in Sector	Many dot-com businesses were unprofitable; Cisco itself reported FY2000 net income of $2.668B on $18.928B of sales (about a 14% GAAP net margin)	Big AI leaders profitable (NVIDIA margin 53%, $215.9B FY2026 rev); but OpenAI projects no profit until 2030
Potential Outcomes	Broad crash 2000–02 (Nasdaq down about 77% from its March 2000 peak to October 2002)	Consensus: selective correction likely rather than systemic collapse; Big Tech AI capex of $650–700B in 2026 creates risk if returns disappoint.

These portray that while both periods have “tech mania” label, the magnitude of hype vs reality differs. Dot-com fortunes almost entirely decoupled from fundamentals. AI mania, by contrast, is partially anchored: some revenues and adoption exist. Thus, the “numbers that don’t make sense” tend to be specific valuations (unicorn worth by revenue) and flows (VC share). The “numbers that make sense” include genuine technology spending forecasts and actual profitability of large firms.

Case Studies (In-Depth)

To ground the analysis, we highlight a few illustrative cases:

• Cisco vs. NVIDIA (Market Caps): Cisco’s market cap peaked at more than $500 billion in late March 2000, briefly making it the world’s most valuable company (^[49]) (various sources; StatMuse cites $371B by Mar 2000 (^[29]) but its intraday high was higher). Cisco earned a net margin of about 14% in FY2000, but its later growth fell far short of what that valuation implied. By contrast, NVIDIA’s February 2026 cap of ~$4.3T is supported by $215.9B in FY2026 revenue (up 65% YoY) and 71% gross margins. Cisco was leveraged to a yet-to-propagate Internet (similar to cloud in 2010). NVIDIA is leveraged to accelerating AI compute demand. So for Cisco, its dot-com-era cap made little sense (it deserved far less long-term), whereas NVIDIA’s cap, while still lofty, reflects tangible bookings and backlog. The comparison suggests Cisco’s valuation was radically inflated, while NVIDIA’s may be high but tied to real growth, though its P/E of ~47× still embeds enormous future expectations.

• Pets.com vs. OpenAI: Both lacked profits when highly valued. Pets.com spent millions on branding with no profits, while OpenAI raised $122B in a round announced in February 2026 and closed in March at an $852B post-money valuation, despite projecting no profit until 2030 and $17–47B in annual cash burn through 2028. In both cases, investors gambled on revolutionary change (e-commerce for pets; AGI for AI). Pets.com went bankrupt; OpenAI’s fate is unknown. The pattern is similar: hype monetized via investment, not results. The critical difference: OpenAI actually built working technology used by 800 million weekly users, generating ~$13B in 2025 revenue with rapid growth, whereas Pets.com’s tech was trivial. So OpenAI is far more substantive than Pets.com, but the extraordinarily high valuation (~56× its 2025 revenue at the $730B pre-money price, for a company that won’t profit this decade) is comparably speculative.

• Amazon (Old) vs. Palantir (Current): Amazon’s market cap peaked at only ~$40B around 2000 (^[50]) (bright today but tiny then), with Amazon still unprofitable. Amazon survived and was worth about $2.8 trillion in October 2026 (^[51]). Conversely, Palantir climbed sharply after its 2020 listing while still loss-making, then fell steeply in 2022 as investors soured on long-term profit potential; it has since recovered to a market value of about $500 billion (October 2026) (^[52]). Both Amazon (old) and Palantir illustrate turning points: Amazon showed that a loss-making dot-com could eventually dominate, while Palantir’s ride shows how quickly impatience with unclear profit timelines can reverse. Magic Software, sometimes cited as a 2025 AI IPO, is not one: its SEC filing states that its Nasdaq IPO occurred on August 16, 1991 (^[53]). These examples underscore that companies with novel tech but no profits can trade at very high premiums, until they can't.

Discussion of Implications

The evidence yields a nuanced picture:

• Bubble Symptoms: High concentration of capital, extreme forward expectations, many speculative valuations (AI “unicorns” whose per-employee numbers exceed realistic benchmarks (^[18])). Surveys indicate investors themselves fear a bubble (^[20]), funding trends are lop-sided, and there are alarm signs (like being named the top tail-risk (^[38])). If unchecked, these conditions historically presaged sharp corrections.

• Counterpoints / Enduring Value: Many in the tech community argue that unlike 2000, fundamental innovation is afoot. AI has verifiable productivity benefits (faster drug discovery, improved software automation, etc.), meaning that even a retrenchment could leave AI stronger. Additionally, the dot-com bust did not stifle infrastructure development (the fiber optics and data centers built around 2000 later supported Web 2.0), similarly AI hardware build-out may yield durable capacity.

• Risk of Overinvestment: Bain’s projection of a $2T compute requirement (^[22]) implies potential over-investment. The gap between spending and returns is stark: Big Tech plans $650–700 billion in combined AI capex for 2026, yet MIT research from August 2025 found that 95% of organizations are getting zero return on their GenAI investments, and US consumers spend only ~$12 billion/year on AI services. If revenues do not scale to match spending, capital destruction (as hedge investor Einhorn warned) is likely (^[54]). Historical cycles suggest that when a new tech requires vast up-front capital, many investors get hurt when realities emerge.

• Policy/Market Responses: Some regulation (like patent lawsuits, antitrust scrutiny, heightened M&A review) could slow hyperspeed deals. Also, any pronounced stock-market correction makes new capital harder to raise, which might naturally cool the bubble.

Looking forward, analysts envision either a “soft landing” or “hard landing” scenario for AI investments. In a soft scenario, delays in progress or modest growth rates allow valuations to moderate gradually, and tech giants absorb the adjustment. In a hard scenario, a sharp “pop” could cause knock-on effects (like tech layoffs, funding droughts, even a small recession if enough capital evaporates).

Given the complex stakes, stakeholders should remain vigilant:

• Investors should stress-test AI company valuations: Are prices pricing in perpetual 100% growth? If yes, that is often unrealistic. Look for businesses with clear revenue pipelines and profitability timelines.

• Companies with AI products must manage expectations: overpromising (e.g. on AGI) can lead to rapid collapses in trust. Transparency on metrics (adoption rates, cost savings, etc.) can help temper speculative valuations.

• Economists and Policymakers must distinguish between AI-driven productivity (which should be encouraged) and pure speculation. Fiscal or monetary intervention might be warranted if excesses threaten the broader economy, but should be balanced so as to not stifle genuine innovation.

Conclusion

In sum, the current AI investment frenzy bears resemblances to 1999’s dot-com bubble but also crucial differences. The “numbers that don’t make sense” today include sky-high valuations of AI startups with minimal or distant profitability (OpenAI at $852B post-money with no profit expected until 2030), extraordinary per-employee funding, the enormous share of capital funneled into AI (61% of all global VC in 2025), and Big Tech’s combined $650–700 billion in planned 2026 AI capex while most enterprises report zero returns on GenAI investments. Those echo bubble-like irrationality. However, the “numbers that make sense” (broad organizational adoption (71% of organizations regularly using GenAI, 87% of large enterprises), explosive revenue growth at major firms (NVIDIA’s $215.9B FY2026 revenue, Anthropic’s 10× revenue growth), and meaningful AI deployments) suggest a genuine technology wave powering those metrics. Much like the Internet in the 2000s, AI may be a generational shift, even if the hype moment is overdone.

Time will tell whether we are simply at the peak of a market cycle (followed by pullback) or at the cusp of a sustained transformation akin to the PC or mobile revolutions. History cautions us that not all that glitters is gold: speculators in dot-coms lost fortunes when the bubble popped. But equally, those who quietly built internet businesses enjoyed decades of growth afterward.

This report shows that disaggregating the data is key: some AI-related numbers (funding concentration at 61% of global VC, private valuations reaching $965B for Anthropic in May 2026, market concentration at 50-year highs) look like 1999, while others (NVIDIA's 53% profit margins, $215.9B revenue, rapid enterprise adoption) do not. The DeepSeek shock of January 2025 and the Nasdaq's 3% decline in February 2026 hint at the corrections that may lie ahead, while record-breaking fundraises by OpenAI and Anthropic suggest the mania has not yet peaked. Whether this turns into a classical bubble burst or a managed market correction, stakeholders should use the full range of evidence, from P/E ratios (^[15]) to surveys (^[20]) to industry forecasts (^[22]), in forming their judgments. What is clear is that, unlike 1999, we are not “in 1999 again” in every respect, and any single narrative (bubble vs revolution) is too simplistic.

Sources: This analysis is grounded in up-to-date research and market data, including Reuters, Bloomberg/AP/Quartz reporting, analyses by Bain & Co., Gartner/IDC forecasts, and academic surveys. Key statistics and quotes are cited inline; some figures summarize widely reported company disclosures and news coverage without an inline link.