Dot-com lessons for paid ads marketers in the AI bubble

The paid-ads lesson in the AI bubble versus dot-com bubble comparison is visible before either cycle reaches a stock-market postmortem. In both cases, a suddenly prominent group of advertisers bought expensive mass attention while investors were still funding expansion. Conspicuous concentration does not prove that a crash is coming, but it identifies the bidders whose withdrawal could abruptly change auction demand.

For a media buyer, the useful question is not whether AI will reproduce the Nasdaq chart. It is what a search or social account would show if bubble-funded advertisers started losing financing, cutting acquisition budgets and leaving auctions. The dot-com sequence provides a more practical template than the headline market loss.

What the dot-com retreat looked like from the ad market

The late-1990s growth model made advertising part of the financing cycle. Venture-backed internet companies pursued “get big fast” strategies, spending heavily on marketing to acquire users and establish brands before proving durable economics. As long as capital remained abundant, customer acquisition could be treated as expansion rather than as a cost that each customer had to repay.

The financing turn came quickly. The Nasdaq Composite peaked at 5,048.62 on March 10, 2000. By October 9, 2002, it had fallen to roughly 1,114—a decline of about 78% associated with approximately $5 trillion in lost market value.[2] Those figures establish the scale of the financial collapse, but the June 2000 reassessment of dot-com advertising campaigns is more relevant to anyone responsible for bids and budgets. The marketing retreat was already becoming an operating decision while the market decline was still unfolding.[2]

The Super Bowl made that change unusually easy to see. Dot-com companies had crowded into the January 2000 broadcast; a year later, the documented count had dropped from 17 to three.[1] Other historical tallies for the 2000 game range from 12 to 19 because researchers draw the category boundary differently. The exact count matters less than the direction and speed: an advertiser class that had been difficult to miss became scarce within one annual buying cycle.

Company failures then removed some demand permanently. Nearly 5,000 US internet companies reportedly failed, although the shakeout was not total: about 48% of dot-com companies survived through 2004.[2][3] A bubble can destroy businesses and investor capital without erasing the technology category or every advertiser associated with it.

Online advertising itself continued to develop. Google launched AdWords in 2000, during the crash rather than after the sector had fully recovered.[2] That timing is a useful corrective to the idea that an AI funding reversal would eliminate automated advertising, generative products or digital media. The dot-com crash separated viable businesses from companies whose advertising depended on a continuing supply of outside capital.

There is an important evidentiary limit here. The historical IAB/PwC annual online-ad-revenue series for 2000–2002 was not available in the reviewed materials. The sourced record supports heavy marketing expenditure, campaign reassessment, a sharp fall in promotional visibility and widespread company failure. It does not support a definitive claim here about whether total annual US online-ad revenue contracted, or by precisely how much.

How fewer funded advertisers reset an auction

Auction effects do not require the entire ad market to shrink. If a subset of advertisers has unusually high budgets, aggressive growth targets and loose payback requirements, it can raise the clearing price for everyone competing for the same queries, audiences or inventory. When those advertisers cut spend or disappear, the composition of demand changes even if established brands keep advertising.

That change will not produce a uniform discount. Competitors may leave one query cluster while crowding into another. Platforms can alter placements, reserve prices, matching or optimization at the same time. Surviving advertisers can absorb newly available inventory. The historical lesson is therefore a mechanism—funding withdrawal removes bidders and resets pricing pressure—not a promise that every surviving account receives cheaper clicks.

The AI advertising boom has concentration, not a confirmed replay

Super Bowl LX supplied the visual parallel. Its 15 AI-related ads turned the February 2026 broadcast into an “AI Bowl,” with roughly 23% of all ads connected to the category.[1] That is evidence of advertiser concentration and willingness to pay for attention. It is not evidence that those companies are insolvent, that their campaigns are unprofitable or that a synchronized withdrawal has begun.

The category is also moving toward ad-funded products. OpenAI has described an $8-per-month ChatGPT Go tier and an approach to testing advertising while keeping paid tiers ad-free.[4] That creates another source of inventory and another monetization path, but it does not establish that chatbot advertising can already carry the cost base of frontier AI. The economics and supply constraints are examined separately in the analysis of AI chat subscriptions and ad supply.

Forecast scale adds pressure without settling the bubble question. eMarketer projects US AI advertising spending of $32.03 billion in 2026, rising to $68.25 billion by 2030. More than 80% of the 2026 amount is expected to appear alongside AI-generated content—such as search ads adjacent to AI Overviews—rather than inside chatbots.[5] Paid-search teams are therefore exposed to AI advertising economics even if they never buy a conversational ad.

The imbalance attracting concern sits behind that advertising activity. Sequoia framed the annual gap between AI infrastructure spending and AI revenue at close to $500 billion.[6] Goldman Sachs has separately questioned whether generative AI spending is producing sufficient benefit.[7] These are pressure indicators, not timers. They say little about which advertiser will cut a campaign first or which auction will feel the withdrawal.

The wider market is not currently forecasting an advertising collapse. IAB projected US ad spending to grow 9.5% in 2026 and reported that five of marketers’ six leading focus areas were AI-driven.[8] Adoption, strategic attention and aggregate growth can coexist with poor economics at individual AI companies. They can also conceal a change in bidder quality until budget approvals tighten.

One plausible synthesis is that advertiser demand financed directly or indirectly by the AI investment cycle is contributing to platform auction pressure. That is this site’s analysis, not a conclusion issued by eMarketer, IAB, Sequoia, Goldman Sachs or an ad platform. The transmission channels—from infrastructure spending through vendor revenue, hiring, customer acquisition and platform monetization—have different levels of verifiability. They are mapped in the analysis of how AI data-center growth can reach ad costs.

What the turn would look like inside an account

Start with who is actually bidding

Advertiser mix is the closest account-level equivalent to watching dot-com companies abandon major media buys. Search teams can record recurring competitors by commercially important query cluster rather than taking an occasional screenshot of Auction Insights. Paid-social teams have less direct auction visibility, but changes in creative-library activity, message frequency and the appearance of newly funded brands can still provide a directional view.

The unit of analysis should be narrower than “AI advertisers.” A model provider, an AI meeting assistant and a cloud platform may compete in different auctions and face different funding conditions. Group bidders only when they overlap with the account’s actual queries, audiences or placements. Otherwise, a broad category count can rise while the competitive set relevant to the account is already weakening.

Look for coordinated movement rather than treating one disappearance as a signal. A competitor can leave because it changed agencies, paused a geography, shifted match types or moved budget into another channel. The stronger pattern is a reduction in several aggressive bidders accompanied by lower overlap, falling impression-share pressure or fewer expensive conquest campaigns. Even then, the account data identifies an auction change; it does not reveal the competitors’ balance sheets.

The current-account companion, Three Signals the AI Ad Bubble Is Already Leaking, tracks evidence such as observed Meta CPM movement. The dot-com comparison adds a sequence to that evidence: identify the funded advertiser cohort, watch campaign reassessment, then test whether reduced participation changes auction pressure.

Reconcile cost drift with business outcomes

Platform-claimed lift deserves reconciliation, not automatic rejection. An automated campaign can discover demand, improve creative matching or increase measured conversions while auction prices rise. The buyer still needs to determine whether the improvement survives after accounting for conversion quality, margin, incrementality and any change in attribution.

A useful monitoring view keeps platform and business measures on the same timeline: CPM or CPC, reported conversion rate, CPA, qualified conversion rate, realized revenue and contribution margin where available. Annotate material budget reallocations, tracking changes, promotions and product launches. Without those notes, an apparent auction signal may simply be a measurement break or a change in the traffic being purchased.

The concerning pattern is sustained cost inflation paired with weaker downstream economics while the platform continues to report optimization gains based on its own conversion surface. That divergence does not prove bubble-funded demand is responsible. It does tell the operator that the claimed gain is not compensating for the price or quality change borne by the business.

If costs later ease, the same discipline applies. A lower CPM can accompany worse inventory, weaker intent or reduced conversion volume. Declining auction pressure is useful only when the account preserves profitable reach. A post-bubble environment does not guarantee improved ROAS.

Measure how much spend is trapped inside aggregated systems

Performance Max, Advantage+ and AI Max can all remain effective during an advertiser retreat. The exposure comes from concentration and observability. When most spend runs through campaign types that combine audiences, queries, placements or creative decisions, the buyer may see the blended result without being able to identify immediately where competitive demand disappeared.

Record the share of total spend in each automated campaign family and the share that can be moved without losing essential coverage or rebuilding measurement. Then preserve whatever independent controls the account can support: brand versus nonbrand separation, geographic comparisons, product-level margin reporting, placement exclusions where available and experiments that do not rely solely on platform attribution.

This is an exposure measure, not an argument for replacing automation with manual bidding. An account can be highly automated and well controlled if it has reliable business-side measurement and credible comparison surfaces. A nominally manual account can still be blind if its conversion data are weak.

Budget decisions may move before auction averages

The June 2000 campaign reassessments matter because committees and finance teams can change demand before failures appear in the news. Watch for shorter approval windows, delayed renewals, reduced test budgets, tighter payback requirements and abrupt pacing caps among clients or business units exposed to the same funding cycle.

Spend-cut velocity is often more informative than the final amount. The site’s examination of how Apple’s Vision Pro ad-spend cut preceded layoffs is a working example of the sequence, not evidence that every sharp cut predicts layoffs or failure. The monitoring value comes from noting when a company moves from expansion to preservation and whether peers begin doing the same.

A disciplined standard for calling an auction turn

As of August 28, 2026, the available evidence supports an AI advertising concentration and a substantial spend-versus-revenue tension. It does not establish the timing of a crash or prove that AI-funded advertisers are already withdrawing as a group. Macro warnings can frame the risk—including their possible ad-tech failure modes—but they cannot replace account observation.

A credible account-level call requires several linked observations: relevant AI advertisers disappear or reduce auction overlap; cost and performance move in ways that platform-claimed lift does not explain; budget approvals and pacing weaken across more than one isolated company; and a meaningful share of spend remains concentrated in automated systems that make the source of the change difficult to inspect.

That standard does not predict when the AI bubble will break, and it does not promise cheaper auctions afterward. It gives the media buyer something more defensible: a way to distinguish a macro story from a deterioration in bidder composition that is already changing the economics of the account.

References

1. Super Bowl Ads as a Bubble Warning — Acadian Asset Management
2. Dot-com bubble — Wikipedia
3. From Dot-Com to DeepSeek: 25-Year Tech Bubble Comparison for AI Era — Cheung Kong Graduate School of Business
4. Our approach to advertising and expanding access — OpenAI
5. US AI Advertising Forecast 2026 — eMarketer
6. AI’s $600B Question — Sequoia Capital
7. Gen AI: Too Much Spend, Too Little Benefit? — Goldman Sachs
8. Outlook Study Forecasts 9.5% Growth in U.S. Ad Spend — Interactive Advertising Bureau