The Hyperinflation Of Search; How AI Might Be About To Break Search's Unit Economics

With thanks to Google Gemini for being my research assistant on this piece. Trying it out to see how good it is - and it aced it!

The End of an Era: Search Before the AI Shockwave

For over two decades, the digital economy has been built on a remarkably stable and profitable foundation: search advertising. This model, perfected by Google, transformed the simple act of finding information into one of the most powerful economic engines in history.

The ground beneath this foundation is beginning to fracture. The emergence of generative artificial intelligence (AI) is not merely an incremental change but a seismic shift that threatens to shatter the long-standing economic principles of search. While many marketers are focused on the surface-level shifts in user behavior, such as declining click-through rates, they are missing the far more profound event happening in the background: a massive, impending explosion in the underlying computational cost of search itself.

This analysis will demonstrate that the rise of agentic AI will trigger an exponential increase in system-level queries, blowing out the energy and infrastructure costs for search providers—a cost that will inevitably be passed on to advertisers, fundamentally reshaping the financial calculus of customer acquisition.

A Brief History of Search's Economic Model

The internet's early days were characterized by manual curation. To find information, users navigated human-edited lists like the Open Directory Project (DMOZ) or Yahoo!'s directory.¹ The first true search engines, such as Aliweb (1993) and WebCrawler (1994), introduced the revolutionary concept of using automated "crawlers" to index web content, making every word searchable.¹

The pivotal moment came with the launch of Google in 1998. Founded by Larry Page and Sergey Brin, Google introduced the PageRank algorithm, which evaluated the importance of a webpage by analyzing the quantity and quality of links pointing to it.² This link-based ranking system provided vastly more relevant results, quickly establishing Google's dominance.

The economic genius of this model was not in charging users, but in monetizing their intent through advertising. While the concept of Pay-Per-Click (PPC) advertising was first introduced by Planet Oasis in 1996, it was a company called GoTo.com (later Overture) that pioneered the auction-based model where advertisers bid on keywords.⁶ Google perfected this system with the launch of Google AdWords in 2000, creating a highly efficient marketplace that connected advertisers with users at the precise moment of their expressed need.⁶ This model proved astoundingly lucrative, becoming the financial bedrock of the modern internet and generating nearly $200 billion in revenue for Google from search-related activities in 2024 alone.⁸

The First Tremors: AI and the Initial Dip in Search Consumption

After two decades of unparalleled dominance, the first cracks in Google's empire are appearing. For the first time in its history, Google is experiencing a sustained, measurable decline in its core search market share. Data shows its global share fell below 90% for the first time since 2015, a small but symbolically significant dip.¹¹ This is not a temporary market fluctuation but a structural shift driven by a new class of competitor: AI-powered "answer engines" like OpenAI's ChatGPT and Perplexity.

These platforms represent a fundamental change in how users, particularly younger demographics, interact with information. Instead of receiving a list of blue links to be clicked and explored, users receive direct, synthesized answers. This has given rise to the phenomenon of "zero-click searches," where a user's query is fully resolved on the search results page itself, eliminating the need to visit a third-party website. The trend is accelerating; one study found that 80% of users reported that AI tools helped them resolve nearly 40% of their queries without clicking a single link.¹³ Another analysis indicated that the percentage of zero-click searches, reported at 25.6% in 2022, had more than doubled to 58.8% by 2024.¹⁵

This shift has not gone unnoticed within Google. An internal document from late 2024 acknowledged that losing search traffic to AI—either its own Gemini platform or a rival like ChatGPT—is "inevitable".¹⁰ The concern is that the very foundation of the search business model—driving traffic to ad-supported websites—is being eroded.

However, this focus on the decline in user-initiated clicks and website traffic is a dangerous red herring. Marketers see these front-end metrics and understandably worry about the diminishing value of search as a marketing channel. Yet, this perspective completely misses the tectonic shift happening on the back-end. The architectural requirements of generative AI mean that for every complex user prompt, the search engine must execute a multitude of its own queries in the background to synthesize a coherent answer. This creates a perilous economic paradox: the perceived value of search (measured in clicks and traffic) appears to be declining at the exact moment that the underlying cost to provide the service is preparing to explode. This financial disconnect between falling monetization opportunities and skyrocketing operational costs is unsustainable and will force a dramatic repricing of the entire search advertising market.

The Physics of Cost: Quantifying the Energy Demands of AI

The coming cost explosion in search is rooted in the fundamental physics of computation. Generative AI is not a marginally more expensive technology; it is an order-of-magnitude leap in computational and, therefore, energy intensity. To understand the financial implications for advertisers, one must first grasp the sheer scale of the energy required to power this new paradigm, from the cost of a single query to the aggregate demand of global data centers.

The Energy of a Single Search: Establishing the Baseline

To quantify the change, we must first establish a baseline. According to a widely cited 2009 report from Google, a single traditional search query consumes approximately 0.0003 kilowatt-hours (kWh), or 0.3 watt-hours (Wh), of electricity.¹⁶ This process emits an estimated 0.2 grams of CO2.¹⁹ While this figure is now over 15 years old, and Google's own search has become more complex in the interim, it serves as the most commonly accepted benchmark for the energy cost of pre-generative AI search.

The Generative AI Energy Multiplier

The energy consumption of a generative AI query is a subject of intense debate, with estimates varying dramatically based on the model, the hardware, and the complexity of the prompt. Early analyses, likely based on older models like GPT-3, suggested a single ChatGPT query could consume around 2.9 Wh of energy—nearly ten times that of a traditional Google search.¹⁷ Some researchers have posited multipliers as high as 30x for generative AI compared to conventional search.²⁰

More recently, OpenAI CEO Sam Altman and independent researchers at Epoch.ai have argued that newer, more efficient models like GPT-4o are far less costly, requiring only about 0.3 to 0.34 Wh per average query.¹⁶ This would place its energy consumption on par with Google's 2009 baseline.

While the per-query multiplier is debated, the macro-level data tells an unambiguous story. The aggregate electricity demand from data centers is exploding, and AI is the primary driver. The International Energy Agency (IEA) projects that global electricity consumption from data centers, driven by AI, will more than double by 2030 to over 945 terawatt-hours (TWh)—an amount greater than the entire current electricity consumption of Japan.²¹ Other analyses project that data centers could account for as much as 21% of total global energy demand by 2030 when factoring in the costs of delivering AI to consumers.²² This surge is forcing utilities to plan for unprecedented load growth, with data centers expected to account for almost half of the growth in US electricity demand between now and 2030.²¹

The debate over whether an AI query costs 1x or 10x the energy of a traditional search is ultimately a distraction from the main economic event. The total electricity bill is a function of two variables: the energy cost per query and the total number of queries. Even if model efficiency gains keep the first variable stable, the architectural shift to agentic systems is about to cause the second variable to explode, overwhelming any per-unit savings and driving total energy costs to astronomical new heights.

The Engine of Inflation: How Agentic Search Multiplies Queries

The true engine of the coming cost inflation is not the energy cost of a single query, but the exponential increase in the total number of queries that will be executed by search systems. This multiplication is not an accident; it is a fundamental architectural requirement of the next evolution of search: agentic AI. This new paradigm, which shifts from passively providing information to actively completing tasks, is designed to break down complex human requests into numerous machine-level queries, creating a computational cost bomb.

From Answering to Acting: Defining Agentic Search

Traditional search engines retrieve information. Generative AI synthesizes information. Agentic AI, however, takes action. An AI agent is a system that can operate autonomously to achieve a defined goal, moving beyond simply responding to a prompt.²⁴ It is a shift from passive document lookup to active, intent-driven orchestration.²⁵ These systems are defined by their ability to use tools, maintain a memory of past interactions, and create multi-step plans to accomplish complex tasks.²⁶ Gartner predicts that search engine volume will decline by 25% by 2026 precisely because these AI agents will revolutionize how users interact with information, handling more complex needs that previously required multiple manual searches.²⁶

The "Query Fan-Out" Effect: A Walk-Through of Decomposition

The core mechanism driving the query explosion is known as Query Decomposition. To handle a complex, multi-faceted human prompt, an agentic system must first break it down into a series of simpler, discrete sub-queries that can be executed and answered independently.²⁹ The system then synthesizes the answers to these sub-queries to form a comprehensive response to the original prompt.

Consider this real-world walk-through:

• A user issues a single, complex prompt: "Plan a 3-day marketing team offsite in Lisbon for 15 people in October, focusing on team-building activities and a budget of €2,000 per person."

An agentic system cannot answer this with a single search. It must decompose the task and execute a multitude of background queries, a process Google itself calls "query fan-out".³² The agent would autonomously issue dozens, if not hundreds, of system-level queries, such as:

• "Average flight prices from Sydney to Lisbon in October"

• "Typical weather in Lisbon in October"

• "Top-rated hotels in Lisbon with conference facilities for 15 people"

• "Cost of private dining rooms in Lisbon restaurants"

• "Corporate team-building activity providers in Lisbon reviews"

• "Availability of AV equipment rental in Lisbon hotels"

• "Ground transportation options from Lisbon airport for large groups"

This is not a theoretical concept. It is precisely how next-generation search systems are being built. Microsoft's Azure AI Search uses an "agentic retrieval engine" that employs "query planning" to transform one complex query into many.³³ Anthropic's multi-agent research system uses a lead agent to decompose complex questions into parallel tasks for subagents to investigate simultaneously.³⁴ Google has stated that its "AI Mode" can issue hundreds of searches on a user's behalf to create a single, expert-level report.³²

Validating the 10x Hypothesis

A 10x increase in the total volume of system-executed queries is not just plausible; it is likely a conservative estimate. This multiplication happens on two levels. First, as Google has already observed with its AI Overviews, when users are given more powerful tools, they ask more complex questions more often, leading to a more than 10% increase in user-initiated queries for these features.³²

Second, and more importantly, each of these more complex user prompts triggers the "query fan-out" effect on the back-end. If a single user request can generate "hundreds" of machine-level searches, the system-wide multiplier will be immense. The average multiplier across all searches—simple and complex—will easily exceed 10x as agentic capabilities become the default.

This architectural shift fundamentally redefines what a "search" is. Historically, the economic model was simple: one human query resulted in one set of results and one ad opportunity, creating a clean 1:1 link between cost and revenue. Agentic AI shatters this link. Now, one human prompt triggers N machine queries. The cost to the search provider is now proportional to N, while the revenue opportunity remains tied to the single human interaction. The cost-per-human-interaction has been multiplied by N, creating an economically unsustainable model that must be re-platformed and, ultimately, re-priced.

The New Advertiser's Calculus: Modeling Future Financial Scenarios

The technical reality of exploding computational demand will inevitably translate into a new financial reality for advertisers. As search providers like Google face an exponential increase in their cost base, they will be forced to recoup these investments. This cost squeeze will occur just as the supply of traditional, clickable ad inventory is being compressed by "zero-click" AI answers. This collision of rising costs and shrinking supply will create a hyper-inflationary environment for search advertising and bifurcate the customer acquisition funnel into two distinct, and starkly different, economic scenarios.

The Coming Cost Squeeze and the Bifurcated Funnel

The pressure to monetize this new, more expensive form of search is immense. Internal Google documents reveal an urgent push to "accelerate monetizing Gemini with Ads ASAP" in anticipation of these structural shifts.¹⁰ At the same time, the number of traditional ad slots is under threat. One eMarketer report estimated that AI agents could cause a 38% drop in ad exposure during the discovery phase of the customer journey and a 47% drop during consideration, dramatically shrinking the available inventory for advertisers to bid on.¹² This dynamic—exploding costs for the provider and shrinking inventory for the advertiser—sets the stage for a radical repricing of the market. This will force advertisers to navigate a new, bifurcated world: paying a steep premium for customers acquired through legacy search, while simultaneously investing in new methods to attract customers through agentic channels.

Scenario A: The Economics of the "Legacy Search Customer"

This scenario models the future for an advertiser seeking to acquire a customer through the familiar, ad-supported search results page. Today, an example cost for a B2C Customer Acquisition Cost (CAC) on Google Ads can around $290.³⁶ In the near future, this cost is set to skyrocket, driven by three primary factors:

1. Amortized Energy Costs: The massive new electricity bill required to power AI data centers represents a new fixed cost that must be spread across all revenue-generating activities, including advertising.²¹

2. The Query Multiplier: The 10x (or greater) increase in system-level queries means the fundamental cost to serve a single human prompt has multiplied. The price of the ad slot associated with that prompt must rise to maintain margins.

3. Inventory Scarcity: With AI Overviews and answer engines resolving more queries directly, fewer users will click through to traditional search results pages. This reduces the number of available ad slots, forcing more advertisers to compete in auctions for a scarcer resource, inevitably driving up the price per click.

Combining these pressures, it is plausible to forecast a 2x to 5x increase in CAC for customers acquired through traditional search channels. The example $290 CAC could easily inflate to between $580 and $1,450.

Scenario B: The Economics of the "Agentic Customer"

This scenario presents a radically different acquisition model. Here, the customer does not see an ad. Instead, their personal AI agent interacts directly with a business's systems. This is the world of "agentic demand," a novel form of commerce where a customer is "fully qualified" and arrives with a "validated payment instrument and explicit purchase order".³⁸

In this model, the traditional marketing funnel of awareness, interest, and consideration is bypassed entirely.³⁹ The agentic transaction is theoretically a 100% conversion event.³⁸ For the advertiser, the economic implications are profound:

• Near-Zero Ad Spend: Because the customer was not acquired via a paid ad, the direct ad spend for that transaction is $0.

• CAC Approaches Zero: For that specific transaction, the Customer Acquisition Cost approaches zero.³⁸ The cost for the business is no longer a marketing expense but an infrastructure expense: the cost of building and maintaining systems that are "agent-compatible."

• The LTV Trade-off: The critical caveat is the potential impact on Customer Lifetime Value (LTV). An AI agent, operating on logic, may prioritize objective factors like price or features over brand loyalty in future purchases. A customer acquired via an agent may be less "sticky" and thus have a lower LTV, disrupting the traditional 3:1 LTV-to-CAC ratio that underpins sustainable growth.³⁸

The future for marketers is not a simple choice between these two scenarios. For the foreseeable future, they must operate in both worlds simultaneously. This creates a significant strategic and organizational challenge. The Chief Marketing Officer will need to justify paying a hyper-inflated $1,000 CAC for a customer from Google Ads, while simultaneously petitioning the CFO for a multi-million dollar engineering budget to build the capabilities for AI Agent Optimisation (AAO) to attract "$0 CAC" customers. This tension will force a redefinition of marketing itself, blurring the lines between the marketing budget and the IT budget, and elevating technical proficiency to a core marketing competency.

From SEO to AAO—Surviving the Cost Bomb

The economic model that has governed digital advertising for a generation is on the verge of a violent repricing. The shift to an AI-driven search paradigm is not an incremental evolution; it is a fundamental disruption with an inescapable causal logic. Understanding this logic and preparing for its consequences is now the most critical strategic challenge facing marketers.

The Causal Chain of the Cost Bomb

The argument for the coming cost explosion follows a clear and direct causal chain:

1. Agentic AI is the new paradigm for search. It moves beyond simple information retrieval to become an active, goal-oriented assistant.

2. Its core architecture relies on Query Decomposition, a process that breaks down a single complex human prompt into a multitude of system-level queries, leading to a 10x or greater explosion in total search volume.

3. This query explosion drives an unprecedented surge in computation and energy consumption, causing the operational costs of data centers to skyrocket.

4. This forces search engines to pass on these massive new infrastructure costs to the only available revenue source: advertisers.

5. The result is a hyper-inflationary environment for traditional search ads, fundamentally altering the economics of digital marketing and rendering old budget models obsolete.

The world of traditional search is about to up-end its unit economics. Something every advertiser and CMO now has to be hyper-aware of in the coming disruption.

Works cited

1. The History of Search Engines - Audits.com, accessed June 15, 2025, https://audits.com/seo/insights/history-of-search-engines/

2. The Complete History of Search Engines | SEO Mechanic, accessed June 15, 2025, https://www.seomechanic.com/complete-history-search-engines/

3. History and evolution of search engines - Telefónica, accessed June 15, 2025, https://www.telefonica.com/en/communication-room/blog/history-evolution-search-engines/

4. The Evolution of Search Engines: From Archie to AI-Powered Tools - Rellify, accessed June 15, 2025, https://rellify.com/blog/evolution-of-search-engines

5. The Evolution of Search Engines - Kenility, accessed June 15, 2025, https://www.kenility.com/blog/technology/evolution-search-engines

6. A Brief History of Search Engine Marketing (SEM) - Intero Digital, accessed June 15, 2025, https://www.interodigital.com/blog/a-brief-history-of-search-engine-marketing/

7. en.wikipedia.org, accessed June 15, 2025, https://en.wikipedia.org/wiki/Pay-per-click#:~:text=advertising%20payment%20system.-,History,web%20directory%20called%20Planet%20Oasis.

8. Pay-per-click - Wikipedia, accessed June 15, 2025, https://en.wikipedia.org/wiki/Pay-per-click

9. A Brief History of Pay-Per-Click | AGI Marketing Solutions, accessed June 15, 2025, https://agims.com/brief-history-ppc/

10. Google Search traffic decline is inevitable, execs say, accessed June 15, 2025, https://searchengineland.com/google-search-traffic-decline-inevitable-455345

11. Google search volume declines for first time in 22 years. Have AI ..., accessed June 15, 2025, https://www.fool.com.au/2025/05/15/google-search-volume-declines-for-first-time-in-22-years-have-ai-powered-tools-taken-over/

12. In Graphic Detail: How AI is changing search and advertising - Digiday, accessed June 15, 2025, https://digiday.com/marketing/in-graphic-detail-how-ai-is-changing-search-and-advertising/

13. End of Google Search? AI Reshaping Search, Ads, and SEO - Connected-uk.com, accessed June 15, 2025, https://www.connected-uk.com/googles-decline-in-search-usage/

14. Marketing's New Middleman: AI Agents | Bain & Company, accessed June 15, 2025, https://www.bain.com/insights/marketings-new-middleman-ai-agents/

15. SEO in the AI Era: Navigating AI Overviews, Zero-Click Searches, and Agentic AI, accessed June 15, 2025, https://cmimediagroup.com/resources/seo-in-the-ai-era-navigating-ai-overviews-zero-click-searches-and-agentic-ai/

16. Everyone Thought ChatGPT Used 10 Times More Energy Than ..., accessed June 15, 2025, https://www.zmescience.com/science/news-science/how-much-energy-chatgpt-query-uses/

17. How Much Energy Do Google Search and ChatGPT Use? - RW Digital, accessed June 15, 2025, https://www.rwdigital.ca/blog/how-much-energy-do-google-search-and-chatgpt-use/

18. AI uses more energy, water than a Google search without AI | wcnc.com, accessed June 15, 2025, https://www.wcnc.com/article/news/verify/ai/ai-electricity-water-use-chatgpt-google-la-fire-fact-check/536-85df815e-4212-4f2d-87bd-f5dd4e8f5eff

19. Search Engines vs AI: energy consumption compared - Kanoppi, accessed June 15, 2025, https://kanoppi.co/search-engines-vs-ai-energy-consumption-compared/

20. GenAI uses 30 times more energy than traditional search engine - ET CIO, accessed June 15, 2025, https://cio.economictimes.indiatimes.com/news/artificial-intelligence/genai-uses-30-times-more-energy-than-traditional-search-engine/113382102

21. AI is set to drive surging electricity demand from data centres while ..., accessed June 15, 2025, https://www.iea.org/news/ai-is-set-to-drive-surging-electricity-demand-from-data-centres-while-offering-the-potential-to-transform-how-the-energy-sector-works

22. AI has high data center energy costs — but there are solutions | MIT Sloan, accessed June 15, 2025, https://mitsloan.mit.edu/ideas-made-to-matter/ai-has-high-data-center-energy-costs-there-are-solutions

23. Power for data centers could come at 'staggering' cost to consumers - Floodlight News, accessed June 15, 2025, https://floodlightnews.org/power-for-data-centers-could-come-at-staggering-cost-to-consumers/

24. Understanding Agentic AI: The Next Frontier of Artificial Intelligence - Coveo, accessed June 15, 2025, https://www.coveo.com/blog/agentic-ai/

25. Agentic Search: Smarter AI for Government Information Use, accessed June 15, 2025, https://techsur.solutions/from-search-to-strategy-how-agentic-search-can-upgrade-information-discovery-in-government/

26. From search to AI agents: The future of digital experiences, accessed June 15, 2025, https://searchengineland.com/from-search-to-ai-agents-the-future-of-digital-experiences-451932

27. Why Enterprises Need AI Query Engines to Fuel Agentic AI - NVIDIA Blog, accessed June 15, 2025, https://blogs.nvidia.com/blog/ai-query-engines-agentic-ai/

28. Search engine - Wikipedia, accessed June 15, 2025, https://en.wikipedia.org/wiki/Search_engine

29. Query Decomposition - FlowHunt, accessed June 15, 2025, https://www.flowhunt.io/components/querydecomposition/

30. Advanced RAG: Query Decomposition and Reasoning - Haystack - Deepset, accessed June 15, 2025, https://haystack.deepset.ai/cookbook/query_decomposition

31. Advanced RAG: Query Decomposition & Reasoning - Haystack - Deepset, accessed June 15, 2025, https://haystack.deepset.ai/blog/query-decomposition

32. AI in Search: Going beyond information to intelligence - Google Blog, accessed June 15, 2025, https://blog.google/products/search/google-search-ai-mode-update/

33. Up to 40% better relevance for complex queries with new agentic ..., accessed June 15, 2025, https://techcommunity.microsoft.com/blog/azure-ai-services-blog/up-to-40-better-relevance-for-complex-queries-with-new-agentic-retrieval-engine/4413832

34. How we built our multi-agent research system \ Anthropic, accessed June 15, 2025, https://www.anthropic.com/engineering/built-multi-agent-research-system

35. Anthropic: How we built our multi-agent research system, accessed June 15, 2025, https://simonwillison.net/2025/Jun/14/multi-agent-research-system/

36. Customer Acquisition Cost Across Digital Marketing Channels - ClickGUARD, accessed June 15, 2025, https://www.clickguard.com/blog/customer-acquisition-cost-across-digital-marketing-channels/

37. Customer Acquisition Cost Across Digital Marketing Channels, accessed June 15, 2025, https://www.clickguard.com/blog/customer-acquisition-cost-across-digital-marketing-channels

38. Pricing Agentic: Economic Models for a New Kind of Demand ..., accessed June 15, 2025,

39. What is Customer Acquisition? Definition, Calculation & Examples | Amazon Ads, accessed June 15, 2025, https://advertising.amazon.com/library/guides/customer-acquisition