Sunday, July 20, 2025

Antitrust Law is Becoming More Subjective: Google Search is a Case in Point

Like everybody else who follows Alphabet, browsers and search, I have trouble figuring out what the judicial system will force Alphabet to do about its “monopoly” in the search business. Among the possible remedies is a forced divestiture of Google Chrome (the browser), among other possibilities. 


That remedy might seem odd in view of the defined problem being a monopoly in “search,” not browser market share. In fact, regulators might have a hard time making the case that “browser monopoly market share” actually matters, in a strict sense, since browser use is “free to use.” 


And, of course, traditional antitrust only applies to products that have an actual price, though the line began to blur when the U.S. government took action against the bundling of the Microsoft operating system with Internet Explorer. 


Even if the browser was essentially a feature, not a “product,” the monopoly danger was seen to exist in the combination of the operating system with the browser. 


So some argue that the remedy in the Google search case has to involve some remedy involving “something else,” as use of search also is “without charge” to the user. 


So the “remedies” now center around “doing something” to a product, or products, that are offered to consumers without charge. In other words, we now are dealing with allegations of “consumer harm” where the “harm” cannot be quantified, as the products are offered free of consumer charge. 


Traditionally, antitrust law has focused on “consumer harm” in the form of  higher prices, reduced output, or diminished quality, though the latter two criteria are subjective and cannot be measured. 


So we wind up with ideas that are basically subjective. The alleged monopoly in search is said to reduce innovation on Google’s part. Or, perhaps, Google search could suffer from quality issues, as it might have little incentive to innovate. 


If consumers "pay" for free services with their data and attention to ads, perhaps that is among the effects of the “monopoly.”


Others might argue that commercial agreements, such as Google paying Apple to be the default search engine on Safari, harms the competitive process.


Others might say the harm falls on advertisers who have to pay higher prices, and therefore incorporate those costs into consumer prices. 


All of those elements might seem logical, but virtually all are subjective or indirect metrics. As always, one has to define “the market” to make a claim of monopolization. 


What are the boundaries of  a market for a "free" product? Is Google Search competing with other search engines, or with social media platforms or apps that offer direct access to information or something else. It is not clear. 


The point is that the Google search monopoly allegation pushes the boundaries of several antitrust concepts. 


Digital platforms often operate in "two-sided" or "multi-sided" markets, connecting different groups (users and advertisers, drivers and riders). The complexity lies in how harm on one side (higher ad prices) impacts the other ("free" users) and how market power is assessed across these interconnected groups. That is quite a bit more subjective than the traditional test of higher consumer prices. 


Also, the arguments against Google often center on its role as a "gatekeeper" to the internet and how it "forecloses" opportunities for competitors through its widespread defaults and exclusive agreements. This avenue does not rely on consumer harm, but rather supposed harm to other competitors. 


Rather than higher consumer prices, which cannot be demonstrated directly, the court focuses on more subjective issues, such as whether dominant firms stifle innovation in the market. 


So we wind up in the novel realm of remedies for problems where consumer harm really cannot be demonstrated in any direct form. So even the remedies involve behavioral or structural changes for products with no actual consumer price. 


The belief is that, although Chrome generates no direct revenue, it is an important distribution channel for other Google services: search, Gmail, YouTube, and Google Drive. If a structural remedy is sought (forcing Google to divest the Chrome browser, for example), it would seem to involve something other than the actual “problem” of search monopoly. 


Divesting Chrome would presumably disrupt this seamless integration, potentially reducing traffic to Google’s other services.


Element

Chrome’s Role

Value to Google Search

Default Search Engine

Chrome sets Google Search as the default engine for most users.

Drives billions of queries and reinforces Google’s search market share.

Data Collection

Chrome tracks user behavior (URLs, clicks, performance, device info).

Enhances ad targeting and search ranking precision.

Search Integration

Omnibox merges the URL bar and search bar.

Increases frequency of search queries, even when users intend direct access.

Cross-product Reinforcement

Chrome integrates well with Google services (Gmail, Docs, YouTube, etc.).

Keeps users in the Google ecosystem, raising switching costs.

User Scale

World's most-used browser (65%+ global share).

Provides a massive funnel of user traffic into Google Search.


Without control over Chrome, Alphabet could lose the ability to set Google Search as the default engine or promote its AI products like Gemini.


Chrome is estimated to contribute significantly to Alphabet’s advertising revenue by driving 35 percent of Google’s search revenue, by driving search activity, enabling data harvesting, ad targeting and therefore ad sales. 


The issue is whether the divestiture of Chrome necessarily destroys that value chain. Commercial agreements could be struck allowing Alphabet access to the data that allows the rest of the stack and ecosystem to function. 


Indeed, perhaps no new owner would be in position to create as robust a revenue model as Alphabet is able to manage. 


Many of us use Google search because we believe it is the best engine. It seems unclear how many of us would switch away from Google no matter what happens to Chrome the browser. 


It would be logical to expect a hit to Alphabet’s equity value if such a development occurs. But some might well argue some of that already is baked into Alphabet’s valuation. There is a reason Alphabet has the lowest price-earnings ratio of the “Magnificent Seven” stocks, for example. 


In the end, we might not know how much impact a Chrome browser divestiture might have on Alphabet. Some might point to pressure on search advertising, which hasn’t been seen to date, but could well happen if large language models disrupt search revenue models. 


The impact on other products in the ecosystem might likewise be hard to pinpoint. Lots of us might already conclude that Gmail, YouTube and other products in the Alphabet ecosystem are preferred and would still be preferred, whether Alphabet owns the Chrome browser or not. 


Some of us would guess that some immediate equity market impact would happen, but that hit would be erased over time. Alphabet obviously has contingency plans and would obviously innovate in other areas. 


Beyond all that, the principles of antitrust for products that “have no price” are being tested. And the tests are increasingly subjective.


Thursday, July 17, 2025

ChatGPT Agent Launches

ChatGPT launches its new agent.

Future AI Energy Costs are Hard to Predict

Much has been made of a Goldman Sachs analysis of the impact of artificial intelligence on data center power costs, using the oft-quoted claim that AI queries require 10 times the electrical power of a search query.


The claim that AI queries require approximately 10 times the electrical power of traditional search queries has been widely cited, primarily based on estimates from the International Energy Agency (IEA) and other sources suggesting a single ChatGPT query consumes about 2.9 watt-hours (Wh) compared to 0.3 Wh for a Google search. 


But model efficiency, hardware improvements, workload variability, and the specific nature of queries can significantly affect energy use, suggesting the 10x claim is not universally the case.


Hardware and software efficiency is improving. For example, AI-related computer chips efficiency roughly doubles every 2.5 to 3 years. 


Software optimizations, such as those developed by MIT (the Clover tool), can reduce carbon intensity by 80 percent to 90 percent by adjusting workloads to off-peak times or using lower-quality models for less critical tasks.


Also, AI queries vary widely in computational intensity, with some tasks (text generation) being less energy-intensive than others (video generation).


So do search queries. Complex searches involving real-time data or multimedia can approach the energy use of simpler AI queries. And some estimates suggest that even if all Google searches were replaced with large language model (LLM)-powered searches, the global energy increase would be modest (an additional 10–29 TWh annually, compared to 460 TWh for all data centers in 2022). 


Some studies suggest AI energy demands actually are minimal. 


Table of Studies Challenging the 10x Claim

Study/Source

Publication Date

Key Findings

How It Challenges the 10x Claim

IEA: Energy and AI

April 9, 2025

Data centers account for a small share of global electricity demand growth; efficiency of AI chips has doubled every 2.5–3 years, reducing per-query energy use.

Suggests that efficiency gains and task variability (e.g., text vs. video) mean not all AI queries are 10x more energy-intensive.

MIT Sloan: AI Data Center Energy Costs

January 6, 2025

Software tools like Clover reduce carbon intensity by 80–90%; simple steps can cut 10–20% of data center energy demand.

Demonstrates that optimized AI workloads can significantly lower energy use, narrowing the gap with search queries.

University of Wisconsin: The Hidden Cost of AI

August 20, 2024

Microsoft improved chatbot servers to use 10x less energy; efficient generalization techniques reduce energy needs.

Shows that specific AI implementations can have much lower energy use, challenging the blanket 10x estimate.

Sustainability by Numbers: Impact of AI on Energy Demand

November 17, 2024

If all Google searches used LLMs, energy demand would increase by 10–29 TWh, a modest fraction of total data center use.

Indicates that per-query energy differences are less dramatic when scaled, due to efficiency and workload factors.

Harding & Moreno-Cruz: Watts and Bots

2024

AI’s energy demand increase, including economic spillovers, is very small.

Suggests that AI’s per-query energy impact is not as significant as claimed, due to broader efficiency trends.

Masanet et al.: Recalibrating Data Center Energy Use

2020

Data center energy use is lower than projected due to efficiency improvements.

Implies that AI’s contribution to energy demand is moderated, reducing the per-query energy gap.

Nature: AI Energy Demands

March 4, 2025

Lack of transparency leads to simplistic extrapolations; calls for more precise data on AI energy use.

Questions the reliability of the 10x claim due to insufficient granular data.

Breakthrough Institute: Unmasking AI Energy Demand

July 9, 2024

Energy intensity per computation has decreased 20% annually since 2010; AI’s energy impact may be overstated.

Argues that efficiency gains significantly reduce AI’s per-query energy use, challenging the 10x figure.


The point is that the energy consumption of AI queries depends heavily on the model, hardware, and task complexity, and the same applies to search queries, which can vary in intensity.


On-Device AI Use Cases Likely Greatest When Internet Access is Unavailable or Unreliable

AI apps that run directly on smartphones, such as Google’s new AI Edge Gallery, are said to unlock a range of use cases that include operating without an internet connection present. That has value for remote locations, while on airplanes or other locations where access is limited or non-existent. 


The other set of values often is said to include greater privacy or security, processing speed or reliability. Processing on a smartphone is said to provide greater privacy since no data is sent away to a remote site. In principle, that means less risk of a theft of data in transit. 


And some operations, such as those related to image processing on a local camera, are likely to work with less latency if the app is locally resident on the device. 


That might also apply for any AI operations that only use locally-resident (on the device) data. My own personal use cases almost always require access to external data, so the locally-resident value is quite limited. 


Each of us will decide how much value is provided by a locally-resident ability to generate email or message responses. 


On the other hand, some might find useful the ability to generate images locally, such as creating images based on prompts when offline.


Photo analysis using the camera might be another value, assuming no external data is required. 


But, most of the time, the value might come from processing that can be done without an internet connection. The other area where AI might have value is to optimize the performance of the device itself, to save battery life, for example. 


For many use cases, though, it seems as though the value is greatest when all required data is available locally and devices operate in areas with no internet access.


AI Benefits Will Often Show Up as "Externalities" (What is Enabled)

Every important new technology, and especially all general-purpose technologies, have benefits and “costs” (externalities), though most would likely assume that, on balance, every GPT brings more benefits than costs. 


Some of us assume that will be true for artificial intelligence as it has been the case for earlier GPTs. 


Quantified Benefits and Externalities of Electricity Use

Category

Benefit / Cost

Estimated Value / Impact

Economic Output Enabled

Benefit

>$20 trillion globally (approx. 25%+ of global GDP tied to electricity-enabled sectors)

Job Creation

Benefit

~25 million direct and indirect jobs globally (generation, transmission, electric equipment)

Household Welfare

Benefit

>1 billion people lifted out of extreme poverty since electrification

Time Saved (Lighting & Appliances)

Benefit

~500 billion hours annually saved globally (valued at >$5 trillion/year)

Health Improvements (e.g., refrigeration, medical equipment)

Benefit

Millions of lives improved or saved (e.g., vaccines, surgery enabled by electricity)

CO₂ Emissions from Electricity

Cost

~13.5 billion metric tons/year globally (40% of total CO₂ emissions); $1.35–$6.75 trillion/year (at $100–$500/ton social cost)

Air Pollution from Fossil Power

Cost

~3–4 million premature deaths/year; ~$2–$4 trillion in health damages (WHO, EPA estimates)

Infrastructure Costs

Cost

~$2 trillion/year (generation, transmission, distribution, maintenance)

Unequal Access (Energy Poverty)

Cost (opportunity lost)

~760 million people without access (loss of productivity, education, healthcare)

Blackouts and Reliability Issues

Cost

~$150–$300 billion/year in economic losses globally

Wednesday, July 16, 2025

AI is a Case of "More" Everything: Investment, Usage, Consolidation

Just a few graphics that confirm what you probably already believe is happening, namely that humans using artificial intelligence in relatively direct ways (as compared to the indirect ways as when AI aids some other process) keeps growing. 


source: Seeking Alpha, edge-ai-vision 


For most consumers, the most-common direct use is the AI chatbot. 

source: Business Insider, Seeking Alpha


As always is to be expected, there will be many more startups than are sustainable long term, and bigger apps will tend to acquire smaller apps, leading to some consolidation. 


source: IEEE Spectrum


And, compared to internet apps, which in many cases were relatively affordable to create, AI models are prodigiously expensive. And that will ultimately favor deep-pocketed firms with access to lots of capital. 


 source: IEEE Spectrum


And, generally speaking, the more capable language models become, the more money it takes to train them. As we move towards agentic AI, more of the cost should shift to inference and “action on inference” operations, though. 

 

source: IEEE Spectrum


On the other hand, the cost of using AI to make inferences keeps dropping, which means it will be used more often, further fueling usage. 

source: IEEE Spectrum



Antitrust Law is Becoming More Subjective: Google Search is a Case in Point

Like everybody else who follows Alphabet, browsers and search, I have trouble figuring out what the judicial system will force Alphabet to d...