For Every Public Purpose There is a Corresponding Private Interest

For every public purpose there are corresponding private interests. OpenAI, for example, has its own interests in language models and the data center infrastructure necessary to support widespread use of such models.

Where it comes to public policy on data centers, that applies to state actors as well.

A report issued by OpenAI discusses two clusters of ChatGPT accounts likely originating from China

“that we banned after they used our models in support of apparent covert influence operations that promoted narratives in an attempt to manipulate a legitimate debate about American AI and wider tech policies,” the report says. 

“The first cluster generated social media comments and images claiming that data center buildouts for AI were increasing electricity prices for average families,” the report says. The issue is not the relevance of the debate, but the effort by state-linked actors to influence that debate, for perceived state interests.

“The operators of the accounts were likely part of a social media operations team at a private Chinese technology company conducting work for Chinese provincial level government clients,” OpenAI says. “This activity appears consistent with a commercial ecosystem that supports Party-state priorities in public opinion guidance.”

“A separate report they uploaded to ChatGPT described their objectives and strategies for influencing public opinion and establishing social media accounts designed to evade platform detection systems,” the report notes.

“They primarily targeted U.S. audiences and generated English-language short comments and images claiming that data centers and AI applications were increasing electricity demand and causing higher

costs for ordinary Americans,” OpenAI says. 

While the campaign does not appear to have gained much authentic engagement, “their significance lies in what they reveal about the intentions of influence operators from China and the narratives they are testing and seeking to amplify,” the report states.

First Amendment Free Speech Protections Apply Only to Government, at the Federal Level, Not Private Firms

Some might believe that U.S. free speech rights apply to most venues we encounter, from shopping malls to public transportation. That is mistaken. The right of free speech as enumerated in the First Amendment to the U.S. constitution only fully applies to government restraint, not that of all private actors or venues. 

So when a passenger on an airline complains about free speech “rights,” that is not a venue of protected speech.

Importantly, the First Amendment generally restricts government action, not private property owners, so it usually does not give a right to speak on private property or in privately run venues just because the public is invited in.

The Supreme Court’s starting point is that private owners do not have to turn their homes, businesses, or other property into expressive forums for others. In Marsh v. Alabama, the Court treated a company town like a municipality because it functioned as a town in all practical respects, so speech rights applied there. 

But later cases narrowed that idea, making clear that ordinary private commercial property is usually not subject to First Amendment speech access rights.

Consider shopping malls, which some litigants have compared to the older “town square.” Courts have ruled that the First Amendment does not force the owner to allow leafleting, protests, or petitioning on private mall property, as in Lloyd Corp. v. Tanner and Hudgens v. NLRB.

But some state constitutions grant greater speech rights in shopping centers than the federal First Amendment requires.

Social media companies are usually treated as private actors, so the First Amendment generally does not make them open public forums for user speech, either. 

Public transportation is trickier because the answer depends on ownership and operation. If the transit system is run by the government, First Amendment limits on viewpoint discrimination can apply because the government is involved. 

If the service or property is privately owned or operated, the First Amendment usually does not itself force access, although specific transit areas can sometimes be treated as public forums for certain government-run advertising or station spaces.

Churches are private property and are not generally subject to First Amendment speech-access claims from outsiders. 

The general rule is that the First Amendment applies directly when the government is restricting speech, but not when a private owner is doing so. 

Language Model Brand Preference is Unclear, Yet

Whether language model usage represents sustained brand preference is probably unclear. One study found satisfaction rankings for the top three platforms (Claude, ChatGPT, and DeepSeek) statistically indistinguishable. 

That study also found that users treat these tools as interchangeable utilities rather than sticky ecosystems. More than 80 percent of respondents use two or more platforms, and switching costs are negligible. 

Also, each platform attracts users for different reasons: ChatGPT for its interface, Claude for answer quality, DeepSeek through word-of-mouth, and Grok for its content policy, the researchers say. 

So usage patterns, though clear enough, should not necessarily equate to sustainable brand preference, for the moment. 

In consumer markets, ChatGPT continues to hold the lead, but Gemini has climbed significantly. 

source: Momentic Marketing 

In the enterprise user market, Anthropic’s Claude leads, followed by OpenAI and Gemini. 

source: Momentic Marketing 

Still, most enterprises aren’t betting on a single model provider, as some 81 percent now use three or more model families in testing or production, according to Andreessen Horowitz. 

Model leadership also varies by use cases. Anthropic is notable for enterprise software development use cases, for example. ChatGPT gets wide use for general-purpose chatbot use cases. 

a16z.com

Trust, compliance, and "authority" positioning matter more to business buyers than to consumers. To the extent we can say brand positioning exists, Anthropic seems to be the brand for enterprises seeking “reliability.” 

In consumer markets, perhaps Gemini's growth is tied heavily to being pre-installed rather than actively chosen by users, and OpenAI's own remaining enterprise strength is partly inertia.

But the basis for brand preference seems clear enough in enterprise markets. Only about 11 percent of enterprise teams reported switching vendors in the past year. Most (66 percent) upgraded the same vendor's model, so Anthropic's enterprise gains represent real vendor displacement, not just organic growth of new spend.

At least at this point, though, language model brand preference durability seems less pronounced than was the case for earlier computing innovations:

• Operating systems and browsers historically showed durable lock-in from network effects, file-format compatibility, and default-setting. Once Windows or Chrome won a segment, share moved in single-digit points per year for a decade.

• Enterprise infrastructure (databases, cloud, ERP) traditionally locks in through multi-year contracts, data gravity, and integration cost, producing switching costs high enough that even a technically superior challenger struggles for years. LLM enterprise switching costs are comparatively low. 

• Multi-homing is common, as  low switching costs and task-specific model preferences seem to exist.

So language model brand preference today functions less like durable ecosystem lock-in (OS, browsers, smartphones) and more like a fluid, feature- and task-driven preference market, except perhaps in enterprises, where trust, compliance positioning, and demonstrated coding performance have already produced one real reversal of market leadership in under three years.

BlackBerry and Consumer Technology Transitions

The demise of Research in Motion and its BlackBerry smartphone, which once was the premier enterprise device, can be blamed on several failures:

• Underestimating the iPhone and touchscreens

• Missing the app platform and app store transition

• Evolution of the smartphone use case, from single-purpose to multi-purpose

• Delayed operating system upgrades. 

For me, BlackBerry lost because it stuck with “email on a phone” when the market was shifting to “internet on a phone.” 

As I recall, it was specifically turn-by-turn directions which was the “killer app” that displaced email functions as the key feature. 

The elimination of the standalone GPS device is perhaps one of the more underappreciated reasons why consumers abandoned devices such as the BlackBerry in favor of touchscreen smartphones. 

Although the iPhone's multi-touch interface, app ecosystem, and media capabilities receive most of the attention, navigation was one of the first "killer utility" applications. 

The BlackBerry dominated secure mobile email but was not designed around rich graphics, mapping, or location services. 

As I recall, the turning point came in 2009, when Google introduced free turn-by-turn navigation for Android devices. 

And though it was for me a relatively minor change, the ability to use the smartphone in place of a dedicated music device also was an advantage. Others might have found replacement of a camera or laptop more compelling. 

As I also recall, turn-by-turn navigation also was the trigger for upgrading early to 4G, even before the standards issue was decided. 

Back then, there was an ecosystem debate between Long Term Evolution and WiMax. 

LTE, backed by 3GPP and the big cellular carriers faced WiMax, backed by the IEEE/802.16 world and Sprint. 

As I recall, Sprint’s version had a half-year deployment lead over LTE, so Sprint got the immediate benefit of the account. We all switched to LTE when the standards debate was settled.

AI ROI: Efficiency First, Effectiveness Longer Term

The distinction between efficiency and effectiveness is a foundational dichotomy in organizational life, and adoption of artificial intelligence will exhibit both tendencies, as most prior computing technologies have tended to do. 

As with earlier waves of information technology adoption, organizations initially justified investments in computers, enterprise software, and the internet through labor savings and operational efficiencies.

Over time, however, the greatest economic gains came from effectiveness:

• creating entirely new products and services (such as online banking, e-commerce, cloud software, and digital media)

• entering new geographic markets

• serving customers in more personalized ways

• accelerating research and product development

• enabling new business models that were previously impractical.

AI appears to be following the same trajectory. 

While efficiency is often the primary driver for early-stage AI adoption: 

• automating manual tasks

• accelerating workflows

• reducing overhead.

But most might agree that greater long-term outcomes are possible by focusing on AI effectiveness, or "doing the right things” (perhaps different things)  with AI. 

Dimension	Operating More Efficiently	Operating More Effectively
Goal	Reduce costs	Increase value created
Focus	Existing work	New opportunities
Typical metric	Cost, hours, headcount, error rate	Revenue, market share, innovation, customer satisfaction
Economic effect	Improve margins	Increase growth
Example	Automate invoice processing	Launch products previously impossible to build

Efficiency normally gets the early attention, as AI excels at removing human friction from repetitive tasks such as data entry, scheduling, routine coding, or basic customer service. 

Business Function	AI Application	Efficiency Gain	Example
Customer service	Chatbots and AI agents	Lower support costs, faster response	Klarna automates a large share of customer inquiries
Software development	Code generation	Faster coding and testing	GitHub Copilot, Microsoft
Manufacturing	Predictive maintenance	Less downtime	Siemens, GE
Finance	Invoice processing	Reduced manual accounting	SAP, Oracle AI
HR	Resume screening	Faster hiring workflows	Workday AI
Legal	Contract review	Reduced attorney review time	Harvey AI
Marketing	Content generation	Lower content production costs	Adobe Firefly
Supply chain	Demand forecasting	Lower inventory costs	PepsiCo, Walmart
Healthcare	Medical documentation	Reduced physician administrative burden	Microsoft Dragon Copilot
IT operations	AI help desks	Faster ticket resolution	ServiceNow

Executing established processes faster, cheaper and with fewer errors improves the throughput of a business. It can help lower operating costs. But does not necessarily change the nature of the business.

Effectiveness is about strategy and outcome; the pursuit of higher-value results. 

As organizations mature in their AI journey, the focus often shifts from "doing more with less" (efficiency) to "doing more of what matters" (effectiveness). 

Category	Definition	AI Implementation Example	Key Benefit
Efficiency (Doing things right)	Focuses on optimizing processes, speed, cost-reduction, and resource utilization.	Automating invoice processing; using predictive maintenance to reduce machine downtime.	Operational excellence, cost savings, time-to-market speed.
Effectiveness Doing the right things)	Focuses on achieving high-value outcomes, strategic goals, and solving the correct problems.	Using generative AI for creative brainstorming; AI-driven drug discovery; personalized patient diagnostics.	Competitive advantage, innovation, improved quality of output, solving complex challenges.

Long-term AI gains are likely to come from greater organizational effectiveness, as that has been the pattern for earlier computing innovations as well.

Business Function	AI Application	Effectiveness Gain	Business Outcome
Sales	Personalized prospecting	More customers reached	Higher revenue
Marketing	Hyper-personalization	Better conversion	Larger customer acquisition
Product development	AI-assisted design	More products launched	Faster innovation
R&D	Scientific discovery	More candidate molecules/materials	New products
Customer success	Predictive recommendations	Higher retention	Greater customer lifetime value
Consulting	Faster research	More client engagements	Revenue growth
Software	AI-native applications	Entirely new product categories	New businesses
Manufacturing	AI-assisted engineering	Better product quality	Competitive advantage
Education	Personalized tutoring	Better learning outcomes	New educational services
Financial services	AI investment analysis	Better decision quality	Higher returns

Analysts at McKinsey tend to find such returns are possible when enterprises implement AI, though both efficiency gains as well as effectiveness advantages can be seen. 

Finding	Interpretation	Source
Companies report both cost savings and revenue increases from AI	AI affects both efficiency and effectiveness	McKinsey Global AI Survey (McKinsey & Company)
Revenue gains are most common in marketing, sales, and product development	AI enables growth, not just automation	McKinsey (McKinsey & Company)
Enterprise-wide AI adoption correlates with substantially higher profit margins and returns on invested capital	Value comes from integrating AI into business strategy	McKinsey Operations Survey 2026 (McKinsey & Company)
AI-supported sales tools increased incoming orders by roughly 40% in one automotive transformation	AI improves commercial effectiveness	McKinsey Operations Survey (McKinsey & Company)
Software engineering productivity improved by up to 44% while enabling faster product delivery	AI boosts both efficiency and innovation	McKinsey Operations Survey (McKinsey & Company)