Friday, October 31, 2025

How Much Real Value Do Consumers Ascribe to "Nutrition Labels" of Any Sort?

Forgive me if I cannot get too exercised about the existence and usefulness of Federal Communications Commission broadband nutrition labels, one way or the other, as there is an argument to be made that they neither help nor hinder consumers very much. 


The reason is drop-dead simple: consumers aren’t dumb. They routinely use “lowest price” searches, so any supplier that bundles the add-ons risks being beaten by another provider that doesn’t do so. One might argue that so long as every supplier has to play by the same rules, there is no advantage to be gained or lost. 


But there is a supplier cost: time and money spent “explaining” all the component costs to every customer, on every interaction involving a potential purchase. 


In competitive markets, the widespread use of comparison apps and "find lowest price" features naturally pushes providers to list the lowest possible base price and separate out other costs (taxes, fees, equipment rentals). 


In a market where products or services are perceived as largely similar (like basic internet speed or a standard movie subscription), price becomes the primary differentiator. Providers, seeing their competitors list a low base price, are compelled to match or beat it to remain visible in the comparison results, leading to a competitive "race to the bottom" on the headline figure.


By segmenting the price into a low "base" and later "fees," providers make their initial offering appear more transparent or competitive than a competitor who lists a higher, all-inclusive price. The competitor with the truly all-inclusive price is penalized by being ranked lower in the comparison search results.


The “nutrition label” is supposed to help, in that regard, but one suspects few consumers are surprised by the fact that the advertised base price is not the total “all in” cost. 


Industry/Product

Advertised Base Price

Common Separated Fees/Costs (The "Drip")

Internet/Cable

Monthly Service Rate ($49.99/mo.)

Regulatory Recovery Fee, Broadcast TV Fee, Regional Sports Surcharge, Equipment Rental Fee (Modem/Router), Installation/Activation Fee.

Mobile Phone Plan

Monthly Data/Talk Cost ($30/mo.)

Regulatory Fees, E911 Service Fee, Universal Service Fund Fee, State/Local Taxes, Device/Installment Plan Payments.

Streaming TV/Content

Standard Monthly Subscription ($15.99/mo.)

Sales Tax (added at checkout), Platform Fees (if purchased via a third-party app store), Premium/Ad-Free Tier Upgrade Costs, Pay-Per-View/Add-on Movie Charges.

Event Tickets

Ticket Price ($50.00)

"Convenience Fee," "Service Charge," "Processing Fee," "Facility Fee," Delivery/E-Ticket Fee.

Software Subscriptions

Standard Plan per User ($10/user/mo.)

Setup/Onboarding Fee, Data Storage Overages, Premium Support Access, Migration Fees, Mandatory Annual Contract Surcharge.


AI "Is Not a Dot-Com Bubble," Federal Reserve Chair Jerome Powell Says

Federal Reserve Chair Jerome Powell doesn’t think the AI boom is another dotcom bubble. Speaking at a meeting of the Federal Reserve Open Market Committee, he argued that the current wave of artificial intelligence investment is grounded in profit-making firms and real economic activity rather than speculative excess. 


“These companies..actually have business models and profits and that kind of thing,” he said. “So it’s really a different thing” from the dotcom bubble, where investment capital flowed to firms with no clear business models. 


Also, one might note, the AI investments we currently see are not especially large, as a percentage of gross domestic product, compared to investment levels of prior general-purpose technologies such as railroads

Technology/Wave

Estimated Peak Investment as % of GDP

Time Period

Railroads

≈6%

Late 19th Century (e.g., 1870s boom)

IT/Dot-com Boom

≈4.6%

Q4 2000 (Peak for private domestic investment in information processing equipment and software)

AI (Current Estimate)

≈1.3%−1.5%

Estimated 2025 (Focus on AI data center capital expenditures)

Telecom

≈1%

≈2000−2001 or 2020 Peak (Specific boom estimates vary)


Also, AI investment levels are far below levels of dot-com era investment. 


The dot-com era investments happened across a wide range of information processing equipment and software categories, while AI investment is much more narrowly focused on high-performance chips and servers, as well as the data centers which house them. 


That doesn’t mean overinvestment is no risk at all. Markets tend to overshoot before correcting, so some “waste” (overinvestment that does not produce a positive outcome)  will happen. 


That might well be a different matter than a general “investment bubble in AI,” though.


Thursday, October 30, 2025

Can OpenAI Exceed All Prior Enterprise Software Market Shares?

If OpenAi does manage, by about 2030 or 2031, to earn $100 billion or so in annual revenues, it might at that point be supplying up to a third of all enterprise software revenues, if one attributes its platform and ecosystem revenues to the "enterprise" category.


Some will argue that makes little sense, since ChatGPT right now is consumer oriented. But the analogy is the Microsoft platform, where even consumer revenues are underpinned by enterprise use cases (software licenses are sold to device suppliers, not directly to consumers, even when consumers buy the appliances).


And if OpenAI does reach such lofty (for now) sales figures, it will be on the back of total platform revenues, not just chatbot revenues.


Enterprise IT Segment

2024 Annual Revenue (USD)

Description

IT Services

$1.5 trillion

Includes outsourcing, consulting, operations & maintenance, cloud migration, and cybersecurity .

Enterprise Software

$280–320 billion

Covers CRM, ERP, business intelligence, content management, and supply chain systems .

IT Hardware

~$140 billion

Includes server, storage, and data center equipment .


That would be an unprecedented level of market share, as the global leaders in enterprise software tend to have mid-single-digit market shares. 


Segment

Leading Vendors

Market Share Estimates (2024/2025)

Enterprise Software (total)

Microsoft, SAP, Oracle, Salesforce, IBM

Each holds low to mid-single-digit percent market share globally. Microsoft and SAP are typically top 2 or 3 by revenue .

ERP (Enterprise Resource Planning)

Oracle, SAP, Microsoft

Oracle (6.5%), SAP (6.5%), Top 10 vendors jointly 26.5% market share .

CRM (Customer Relationship Management)

Salesforce, Microsoft, Oracle, SAP, Adobe

Salesforce (26.1%), Microsoft (5.9%), Oracle (4.4%), SAP (3.5%), Adobe (5–6%) .

Cloud Infrastructure Services

Amazon AWS, Microsoft Azure, Google Cloud

AWS (30%), Azure (20%), Google Cloud (12–13%), together >60% global share .

IT Services

Accenture, IBM, Tata Consultancy, Infosys, Cognizant

Each between 2–5% of global services revenue, with significant fragmentation .


OpenAI IPO in 2026?

Reuters reports that OpenAI is considering an initial public offering with a firm valuation of $1trillion as early as 2026. That might make it the largest IPO ever. 


A $1 trillion valuation today implies that investors believe OpenAI is on an accelerated track to be a $50-100 billion revenue company in the near term, with a defensible technology that will allow it to capture a dominant share of the multi-trillion dollar global AI market.

OpenAI's current projected revenue is around $12 billion to $15 billion (annualized for 2025). To reach a $1 trillion valuation at its current revenue, it would need a 66x to 83x revenue multiple.


That implies annual revenue somewhere between$50 billion to $125 billion. 


That also implies that OpenAI maintains triple-digit or very-high double-digit growth rates for five to seven years. 


It also might imply that OpenAI is not just conquering a new software category, but is capturing a significant portion of global information technology spending, including developer tools, cloud services, and enterprise software across multiple industries.


That might seem unrealistic, but would fit the “winner take most” or “winner take all” character of recent internet markets. 


Such a valuation might also imply difficulty for traditional suppliers of enterprise software, as ChatGPT increasingly becomes a functional substitute for many traditional functions and apps. 


Of course, rivals including Anthropic, Google DeepMind, and Meta also are building platforms, distribution networks and developer ecosystems designed to secure long-term dominance in the field. 


You can make your own guess as to the likelihood of OpenAI achieving its goal of supporting its $1 trillion valuation with revenues. 


Wednesday, October 29, 2025

Why Bitcoin Miners are Pivoting to High-Performance Computing

There is a good reason why many bitcoin mining companies are pivoting to high-performance computing: the revenue per kWh is significantly higher, while doing so also smooths out income performance because there is a shift to recurring service revenue and away from the commodities nature of bitcoin valuation.


AI workloads, particularly for training large language models, command a premium price. Leasing out a megawatt of infrastructure capacity to a creditworthy AI customer can generate revenue significantly higher than using that same MW to mine Bitcoin.


Metric

Bitcoin Mining (Post-Halving)

AI / HPC Hosting (H100/A100 Pods)

Revenue per kWh Equivalent

∼$0.07 to $0.09

∼$0.25 to $0.35

Annual Revenue per MW

∼$613,000 to $788,000

∼$2.2 million to $$3.1 \text{ million}

EBITDA Margin

∼55% to 65%

∼70% to 80%

Revenue Source

Volatile block rewards (Bitcoin price-dependent)

Predictable, multi-year contracts with creditworthy clients


There also are equity valuation implications. Bitcoin Miners typically trade at a valuation multiple of 6x to 12x EV/EBITDA. Leading data center operators (such as Equinix or Digital Realty) trade at multiples of 20x to 25x EV/EBITDA.


Yes, Follow the Data. Even if it Does Not Fit Your Agenda

When people argue we need to “follow the science” that should be true in all cases, not only in cases where the data fits one’s political pr...