When is a "Commodity" Not a Commodity?

Real markets often do not conform to our efforts to categorize or change them. Sometimes products are hard to classify, in that regard. Consider home broadband and mobile phone services, which might be considered commodities by some; but the opposite by others. 

Any “commodity market” requires products that are standardized and therefore interchangeable across providers, with competition primarily driven by price, with  little differentiation in quality or features. Agricultural products, metals and energy products provide good examples. 

But it is a judgment call whether mobile services have such properties, as competitive as the market might be. 

Mobile phone service has some commodity traits (price sensitivity, partial standardization), but significant differentiation through network quality, branding, and bundled services, along with limited fungibility (one provider’s product is quite similar to all others), makes it a non-commodity.

Broadband is the most commodity-like (compared to either mobile phone service or electricity) due to high standardization, fungibility, and price-driven competition. However, bundling and infrastructure variations introduce slight quasi-commodity traits. 

Product	Standardization	Price Competition	Fungibility	Differentiation	Category
Electricity	High	High (deregulated); Low (regulated)	High (deregulated)	Low	Quasi-commodity
Mobile Phone Service	Moderate	Moderate	Low	High	Non-commodity
Home Broadband	High	High (competitive markets)	High (competitive markets)	Moderate	Commodity

Home Broadband might be the most commodity-like due to high standardization, fungibility, and price-driven competition in markets with multiple providers. Its differentiation through bundles (streaming services, mobility, linear video) is secondary to price and speed.

Electricity might be a quasi-commodity, highly standardized and fungible in deregulated markets but constrained by regulated monopolies and infrastructure dependence in many regions.

Mobile phone service might be considered a non-commodity, with significant differentiation through network quality, branding, and bundled perks, despite some price sensitivity.

Whether an industry’s products are commodities or not also shapes business and revenue models. Even products we might assume are definitely commodities, such as electricity, might actually be “quasi” commodities. 

In most regions, electricity, gas, and water are delivered by regulated monopolies, meaning consumers have no choice in providers. This lack of competition undermines the commodity trait of market-driven pricing. 

Unlike true commodities (oil traded globally), these utilities are tied to local infrastructure (grids, pipelines, water systems), which effectively means an electron for sale in Atlanta, Ga. is not interchangeable with an electron to be sold in Los Angeles, Calif. 

That might not seem so important, but it means that retail electricity for consumers is not an actual commodity (implying alternate suppliers), but more a “quasi-commodity” as there are, in many markets, no competitors. 

The point is, our efforts to categorize and understand ”commodity” products, and company and industry efforts to prevent products becoming commodities, is harder than it might seem.

Will AI Operations Dominate Data Center Networking by 2030?

CoreWeave appears to be the first data center operator to deploy Spectrum-XGS Ethernet, an Nvidia term for  “scale-across” capability linking remote data centers so they can operate as though all resources were locally present. 

The innovation is a third approach to AI computing  that complements existing “scale-up” (more powerful processors) and “scale-out” (more processors at the same location) strategies.

Spectrum-X Ethernet aims to provide:

• Distance-adaptive algorithms that automatically adjust network behavior based on the physical distance between facilities

• Advanced congestion control that prevents data bottlenecks during long-distance transmission

• Precision latency management to ensure predictable response times

• End-to-end telemetry for real-time network monitoring and optimisation

“Spectrum-XGS Ethernet nearly doubles the performance of the Nvidia Collective Communications Library,” the company says. 

At least in principle, Spectrum-XGS Ethernet could allow data center operators to build smaller facilities that do not strain local power grids. 

The dramatic rise in AI workloads is causing network traffic inside and between data centers to surge at rates much higher than historical averages, with projections of 30%+ annual growth for AI-related data traffic—versus the long-term rate of about 20–30% pre-AI. Surveys indicate that within 2-3 years, over half of data center operators expect AI workloads to dominate inter-facility bandwidth needs, outpacing even traditional cloud computing. This is prompting a shift towards scalable fiber-optic connectivity and advanced network fabrics, enabling fast, reliable data transfer at unprecedented scale.

Demand Growth Estimates

Analyst estimates highlight explosive growth:

• By 2030, global data center capacity is expected to increase 2.5-fold, and about 70% of data center capacity will need to support advanced AI workloads.

• Demand for AI-ready data center capacity is projected to rise at an average rate of 33% annually between 2023 and 2030.

• By 2025, AI workloads could make up nearly 30% of all data center traffic, with AI inference and training driving massive “east-west” data flows between facilities, rather than traditional “north-south” patterns.

• Operators are increasing investment in high-capacity, low-latency WAN and fiber interconnections, with single-mode fiber and optical transceivers—often at 800 Gbps or higher—becoming the norm for these links.

And such “AI data center to AI data center” connections are projected by some to dominate data center networking demand within five years. 

Projected Data Center Network Demand

Year	% AI Workload Traffic	AI-ready Capacity Growth	Main Driver
2025	30%	31.6% CAGR	AI inference/training
2030	70%	33% CAGR	Advanced AI workloads/interconnects

AI Might be a Threat that Computer-Generated Graphics Have Not Been

Netflix has guidelines for use of generative AI based on five main points:

• The outputs do not replicate or substantially recreate identifiable characteristics of unowned or copyrighted material, or infringe any copyright-protected works (respect for copyright)

• The generative tools used do not store, reuse or train on production data inputs or outputs (data security)

• Where possible, generative tools are used in an enterprise-secured environment to safeguard inputs 

• Generated material is temporary and not part of the final deliverables

• GenAI is not used to replace or generate new talent performances or union-covered work without consent.

The guidelines also caution against creating content that could be mistaken for real events, people, or statements. 

Of course, as a practical matter, all that will have to be monitored and verified. Perhaps the areas of greatest concern are final character designs and key visuals; talent replication and use of unowned training data. 

Proposed Use Case	Action	Rationale
Using GenAI for ideation only (moodboards, reference images)	✅	Low risk, non-final, likely not needing escalation if guiding principles are followed.
Using GenAI to generate background elements (e.g., signage, posters) that appear on camera		Use judgment: Incidental elements may be low risk, but if story-relevant, please escalate.
Using GenAI to create final character designs or key visuals		Requires escalation as it could impact legal rights, audience perception, or union roles.
Using GenAI for talent replication (re-ageing, or synthetic voices)		Requires escalation for consent and legal review.
Using unowned  training data (e.g., celebrity faces, copyrighted art)		Needs escalation due to copyright and other rights risk.
Using Netflix's proprietary material		Needs escalation for review if outside secure enterprise tools.

Some observers might liken the use of generative AI to the use of computer-generated graphics. It might be argued that CGI did not broadly automate creative work, as AI might threaten to do, in some cases. 

While CGI technology does automate certain repetitive or technical tasks, the work typically requires direct human input, creative intent, and iterative collaboration, some would argue. And while CGI shifted some jobs from traditional effects (such as practical props) to digital, it did not broadly automate creative work. 

AI, on the other hand, arguably can drastically reduce the need for human artists, writers, and designers, especially for routine or template-based tasks. A reasonable view held by creatives is that generative AI creates extensive automation threats to creative jobs, challenging the role, compensation, and rights of human creators in ways CGI never did. 

Issue	CGI	Generative AI
Labor Replacement	Redistributes labor, limited direct job loss	Automates substantial creative tasks, risks widespread job loss
Human Creativity	Essential for most tasks	Can fully automate or diminish creative input
New Job Creation	Created new specialist roles	Some new roles, but net job losses expected
Worker Rights/Ethics	Tied to work conditions, overtime	Issues of data exploitation, loss of control, IP and consent
Value Perception	Value linked to expertise and collaboration	Value eroded by commoditization, especially for freelancers
Legal Uncertainty	Relatively mature standards	Significant legal and ethical ambiguity

Do We "Need" 6G? Yes and No

By about 2030, standards bodies and suppliers will have gotten quite a ways down the road of preparing the next generation of mobile networks to succeed 5G. There will be claims about how “revolutionary” it might be, as we heard about 3G, 4G and 5G before. 

There will be requirements for additional spectrum, as always has been the case when a mobile next-generation network has launched. The issue is how much new spectrum might be required. 

And even if the general rule is that users consume more data, and therefore use more bandwidth, over time, there are some questions about the degree to which mobile operators will need to spend heavily on new spectrum, though governments who make money selling spectrum will prefer higher amounts and costlier prices. 

Mobile service providers obviously will want to limit their investment in new spectrum resources. Keep in mind that they have other avenues for doing so. They can create smaller cells; they can use more-efficient radios and network elements; better air interfaces and reclaim spectrum supporting older networks that are decommissioned (2G and 3G being the best examples at the moment). 

But offloading demand to fixed networks has become a huge tool as well. 

Wi-Fi handles 70 percent to 80 percent of total smartphone data consumption, with the exact figure varying by region. Wi-Fi data consumption in the United States is about  85 percent to 90 percent, for example, while lower in emerging markets (around 50 percent to 70 percent), according to estimates fromCisco, Ericsson, and OpenSignal, for example. 

Year	Wi-Fi Data Consumption (EB/month)	Mobile Network Data Consumption (EB/month)	Percentage on Wi-Fi
2024	383	164	70%
2025	460	197	70%

Beyond the Wi-Fi role, technologists and operators have gotten better at using older platforms to ease the transition to a next generation of networks, even if that means not all the touted features are available. Network slicing on 5G networks requires “standalone” platforms that in many cases are lightly deployed at the moment, for example. 

On the other hand, the faster speeds and higher bandwidth, plus lower latency of every next-generation network already is producing commercial revenue in significant amounts, such as using 5G platforms to support fixed wireless for home broadband. 

That might not be among the futuristic capabilities 5G was supposed to provide, but it has created new revenue and product possibilities. 

So perhaps some skepticism about the market “need” for 6G, and the resources needed to support it, are reasonable. Already, 6G is touted as supporting a new array of sensory information such as touch, taste and smell. 

Some of us would be that if such innovations actually arrive, it will be about the time 7G arrives, as that has been the pattern for past next-generation network innovations as well: the promised futuristic apps need twice as long to reach commercial success as predicted. 

So promised 3G innovations don’t arrive until 4G; 4G innovations don’t arrive until 5G. That isn’t to deny the practical advantages for each next-generation network: more capacity and lower latency. 

But those improvements are akin to the need fixed network operators have to upgrade copper access to optical fiber; satellite providers to upgrade from geostationary platforms to low-earth-orbit constellations, all of which support higher capacity networks. 

Mobile networks will need to continue to evolve to support higher speeds. The “revolutionary new applications and use cases”  might ultimately be less important.

Gemini has Gotten Much Better at Power and Water Consumption for Inference

 A median Gemini Apps text prompt generates 0.03 gCO2e and consumes 0.26 mL of water (or less, using a different methodology) researchers say in a study of energy and water consumption for artificial intelligence inference operations

source: Google

“To put this into context, a modern television consumes approximately 100 watts of electricity, so 0.24 Wh represents less energy than watching TV for nine  seconds, the researchers say. 

The water use of 0.26 mL equals five drops of water (based on a standard 0.05 mL drop).

source: Google