By Keshavan Thiagaraja
A new name can attract attention. Enterprise value still depends on what a system can reliably deliver. For CEOs and CTOs, the discussion about “superintelligence factories” raises practical questions about investment, operating costs and accountability.
What changed
On 29 September 2026, President Trump signed Executive Order 14434, directing the executive branch to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI”, to the maximum extent permitted by law. The order retains the existing statutory AI definition for implementation and does not require rewriting previously issued regulations, contracts or historical documents. [1]
Jensen Huang has also described the facilities being built for advanced AI as “superintelligence factories”. Public reporting places those remarks at an America.gov panel. The available evidence supports the phrase, but I would not treat the claimed “just hours after” sequence as independently established. [2]
The policy language and the industrial metaphor deserve separate examination. A government naming decision does not demonstrate that a system exceeds human intelligence across a broad range of tasks.
What the factory metaphor explains
NVIDIA describes AI factories as integrated systems connecting energy, chips, infrastructure, models and applications. Their work includes model training and inference, which means running trained models to produce outputs. [3]
GPUs accelerate computation. Networks move data between processors. Electricity and cooling support sustained operation. Software coordinates workloads. These components must work together before an enterprise can use the output effectively.
For language models, tokens are units of text processed or generated by the model. Producing more tokens measures output volume. It does not, by itself, measure accuracy, understanding or customer value.
The benefits
The factory metaphor makes the physical foundations of AI easier to understand. It encourages leaders to consider capacity, reliability, maintenance and energy alongside software. That is useful when moving from an experimental tool to a service people depend on.
Shared infrastructure can support several enterprise applications and reduce duplicated effort when demand is sufficient. An integrated platform may simplify deployment and improve utilisation. These are potential benefits that require measurement in the buyer’s own environment.
The new terminology may also bring political and investor attention to infrastructure, skills and research. However, attention becomes useful only when it supports projects with clear demand and credible delivery plans.
The drawbacks
“Superintelligence” can create expectations that evidence does not support. Employees or customers may assume a system can judge correctly in unfamiliar situations. That assumption becomes dangerous when people stop checking consequential outputs.
Large facilities also create substantial commitments. Hardware, electricity, cooling, staffing and upgrades all affect total cost. High installed capacity can become expensive idle capacity when projected demand fails to materialise.
Energy efficiency and total electricity demand are different measures. A system may use less electricity per task while consuming more overall as usage grows. A 2024 US Department of Energy report projected data centres could consume 6.7 to 12 per cent of US electricity by 2028. This is a dated projection for all data centres, not a measurement of AI alone. [4]
An integrated supplier stack may accelerate deployment while increasing switching costs. Concentrated dependence also creates exposure to pricing changes, outages and supply constraints. Enterprises should assess those trade-offs before committing.
A practical comparison
Compare ambition with evidence: does the proposed capability pass tests on your real work? Compare scale with utilisation: will demand justify the capacity? Compare token efficiency with task efficiency: how much does a correct, approved result cost? Compare integration with portability: what would it take to change suppliers?
Consider an illustrative customer service assistant. Faster replies may reduce waiting time. Incorrect answers may create repeat contacts and complaints. A useful business measure is cost per correctly resolved enquiry, including human review and rework. Token throughput alone cannot answer that question.
What CEOs and CTOs should do
CEOs should select a specific business problem and establish its current cost and service quality. Agree the improvement required before approving a pilot. Include training and changes to staff responsibilities in the business case.
CTOs should test performance on representative enterprise data, including difficult cases and failures. Define human approval points, access controls, incident handling and a way to return to the previous process. Keep the people responsible for the service involved in acceptance testing.
Together, compare owned infrastructure, managed cloud capacity and existing model services. Examine utilisation, data sensitivity, latency and exit costs. Buying access may suit uncertain demand; ownership may suit sustained workloads with the necessary operational capability.
Approve expansion only when measured benefits remain after infrastructure costs, review and rework. Track quality alongside speed. Require a clear owner for each consequential automated action.
The term “superintelligence factory” is a useful invitation to examine the whole production system. My advice is to keep investment decisions tied to demonstrated capability and a business outcome that somebody can verify.
If your enterprise were offered a superintelligence factory tomorrow, what measurable result would justify the investment?
Think mathematically. Engineer intelligently. Lead strategically. Build responsibly. Scale globally. Create real business value through artificial intelligence.
#ArtificialIntelligence #AIStrategy #NVIDIA #Superintelligence #EnterpriseAI #ResponsibleAI
Originally published on LinkedIn by Keshavan Thiagaraja. Read the original on LinkedIn.
