The AI companies revolution: Real benefits, extraordinary costs and a race for computing power.

AI companies promised artificial intelligence. What we got instead

AI companies have transformed artificial intelligence into a vast commercial infrastructure race centred on large language models, enormous data centres and unprecedented capital expenditure, while delivering systems that remain fundamentally probabilistic, unreliable and considerably narrower than the intelligence once envisioned by the field’s pioneers.

The original objectives of artificial intelligence were concerned with machine reasoning, learning, abstraction, language, problem-solving and the possibility of constructing systems capable of exhibiting meaningful aspects of intelligence rather than merely producing convincing imitations of human communication.

The modern generative AI industry has achieved remarkable advances in statistical prediction and machine-generated content, but its progress has also created an extraordinary appetite for semiconductors, electricity, computing infrastructure, construction capacity and investment capital.

Critics such as technology writer Ed Zitron argue that the industry has confused increasingly impressive demonstrations with genuine intelligence while constructing an economic model whose future profitability depends upon enormous assumptions about adoption, productivity and continued capital availability. The resulting debate is therefore about considerably more than whether chatbots are useful, because it concerns whether the technological and financial resources being consumed by AI companies are proportionate to the value their systems are actually creating.

Key Takeaways

Artificial intelligence originally sought to reproduce important characteristics of human intelligence through computational systems.
Large language models generate remarkably convincing language without guaranteeing understanding, factual accuracy or reliable reasoning.
The AI infrastructure race is consuming enormous quantities of capital, electricity, computing hardware and construction capacity.
Circular investment and debt create questions about whether present spending can ultimately be supported by genuine end-user revenue.
The decisive test for AI companies will be whether their systems create sustainable economic value at a cost society can afford.

The original purpose of artificial intelligence

The modern artificial intelligence industry is sometimes discussed as though the technology emerged with ChatGPT, generative images and autonomous coding assistants, yet the intellectual project is considerably older and was built around questions that remain surprisingly relevant to the present controversy. In 1955, John McCarthy, Marvin Minsky, Nathaniel Rochester and Claude Shannon proposed the Dartmouth Summer Research Project on Artificial Intelligence, which subsequently became one of the foundational events in the history of the discipline.

Their proposal argued that aspects of learning and intelligence could potentially be described with enough precision to allow machines to simulate them, and it specifically identified problems involving language, abstraction, concept formation, problem-solving and self-improvement as areas worthy of investigation.

The ambition was therefore profound. Researchers were not principally attempting to construct machines that could produce paragraphs that sounded like people, nor were they attempting to create a mechanism for predicting the next word in a sentence at industrial scale. They were asking whether aspects of intelligence itself could be formalised and reproduced computationally. That distinction matters because the commercial AI industry has inherited the terminology and cultural expectations of that earlier research programme while pursuing technologies that operate according to a substantially different mechanism.

The history of AI subsequently became a succession of competing approaches, including symbolic reasoning, expert systems, neural networks, statistical machine learning and increasingly sophisticated forms of deep learning. Some approaches failed to meet their expectations, producing periods of declining investment known as AI winters, while others became extremely useful without ever approaching general human intelligence. Machine learning eventually demonstrated that computers could identify patterns in enormous datasets with extraordinary efficiency, creating practical applications in areas ranging from speech recognition and computer vision to recommendation engines, fraud detection and scientific modelling.

The breakthrough that ultimately produced the modern generative AI industry came from the convergence of deep learning, enormous datasets, specialised semiconductor hardware and increasingly powerful neural-network architectures. The Transformer architecture introduced by researchers at Google and the University of Toronto in the 2017 paper Attention Is All You Need became particularly important because its attention mechanism proved highly effective at processing relationships between elements in sequences and could be efficiently scaled using modern computing infrastructure. That development provided the foundation upon which today’s large language models were constructed.

IBM Data Engineering Professional Certificate

Prepare for a career as a Data Engineer. Build job-ready skills – and must-have AI skills – for an in-demand career. Earn a credential from IBM. No prior experience required.

When intelligence became prediction

Large language models are extraordinary computational systems, but understanding what they actually do is essential to understanding the argument surrounding them. During training, a model is exposed to enormous quantities of text and adjusts billions of internal parameters so that it becomes increasingly effective at predicting patterns in language. When the trained model subsequently receives a prompt, it generates an output by calculating probabilities for possible tokens within the context provided, repeatedly selecting or sampling from those possibilities until a response has been produced.

The mathematics behind this process is extraordinarily sophisticated, and reducing a modern language model to a simple autocomplete mechanism understates the engineering achievement involved. Nevertheless, the fundamental distinction remains important because statistical capability does not automatically establish human-like comprehension, consciousness, intention or independent understanding of the world. A system can become extraordinarily good at reproducing patterns associated with explanation and reasoning without necessarily possessing the kind of grounded understanding that humans normally associate with intelligence.

This is central to Ed Zitron’s criticism of generative AI. Zitron, the writer and podcaster behind Where’s Your Ed At and Better Offline, has become one of the most prominent and outspoken critics of the commercial generative AI boom. His argument is directed primarily at large language models and the economic system built around them rather than at every form of artificial intelligence or machine learning. In his writing, including his August 2026 AI Hater’s Manifesto, he argues that the industry routinely presents probabilistic systems as though they were autonomous intelligent entities and then builds enormous economic expectations around that presentation.

The criticism becomes particularly relevant when a language model is used in circumstances where accuracy matters. A model can produce a sophisticated legal explanation containing an invented case, generate computer code containing a subtle security flaw, summarise a document while introducing information that was never present or confidently answer a factual question incorrectly. These failures are commonly described as hallucinations, although the term can obscure the underlying technical problem, which is that the system is optimised to generate plausible sequences of tokens rather than to guarantee that every proposition it produces corresponds to reality.

Human beings are also capable of making mistakes, of course, but humans ordinarily possess mechanisms for grounding claims in perception, experience, physical interaction and deliberate verification. A language model does not automatically acquire those mechanisms merely because it has absorbed an enormous quantity of written material. Consequently, a system can be extremely useful in supervised circumstances while remaining unsuitable for unsupervised authority over consequential decisions.

What we actually built

The result is an unusual technological compromise in which machines have become extraordinarily capable at producing the appearance of intellectual labour without necessarily reproducing all of the underlying processes that make human intellectual work reliable. Generative AI can draft correspondence, transform prose, translate material, generate software, analyse structured information, produce images and accelerate certain research activities, and those capabilities are sufficiently valuable that millions of individuals and organisations have adopted them.

The important issue is therefore not whether AI works, because clearly some AI systems work remarkably well for specific purposes. The more important question is what kind of work they perform reliably, under what conditions they require human supervision and whether their economic cost is proportionate to the productivity they create.

This is where the distinction between artificial intelligence and artificial fluency becomes useful. A system that can explain quantum physics in elegant prose may still make elementary errors in the mathematics. A system that can write an impressive computer program may still produce code that fails under real-world conditions. A system that can generate a convincing photograph can produce an image of a physically impossible object without recognising the contradiction. Fluency is therefore an important capability, but it should not automatically be interpreted as comprehension.

The industry’s most ambitious claims about artificial general intelligence create an even greater conceptual problem. Artificial general intelligence has no universally accepted technical definition, no universally agreed benchmark and no established threshold that would objectively demonstrate that such a system had been achieved. Predictions about when AGI will arrive therefore involve assumptions about intelligence, autonomy, reasoning and transfer learning that remain subjects of active scientific disagreement.

Zitron’s language on this subject is deliberately extreme, and his description of AGI as a fictional concept peddled by dishonest actors represents his own polemical position rather than an established scientific conclusion. Nevertheless, his broader criticism raises a legitimate question about whether the industry has sometimes used an undefined future capability to justify present-day investment at a scale that would be difficult to defend solely on the basis of today’s products.

Data Engineering Courses

Data engineering courses can help you learn data modeling, ETL (extract, transform, load) processes, and data warehousing techniques. You can build skills in data pipeline construction, database management, and ensuring data quality and integrity. Many courses introduce tools like Apache Spark, Hadoop, and SQL, that support processing large datasets and optimizing data workflows. You’ll also explore cloud platforms such as AWS and Azure, which facilitate scalable data solutions and enhance your ability to manage data in various environments.

The trillion-dollar infrastructure race

The economic side of the AI revolution is arguably more consequential than the philosophical argument about whether a language model is intelligent. AI companies and the hyperscale technology companies supporting them are building data centres at extraordinary speed, purchasing enormous quantities of advanced processors, securing long-term electricity supplies and committing vast sums of capital to infrastructure that assumes continued growth in computational demand.

The scale of this investment is difficult to comprehend because AI is no longer primarily a software story. The industry increasingly depends upon physical infrastructure involving land, buildings, electricity generation, transformers, cooling systems, networking equipment, semiconductor fabrication and specialised construction. Every additional AI workload therefore requires a physical chain extending from semiconductor manufacturing through data-centre construction and electrical generation to the customer’s device.

Recent estimates illustrate the magnitude of the bet. The Bank for International Settlements has warned about the financial stability implications of the AI investment boom, while reporting in September 2026 indicated that the world’s five largest technology companies could collectively invest more than US$1 trillion in AI during 2025 and 2026. Such figures should not be interpreted as saying that AI companies have literally lost US$1 trillion, because capital expenditure is an investment rather than an accounting loss. They nevertheless demonstrate how much capital is being committed to the expectation that future AI revenues will justify today’s extraordinary spending.

This creates a scarcity problem that extends beyond the AI industry itself. When large technology companies simultaneously demand advanced GPUs, high-bandwidth memory, electrical equipment, data-centre construction and enormous quantities of electricity, those resources become more expensive and more difficult for other industries to obtain. The consequences can reach consumers who have never used an AI chatbot because the same semiconductor supply chains, energy infrastructure and capital markets support many other areas of the technology economy.

The AI boom therefore has an unusual capacity to increase the cost of technological development more generally. A company developing a conventional computing product must compete for hardware and infrastructure against organisations willing to spend billions or even tens of billions of dollars in pursuit of computational supremacy. The question becomes whether that concentration of resources produces enough additional economic value to compensate for the scarcity it creates.

The economics behind the race

Zitron’s most powerful criticism is economic rather than technical because a technology does not need to be conscious, intelligent or revolutionary to be commercially successful, but it does need an economic model capable of supporting its costs.

Generative AI companies have required extraordinary amounts of capital because developing and operating frontier models is expensive. Training requires enormous computational resources, while serving millions of users requires continuous inference capacity. The hardware itself becomes obsolete as new generations of accelerators appear, while data centres require long-term maintenance, power and cooling. Unlike a conventional software product whose marginal distribution cost can be extremely low, AI inference has a substantial computational cost every time a customer uses the system.

The industry’s financial structure consequently depends upon a difficult proposition: future revenue must grow sufficiently rapidly to compensate for today’s capital expenditure and operating costs. That proposition may prove correct, but it has not yet been demonstrated at the scale implied by the infrastructure being constructed.

This is where the accusation of a “Ponzi scheme” requires careful handling. There is no sound basis for claiming that the AI industry literally constitutes a Ponzi scheme, which is a specific form of fraudulent investment operation. Zitron’s argument is better understood as an accusation that parts of the AI economy exhibit circular or self-reinforcing financial characteristics in which technology companies invest in infrastructure suppliers, infrastructure suppliers benefit from rising AI demand, investors finance AI companies based upon expectations of future growth and customers themselves may depend upon venture capital that is ultimately tied to the same investment ecosystem.

Such circular economics can exist without fraud. Venture capital routinely moves through interconnected technology markets, and corporations frequently become both suppliers and customers of one another. The concern arises when the apparent scale of economic activity begins to depend more heavily upon capital circulation and expectations of future growth than upon profitable end-user demand.

Google UX Design Professional Certificate, offered by Google

This is your path to a career in UX design. In this program, you’ll learn in-demand skills that will have you job-ready in less than 6 months. No degree or experience required.

Enroll now for free

NVIDIA, hyperscalers and the feedback loop

NVIDIA provides one of the clearest examples of the extraordinary feedback mechanism created by AI infrastructure spending. The company supplies many of the accelerators that power modern AI workloads, so expanding investment in AI infrastructure creates enormous demand for its products. Hyperscalers purchase processors, construct data centres and sell computing capacity to AI developers, while AI companies use that capacity to develop and operate models that are subsequently sold to consumers and businesses.

As long as the market expects future AI demand to expand dramatically, every participant has an incentive to continue investing. Semiconductor manufacturers expand capacity because customers require more processors, data-centre operators expand because AI companies require more computing, investors provide capital because future AI revenue is expected to increase and AI companies spend that capital because falling behind competitors could mean losing access to the market.

The resulting system can produce extraordinary growth without proving that the final economic return will be sufficient to support the entire structure indefinitely. That is the central financial risk identified by critics of the current boom, because the system remains highly dependent upon assumptions concerning future productivity and revenue.

If those assumptions prove accurate, today’s infrastructure could become the foundation of a major technological transformation. If they prove excessively optimistic, the industry could find itself possessing enormous amounts of expensive infrastructure whose economic utilisation falls below the levels required to justify the investment.

Why everything around AI is becoming more expensive

The physical requirements of artificial intelligence also expose a problem that is rarely visible when consumers interact with a chatbot through a browser. Behind a seemingly inexpensive subscription lies a global infrastructure network involving semiconductor fabrication, advanced packaging, memory production, high-speed networking, data-centre construction and electricity generation.

Energy is particularly significant because AI data centres consume electricity continuously rather than intermittently. As computing demand grows, technology companies are increasingly negotiating long-term power arrangements and exploring dedicated generation, while governments and utilities face the challenge of accommodating rapidly expanding electricity requirements. In regions where electricity supply, transmission capacity or transformer availability is constrained, new data-centre development can compete with residential, industrial and commercial demand.

This creates a paradox in which a technology promoted partly as a mechanism for improving efficiency can itself become a major consumer of scarce resources. If AI eventually increases productivity sufficiently to compensate for that consumption, the economic equation could work. If productivity gains remain modest, the opportunity cost becomes increasingly difficult to ignore.

Google IT Support Professional Certificate

The launchpad to a career in IT. This program is designed to take beginner learners to job readiness in about three-to-six months.

The dot-com comparison needs more precision

Supporters of the AI boom often argue that critics misunderstand technological bubbles because the dot-com crash ultimately left behind infrastructure that became essential to modern life. That argument contains an important historical truth, but it does not automatically establish that today’s AI investment will produce equivalent results.

The internet created a communications infrastructure whose usefulness increased as more people connected to it and more information moved across it. Fibre-optic networks, data centres and routing infrastructure could continue generating value for many years after the companies that originally financed them disappeared.

AI infrastructure has different characteristics because computational capacity requires continual upgrading and substantial ongoing expenditure. A data centre built for one generation of processors can remain physically useful, but its economics can change rapidly as more efficient processors become available. Models also require continuing development, while inference consumes resources every time the system is used.

A speculative bubble can therefore leave valuable infrastructure behind, but that fact does not establish that every investment made during the bubble was economically rational. The railway boom of the nineteenth century and the internet boom of the late twentieth century both demonstrate that transformative infrastructure and financial speculation can coexist within the same historical period.

What AI companies should ultimately be judged on

The debate should therefore move away from the increasingly sterile argument over whether artificial intelligence is “real” and towards a more rigorous assessment of measurable outcomes. An AI system that saves a doctor significant administrative time, helps a scientist identify a promising hypothesis, enables a small company to compete with a much larger organisation or allows an engineer to analyse thousands of documents rapidly can create genuine value regardless of whether the system possesses consciousness.

The same principle applies in the opposite direction. An AI deployment that produces mediocre marketing copy requiring extensive editing, generates unreliable code requiring more supervision than conventional development or adds an expensive chatbot to software without solving a meaningful customer problem may have little economic value regardless of how impressive its demonstration appears.

This is where the AI industry’s future will ultimately be determined. The technology does not need to achieve science-fiction levels of intelligence to succeed, but it does need to generate enough measurable value to support the infrastructure, energy consumption, salaries, financing costs and continuing research expenditure required to operate it.

The companies that survive the coming period will therefore not necessarily be those making the grandest predictions about AGI. They may be those that discover where probabilistic machine intelligence provides reliable economic advantages and deploy it within carefully controlled boundaries.

AWS Certified Solutions Architect Associate (SAA-C03) Specialization

Master AWS Cloud Architecture and Get Certified. Gain hands-on experience with AWS, learning cloud architecture, security, and cost optimization.

We asked for intelligence and built a prediction engine

There is an uncomfortable historical irony at the centre of the AI revolution. The pioneers of artificial intelligence imagined machines capable of learning, reasoning, abstracting and solving problems in ways that would illuminate the nature of intelligence itself. Seven decades later, the world’s most valuable AI companies have built extraordinarily powerful systems that can manipulate language and other forms of information at a scale no previous technology could approach, yet those systems remain fundamentally different from the autonomous, generally intelligent machines once imagined.

Calling today’s AI a “glorified Magic 8-Ball” is intentionally provocative and technically incomplete, because modern language models are vastly more sophisticated than a toy that produces one of twenty predetermined answers. The analogy nevertheless captures a legitimate warning about the difference between generating a plausible response and possessing dependable knowledge of why that response is correct.

The opposite mistake would be equally serious: treating current AI systems as worthless because they do not possess human consciousness or general intelligence. Their practical capabilities are already substantial, and businesses, researchers, programmers, educators and consumers are finding legitimate uses for them every day.

The real controversy is therefore not whether AI exists, nor whether it can be useful. The question is whether the extraordinary economic machinery now surrounding AI is proportionate to the capabilities being delivered.

AI companies are consuming vast quantities of capital, processors, electricity and construction capacity because investors and executives believe today’s spending will produce tomorrow’s transformative productivity. That may eventually prove to be one of the most consequential investments in technological history. It may also prove to be an extraordinary period of overinvestment in a technology whose practical limitations were visible long before its financial commitments reached this scale.

History will not ultimately judge the AI industry by the eloquence of its chatbots, the size of its GPU clusters or the valuations achieved during the boom. It will judge the industry by whether the machines produced enough genuine economic and human value to justify everything society sacrificed to build them.

That is the finish line that matters, and it is considerably more difficult to reach than producing the next plausible sentence.

The genuine benefits of artificial intelligence

A balanced assessment of AI companies must acknowledge that the technology has already delivered substantial benefits, even if those benefits do not validate every prediction made about artificial general intelligence or justify every dollar being invested in the industry. The most important distinction is between the capabilities of artificial intelligence as a broad technological field and the claims made about particular generations of generative AI.

Machine learning has been producing measurable results for years in fields such as medical imaging, scientific research, financial fraud detection, industrial quality control, weather modelling, logistics, language translation and computer vision, while newer generative systems have expanded the range of tasks that ordinary people and businesses can perform with relatively little specialised training.

One of the most significant benefits of generative AI is its ability to reduce the amount of time required for certain forms of information processing. A person who previously needed hours to organise a large collection of documents, extract recurring themes, translate material, produce a preliminary summary or transform information into another format can often accomplish the first stage of that work considerably faster with an AI system.

The resulting material may still require careful human review, particularly when accuracy is important, but reducing the amount of repetitive processing involved can itself represent a meaningful productivity improvement. For small businesses and individuals without specialist staff, that capability can provide access to analytical and creative assistance that would previously have been prohibitively expensive.

Software development is another area where the benefits are becoming increasingly tangible. AI coding systems can explain unfamiliar programming concepts, generate routine sections of code, suggest alternative implementations, identify certain classes of errors and help developers navigate large codebases.

They do not eliminate the need for competent programmers because generated code can contain security vulnerabilities, logical errors or inappropriate assumptions, yet a skilled developer working with an AI assistant can sometimes complete routine tasks more rapidly than a developer working entirely manually. The distinction between assistance and autonomous replacement is therefore important. AI can increase the productivity of an experienced worker without demonstrating that the worker has become unnecessary.

The technology also has considerable potential in education, particularly when used as an interactive supplementary tutor rather than an unquestioned authority. An AI system can explain the same concept in several different ways, adjust the complexity of an explanation, generate practice questions and provide immediate feedback.

For students who have difficulty asking questions in a conventional classroom, an interactive system can provide another avenue for practising concepts without the social pressure associated with speaking in front of peers. The educational value depends heavily upon the quality of the system and the supervision surrounding it, because an AI tutor that confidently teaches incorrect information can create problems rather than solve them.

Scientific research may ultimately provide an even more important test of AI’s value. Machine-learning systems can examine datasets containing millions of observations, identify statistical relationships that would be difficult for humans to detect manually and assist researchers with tasks such as protein structure prediction, materials discovery and molecular analysis.

AlphaFold’s success in predicting protein structures demonstrated that artificial intelligence could make a significant contribution to a genuine scientific problem rather than merely producing commercially attractive content. Such applications illustrate why criticism of generative AI should not become a rejection of artificial intelligence as a whole. There is a fundamental difference between a system that assists scientists with a constrained computational problem and a chatbot being marketed as an emerging digital person.

AI can also provide significant accessibility benefits. Speech recognition, text-to-speech systems, automated captioning, translation and computer-vision technologies can help people interact with digital information in ways that were previously difficult or impossible. Generative systems can further assist people with disabilities by helping them communicate, organise information, interpret documents or interact with software.

These applications are particularly significant because the value of a technology cannot always be measured through corporate revenue alone. A tool that enables an individual to participate more independently in education, employment or everyday communication can have social value even when the immediate financial return is difficult to quantify.

There are also substantial opportunities in developing economies and smaller organisations. Advanced computing and software traditionally required specialist expertise and expensive infrastructure, whereas AI interfaces can increasingly allow users to perform sophisticated information-processing tasks through natural language.

A small company can use AI to translate product information for overseas customers, analyse customer feedback, prepare preliminary market research or automate parts of its administrative workload without maintaining a large technical department. In countries where specialist expertise is scarce, that reduction in the barrier to accessing computational capabilities could become economically significant.

The strongest case for AI therefore does not require claims that machines are becoming conscious, that human beings are about to become economically obsolete or that artificial general intelligence is imminent. A much more defensible proposition is that computers have acquired increasingly powerful abilities to recognise patterns, generate information and assist with tasks that previously required substantial human time. If those capabilities can be deployed at an acceptable cost while maintaining appropriate human oversight, they can increase productivity and broaden access to expertise.

This is also why the economic criticism of AI should not be confused with an argument against investment in the technology itself. Society routinely invests heavily in technologies whose ultimate applications are not immediately obvious, and some of the greatest technological advances emerged from periods of experimentation in which many individual projects failed. The fact that an AI company might eventually collapse does not mean that the underlying technology has no value, just as the failure of individual internet companies did not invalidate the internet.

The more serious question concerns proportionality. If an AI application can save a medical researcher months of analytical work, assist a programmer in maintaining a complex system or provide an accessible educational resource to a student, substantial investment can be rational even when the technology has significant limitations. When billions of dollars are spent developing systems that provide only marginal improvements over existing software, however, the justification becomes weaker. The same applies when businesses deploy AI because they fear appearing technologically outdated rather than because the technology solves a genuine problem.

The benefits of AI are therefore real, but they do not require the mythology surrounding AI to be true. Artificial intelligence can be useful without being conscious, generative systems can increase productivity without possessing human understanding and machine-learning models can contribute to scientific discovery without becoming artificial scientists. Recognising those benefits makes the criticism of the industry’s excesses stronger rather than weaker because it establishes a reasonable standard against which the enormous financial and physical costs of the current AI race can be measured.

The central challenge for AI companies is consequently not to persuade the public that every limitation represents an approaching breakthrough, but to demonstrate where their technology delivers measurable improvements over the alternatives. If an AI system makes workers more productive, helps researchers reach discoveries more quickly, improves accessibility, reduces waste or creates entirely new capabilities, those achievements provide a substantive foundation for continued investment. If the technology instead requires ever-increasing quantities of capital and infrastructure to produce increasingly modest improvements, investors, governments and consumers have legitimate grounds to question whether the race for larger models has become an objective in itself.

A mature AI industry should ultimately be capable of accommodating both realities at the same time: artificial intelligence can be one of the most useful technologies developed in the modern era, while portions of the commercial AI boom can still be excessively speculative, financially fragile and driven by expectations that exceed demonstrated capability. Those propositions are not mutually exclusive, and recognising that distinction provides a more credible basis for evaluating what AI companies have actually built and what society should reasonably expect them to build next.

Coursera | Degrees, Certificates, & Free Online Courses

What AI could actually do for Trinidad and Tobago

For Trinidad and Tobago, the most compelling argument for artificial intelligence has little to do with building the Caribbean’s largest data centre or competing with Silicon Valley to create the world’s biggest language model. A country of approximately 1.5 million people does not need to spend billions of dollars attempting to reproduce the infrastructure of the United States, China or the largest technology economies in order to benefit from AI.

The more rational opportunity is to apply existing and emerging AI technologies to problems that are particularly important to Trinidad and Tobago, while developing enough local expertise to ensure that the country becomes a creator, integrator and exporter of useful solutions rather than merely another market for foreign technology. That approach is already beginning to take shape, with the Government pursuing a national AI assessment and policy framework, while The University of the West Indies has established an Artificial Intelligence Innovation Centre focused on research, capacity building, commercialisation, policy and governance.

One of the clearest opportunities is public administration, where AI could reduce the amount of time citizens and government employees spend navigating repetitive bureaucratic processes. Trinidad and Tobago has numerous government agencies that maintain separate databases, forms, procedures and communication channels, creating administrative friction for citizens and businesses.

Carefully governed AI systems could help people determine which government service they require, identify the documents needed for an application, explain regulations in accessible language, classify incoming correspondence and assist public servants with retrieving information from large collections of legislation, policies and administrative records. The Government’s digital transformation agenda already identifies improved digital access to services as an important objective, while its current AI readiness work is explicitly examining technological infrastructure, governance, education, society and economic development.

Healthcare represents another potentially significant area, particularly because Trinidad and Tobago has a relatively small population but a healthcare system dealing with substantial administrative and resource-allocation demands. AI could assist clinicians by helping organise medical records, identifying patterns in diagnostic images, supporting clinical decision-making and predicting which patients may require additional monitoring, provided that these systems remain subject to professional oversight and appropriate data-protection safeguards.

AI could also improve administrative scheduling, reduce duplicated paperwork and help public-health authorities analyse patterns in disease, hospital utilisation and medication demand. The objective should not be to replace doctors or nurses with chatbots, but to reduce the administrative and analytical burden placed on healthcare professionals so that scarce human expertise can be directed towards patients.

Education may provide an even more accessible national opportunity because Trinidad and Tobago already has a substantial educational infrastructure and a population with strong familiarity with English-language digital services. AI tutors could provide students with additional opportunities to practise mathematics, science, reading, writing and foreign languages outside normal classroom hours, while teachers could use AI-assisted systems to prepare differentiated exercises, identify recurring areas of difficulty and generate preliminary teaching materials.

The technology would need to operate within clear educational standards because a fluent but inaccurate AI tutor could reinforce misconceptions, but the potential benefit is considerable when AI is treated as a supplementary educational resource rather than an automated replacement for teachers. The Government’s FutureReadyTT initiative, announced in 2026, is already intended to expand access to enterprise AI tools across the education system, illustrating how AI adoption is moving from theoretical discussion towards practical deployment.

The country’s universities could become one of the most important foundations for this transformation because Trinidad and Tobago’s greatest long-term AI asset is unlikely to be its electricity supply or physical infrastructure; it is its people. The UWI St Augustine Campus launched its AI Innovation Centre in December 2025, with research interests extending across areas including robotics, cybersecurity, energy, sustainability, climate resilience, agriculture and the digital humanities.

In July 2026, UWI also announced new postgraduate master’s and doctoral pathways intended to develop a critical mass of AI researchers capable of producing knowledge and applying AI to Caribbean challenges. That development is particularly significant because it shifts the conversation from importing AI products towards developing people capable of questioning, adapting, governing and creating them.

Agriculture offers another area where Trinidad and Tobago could obtain practical value without attempting to build frontier AI models. Computer vision and machine-learning systems can potentially identify crop diseases, monitor plant health, analyse weather patterns, optimise irrigation and improve agricultural planning when combined with suitable sensors and reliable local data.

For a country seeking greater food security, even relatively modest improvements in agricultural productivity could have economic value. The same technologies could eventually be adapted for other Caribbean countries facing similar challenges, creating an opportunity for Trinidad and Tobago to develop specialised agricultural software that can be exported throughout the region rather than attempting to compete globally on general-purpose AI.

Energy may be an even more strategically important field because Trinidad and Tobago already possesses extensive expertise in oil, natural gas, petrochemicals and industrial operations. AI-assisted predictive maintenance could analyse sensor data from industrial equipment to identify patterns associated with impending failures, allowing maintenance teams to intervene before expensive breakdowns occur.

Machine-learning systems could also assist with energy forecasting, industrial optimisation, process control and environmental monitoring. The country’s existing engineering and energy expertise therefore provides an advantage that a technology company starting from scratch in another country might not possess. The opportunity lies in combining local domain knowledge with modern AI rather than assuming that technological progress requires abandoning existing industrial strengths.

The country’s geographical position also creates opportunities in logistics, shipping and aviation. Trinidad and Tobago sits at an important point in the southern Caribbean and already has established connections with regional and international markets. AI could be used to improve cargo forecasting, route optimisation, port scheduling, inventory management and supply-chain risk analysis.

Similar systems could assist airlines and tourism operators by analysing booking patterns and demand fluctuations, potentially helping businesses respond more efficiently to changes in visitor behaviour. The value in these applications does not depend upon an AI system pretending to be a human employee; it depends upon finding patterns in large datasets that humans would struggle to process manually.

Tourism could similarly benefit from AI that understands Trinidad and Tobago’s actual cultural and geographic characteristics rather than presenting the country through generic Caribbean marketing. AI-assisted translation, personalised travel planning, visitor analytics and customer-service systems could help hotels, attractions and tourism operators communicate with international visitors more efficiently.

More importantly, locally trained or locally curated systems could incorporate information about Carnival, Tobago’s marine environment, the country’s culinary traditions, heritage sites, festivals and regional communities in ways that generic international travel systems may not understand adequately. This creates an opportunity to use AI as an amplifier of Trinidad and Tobago’s cultural identity rather than allowing algorithms designed elsewhere to define how the country is presented to the world.

Small and medium-sized enterprises could be among the biggest beneficiaries because they often lack the financial resources to employ dedicated teams for marketing, data analysis, customer service, bookkeeping and software development. A small manufacturer could use AI to analyse inventory and purchasing patterns, a restaurant could examine customer feedback, a retailer could improve demand forecasting and a professional services firm could automate portions of document processing. The economic significance of these relatively mundane applications could exceed the value of spectacular demonstrations because thousands of small improvements distributed throughout an economy can collectively produce meaningful productivity gains.

There is also a major opportunity in cybersecurity. As Trinidad and Tobago becomes increasingly dependent upon digital services, government networks, financial institutions, utilities and businesses face growing exposure to cyber threats. AI can assist security teams by identifying unusual network behaviour, prioritising alerts, analysing large quantities of security logs and detecting patterns associated with fraud or intrusion attempts. The technology itself is not a complete defence because attackers can also use AI, but developing local expertise in AI-assisted cybersecurity could strengthen national resilience while creating specialised skills that are commercially valuable throughout the Caribbean.

Financial services provide another natural application. Trinidad and Tobago has a sophisticated banking and financial sector that could use machine-learning systems for fraud detection, transaction monitoring, customer-service automation and risk analysis. AI could identify unusual patterns across millions of transactions far more rapidly than human analysts, allowing financial institutions to investigate suspicious activity earlier. Such applications also demonstrate why the distinction between narrow AI and generative AI matters. A carefully designed fraud-detection model does not need to write poetry, simulate a personality or claim to possess general intelligence. It needs to identify anomalous patterns reliably, and that is a considerably more measurable engineering problem.

Climate resilience is particularly important for Trinidad and Tobago because Caribbean states face exposure to hurricanes, flooding, extreme rainfall, coastal hazards and other environmental pressures. AI can contribute to disaster preparedness by analysing meteorological information, satellite imagery, historical flood data and geographic information to improve forecasting and risk assessment.

The technology could potentially help emergency-management authorities identify vulnerable infrastructure, optimise evacuation planning and allocate resources more efficiently. The UWI AI Innovation Centre has specifically identified climate resilience among its areas of applied AI research, making this a promising example of a technology agenda aligned with Caribbean conditions rather than imported wholesale from larger economies.

Perhaps the greatest opportunity, however, is the development of a Caribbean AI industry that sells specialised expertise to the wider region. Trinidad and Tobago has an established educational system, a large English-speaking professional workforce, telecommunications infrastructure, financial institutions, energy expertise and longstanding connections throughout CARICOM.

Those characteristics could support a technology-services industry specialising in AI implementation, data preparation, cybersecurity, governance, model evaluation, business-process automation and sector-specific applications. Instead of attempting to build a trillion-dollar frontier model, local companies could develop solutions for Caribbean governments, banks, insurers, hospitals, schools, tourism operators and energy companies facing similar problems across multiple jurisdictions.

This approach would also create a more realistic employment opportunity for young Trinidadians. The country does not need every student to become a machine-learning researcher, because an AI economy requires many different forms of expertise, including data analysis, cybersecurity, software engineering, mathematics, statistics, product design, business analysis, digital marketing, law, ethics, governance and sector-specific knowledge.

The UWI and American Chamber of Commerce of Trinidad and Tobago partnership announced in 2026 specifically emphasises research, workforce development, professional training, internships, AI literacy, cybersecurity and data analytics, demonstrating the importance of connecting education with actual industry requirements.

The critical point is that Trinidad and Tobago should avoid measuring AI success by the same metrics used by Silicon Valley. The country does not need to win a race for the largest model, the greatest number of GPUs or the most expensive data centre. It needs to determine where AI can increase national productivity, improve public services, strengthen resilience, create high-value employment and generate exportable intellectual property without imposing disproportionate financial, environmental or social costs.

That approach would also provide a useful counterargument to the excesses criticised by Ed Zitron. If AI is going to consume enormous quantities of computing power and capital, Trinidad and Tobago has an opportunity to demonstrate a different model in which the technology is evaluated according to concrete outcomes rather than valuation, hype or model size. A successful Caribbean AI strategy could therefore be deliberately modest in infrastructure while being ambitious in application, concentrating investment on people, education, data, cybersecurity, research and locally relevant software.

The country is already establishing some of the institutional foundations required for that approach. The national AI readiness assessment being developed with UNESCO and UNDP is intended to inform policy, investment priorities and a longer-term national AI strategy, while UWI’s AI initiatives are creating research and postgraduate capacity and the Caribbean AI Task Force has begun developing a regional framework for coordinated AI governance and innovation.

For Trinidad and Tobago, the most valuable AI may therefore be the least glamorous. It may be the system that helps a farmer identify a disease before an entire crop is lost, enables a small exporter to enter a new market, helps a teacher support a struggling student, allows a doctor to process information more efficiently, identifies a cyberattack before it becomes a national incident, helps a government agency process application faster or allows an engineer to prevent an industrial failure.

Those achievements would not make Trinidad and Tobago an AI superpower in the conventional sense, and they would not require the country to participate in the trillion-dollar race for ever larger models. They would accomplish something considerably more relevant: using artificial intelligence to solve Trinidad and Tobago’s actual problems while developing enough local knowledge and intellectual property to ensure that the economic benefits remain within the country and can subsequently be exported across the Caribbean.

That may ultimately be a more sustainable vision of artificial intelligence than the one currently being sold by the world’s largest AI companies. Instead of asking how much computing power a country can consume, Trinidad and Tobago can ask how much human capability it can augment, how much productivity it can create and how many distinctly Caribbean problems it can solve. In that context, AI becomes neither a miracle nor a threat, but a tool whose value can be measured by what it enables people to accomplish.

Further reading on Ed Zitron’s views on AI (primarily generative AI / LLMs):

His own writing

The AI Hater’s Manifesto (Where’s Your Ed At, Aug 2026) — comprehensive statement of his position on LLMs, costs, limitations, and the broader industry.
Archive of his newsletter (Where’s Your Ed At): wheresyoured.at — frequent deep dives on finances, data centers, OpenAI/Anthropic, NVIDIA, circular financing, and related topics.

Major interviews and profiles