The biggest debate in technology this year has been about scale: chips, training data, model parameters, and how many billions it takes to build the next frontier system. But that conversation might be missing the point.
“We need to stop talking about AI as if it’s a function of GPUs and data and model size and parameters,” says Vilas Dhar, president of the Patrick J McGovern Foundation, a US philanthropic organisation that funds artificial intelligence research aimed at social impact. “What does this enable us to do as a society? Where are the risks that come not because we’re worried about superintelligence, but because we’re deeply worried about the concentration of power?”
That concentration is the part of the story that gets less attention. Decisions about how AI will reshape work, healthcare, education, and governance are being made in a handful of rooms in Silicon Valley, London, and Beijing. The rest of the world, including most of Africa, is expected to adapt.
Bigger budgets keep missing the bigger questions
History offers a useful reference point. The Industrial Revolution is usually framed as a technological shift, but new institutions, labour arrangements, and ideas about citizenship emerged alongside the steam engine. The technology mattered, but so did the systems built around it.
“We can talk about technological innovation all day long, but if we don’t also innovate our human systems, our institutions, our political and economic models, we’re missing the point,” Dhar says.
The current obsession with bigger models reveals something almost juvenile about how the field discusses itself. The focus was on more computing power, more data, and more capital, as if scale alone produces progress. Then a Chinese lab released DeepSeek, a model that matched frontier US systems at a fraction of the training cost.
“Innovation isn’t constrained to those who can have the biggest chequebook, or even at this point the most computing power,” Dhar says. “What we need is real innovation, and innovation with purpose and soul.”
Ethics has to live inside business models, not slide decks
The phrase “ethical AI” appears in almost every corporate AI policy document. But while the principles are easy, the application is hard.
“Ethics is spoken at a very high level,” says Alessandra Sala, senior director of AI and data science at Shutterstock, the global platform that licenses images, video, and music to creative professionals. “But maintaining this discussion at those levels doesn’t help us to progress and make this technology useful and usable in practical settings.”
Sala also leads work with UNESCO on the ethics of AI, looking at the technology through the lens of those most likely to be affected by it: women, minorities, and groups historically underrepresented in how it gets designed.
Shutterstock had a particular problem to solve. The company holds about 500 million images, videos, and pieces of music, which put it in the rare position of being able to train generative AI on properly licensed content. That raised another question about the artists who created the source material.
“In order to bring this technology in an ethical way, we need to compensate the artists who were the first to create the content that is then used by those models,” Sala says. “How do we rethink the complete end-to-end value chain in AI? And how do we compensate the producer of data that is so important to create a high-quality model?”
The Shutterstock approach treats data contributors as partners rather than free inputs, a small example of what ethics looks like when it stops being a slide deck and starts being a business decision.
Enterprise is where the real change is happening
Most public attention on AI focuses on consumer products such as Gemini and Claude. These have created a perception that AI hasn’t changed much in daily life since ChatGPT arrived in late 2022.
That perception is misleading, according to Martin Kon, the former president and chief operating officer of Cohere, a Toronto-based AI company that builds enterprise-grade systems for large organisations.
“This is the biggest disruption and transformation in 30 years, since the Mosaic browser came out in 1993,” Kon says. “Initially, companies would hire a design firm to create an HTML page, put a photo of a CEO and her phone number underneath, and say, ‘We’re digital now.’ Of course, that wasn’t true, but everything changed after that. Every job, every customer service interaction, every business.”
The same pattern is unfolding with AI. Early hype generated demos and pilot projects that never moved into production. What’s different now is that some companies are pushing past that stage. For example, Fujitsu deployed Cohere’s technology to power Takane, currently the strongest-performing Japanese language model. Saudi telecoms operator STC also partnered with Cohere to build an AI platform for other telcos.
“Enterprises that were quick to adopt in production at scale are the ones who were the winners of the next 10 or 20 years,” Kon says. “And those that didn’t, many don’t exist today.”
The valuable applications often involve what Kon calls “lost knowledge”, information sitting inside an organisation that nobody can find.
“We can use this technology deployed privately inside your data environment to basically help you find needles in haystacks,” he says. “Every year those haystacks get bigger and the needles get more valuable.”
Regulation still confuses two very different things
The European Union has moved fastest on AI regulation. Its AI Act categorises systems by risk level and imposes corresponding obligations. The approach has merit and significant gaps. Kon argues that one of those gaps is the failure to distinguish properly between consumer and enterprise applications.
“It’s not the same thing if you’re putting out a consumer chatbot to 300 million consumers that may have issues with bias or toxicity,” he says. “In enterprise, it’s very different. You know how a bank would want to talk to its high-net-worth individuals or make decisions around insurance policies. They’re going to put their own rules on that.”
Cohere released its Secure AI Frontier Model Framework in Paris in early 2025, focused on evidence-based risks rather than “some of the hyperbole around existential risks, maybe because we watched too many 1950s science fiction movies,” Kon says.
The Paris discussions also raised an uncomfortable question about who gets to write the rules. OpenAI’s Sam Altman has spoken about the need for a new social contract around AI. The question is who exactly negotiates it.
“If you only have big established tech companies, that’s a bit like in the mid-90s, having Olivetti, EDS, and Nortel talking about the internet,” Kon says. “None of them is around right now.”
Creativity gets democratised, not destroyed
A common fear about generative AI is that it will hollow out creative work. Sala has a different take. The brain develops and stays healthy through creative engagement, she points out, and AI can extend who gets to participate in it.
“I’m a person who works in the media industry but has zero creative capability. I can write code, but no one asked me to draw anything,” she says. “Now AI is giving us an opportunity to voice our ideas and our creativity that can be manifested through different channels. It has democratised the possibility to express ourselves.”
The harder question for African enterprises, governments, and citizens is whether the systems being built will give regions outside the AI heartlands any meaningful agency in how this unfolds. That depends less on which chatbot wins and more on whether the institutions, training pipelines, and regulatory frameworks being designed now leave room for participation.
“AI policy and these questions about how AI will transform our lives are truly universal across the planet,” Dhar says. “If we don’t have mechanisms to have scientific expertise, to have practitioners, to have enterprises and civil society at the table, then all we’re going to do is keep asking questions about whether one particular device will change one particular facet of our experience. Undergirding all of this is a societal transformation at scale. This is the opportunity ahead.”
Africa’s AI maths
In late 2025, South Africa hosted the first G20 summit ever held on African soil. The presidency anchored the agenda around digital inclusion and AI capacity, launching the “AI for Africa” cooperation platform and a $100 million B20 Digital Inclusion Fund. UNESCO estimates the continent could add $2.9 trillion to its economy by 2030 but currently captures just 2.5% of the global AI market and 0.3% of projected worldwide investment.
The African Union adopted its Continental Artificial Intelligence Strategy in July 2024, accompanied by a $60 billion Africa AI Fund. Through mid-2025, 159 African AI start-ups had raised $803 million cumulatively over five years. Global private investment in AI hit $100 to $130 billion in 2024 alone, and AI took 46% of global venture funding by the third quarter of 2025. Within Africa, Nigeria, Kenya, South Africa, and Egypt account for around 83% of that start-up investment.
The capacity gap surfaced in painful form earlier in 2026, when South Africa published a draft National AI Policy proposing a national AI commission, an ethics board, a regulatory authority, an ombud, a safety institute, and an AI insurance superfund to compensate citizens harmed by AI systems. Within weeks, News24 journalists discovered that at least six of the draft’s 67 academic citations were AI-generated hallucinations, referring to journals that don’t exist and papers that were never written.
The minister withdrew the document and suspended two officials who had used AI to draft it without disclosure. An independent expert panel, chaired by machine intelligence professor Benjamin Rosman, has been rebuilding the policy, with a new public draft expected in January 2027. The continent’s leverage in this transition will sit less in matching frontier computing budgets than in building the human capacity to decide how AI gets used.
