← Back to essays

    Africa’s Real AI Prize Is Physical

    32 min read
    Africa’s Real AI Prize Is Physical

    In Greek mythology, Antaeus was a giant of terrifying power. The son of Poseidon and Gaia, of the sea and the earth, he challenged all travellers who crossed his land to wrestle him, and he killed every one of them. He was, for all practical purposes, invincible. His power had a condition. If his feet touched the ground, if he was in contact with the earth, with his mother Gaia, no force could defeat him. Every time he was thrown down, the earth restored his strength. Hercules, who eventually overcame him, did so not by out-wrestling him but by discovering the secret: he lifted Antaeus off the ground and held him in the air until all the power drained from his body. Suspended above the world he was born to touch, the giant became nothing.

    That story is very old. It is also, I would argue, the most precise description of what happens to technology that never meets the physical world.

    Ideas that stay in the air are weightless. Systems that never touch real environments, real bodies, real soil, real machinery, systems that exist only in the abstraction of screens and software, carry none of the power they promise. They can be beautiful. They can be clever. They can raise money and generate press releases and fill conference stages. Yet they do not change the weight of anything. They do not save a life, repair a road, recover a harvest, or keep the lights on. In my own Shona tradition, a similar instinct runs deep: the earth is not backdrop. It is ‘mubereki’, the parent, the source, the place where power is rooted and legitimacy is earned. Ideas and tools that never touch the soil of real life are, in that understanding, unanchored. Powerful in neither world.

    I am writing this for the founders, researchers, and students who are deciding right now what to build next. Many of them are among the most talented technology builders this continent has produced. Most of them are looking in the same direction: software. Apps. Platforms. Fintech tools. SaaS products. The reasons are understandable. Software is cheaper to build, faster to iterate, easier to fund. The global playbook is legible. You can see what success looks like.

    The biggest technology opportunity in Africa is not there. It is in the physical world. In sectors that define the daily reality of hundreds of millions of people. In industries that are dangerous, inefficient, undercapitalised, and almost entirely underserved by people who understand both the technology and the terrain.

    Qasar Younis is the co-founder and CEO of Applied Intuition, a $15 billion physical AI company whose customers include 18 of the top 20 global automakers, the largest mining and construction companies on earth, and the United States Department of War. He is not a peripheral voice. He recently said plainly that the real impact of AI in the next five to ten years is going to be in farming, mining, and construction, not in the software platforms that dominate the technology conversation. That is the view from the centre of the industry. It is worth taking seriously.

    Physical AI, machine intelligence embedded in real environments, in machinery, infrastructure, vehicles, and landscapes, is arriving faster than most people in African technology circles are accounting for. Almost nobody here is building it. That is the problem. It is also, for those willing to put their feet on the ground, the greatest opportunity this generation of African builders has.

    Mining: the physical imperative at the centre of the global economy

    Sub-Saharan Africa holds approximately 30% of the world’s proven critical mineral reserves. The DRC produces over 70% of global cobalt. South Africa produces roughly 70% of global platinum. Zimbabwe, Mali, and Namibia are emerging as vital sources of lithium and graphite for electric vehicle batteries. The IEA projects that global demand for lithium will rise tenfold and cobalt will triple by 2050. The physical inputs for the global energy transition are, to an extraordinary degree, sitting under African soil. The world’s ability to decarbonise depends substantially on Africa’s ability to extract.

    That extraction is currently happening under conditions that are dangerous, expensive, and increasingly incompatible with what global capital markets are willing to accept. Mining accounts for roughly 1% of the global workforce but around 8% of fatal workplace accidents. South Africa’s 2024 figures, 42 fatalities and 1,841 occupational injuries, were the lowest on record and still represent South Africa accounting for 15 of the 42 fatalities reported globally by ICMM member companies that year. The continent’s deepest and most productive mines are also its most dangerous, and the gap between African safety performance and that of peers in Australia or the Americas remains wide.

    The equipment failures that define daily operational reality are not administrative problems. Mining companies spend between 35 and 50% of their annual budgets on equipment maintenance, and a single unplanned breakdown of a primary crusher or haul fleet costs an average of $180,000 in lost production. Some open-pit mines in South Africa suffer up to 20% operational downtime, far behind the global benchmark of 85% availability. Energy instability compounds every mechanical failure: South African load-shedding disrupted commercial operations for over 1,200 hours in 2023, costing the broader economy an estimated $6 billion. For a deep-level mine, a sudden loss of power does not merely pause production, it triggers safety evacuations, shuts down underground ventilation and dewatering pumps, and causes severe physical damage to the heavy electrical motors driving processing mills.

    Physical AI addresses these failures directly, and the evidence is no longer theoretical. The Syama gold mine in Mali, operated by Resolute Mining, was designed from inception as the world’s first fully automated underground mine. Running Sandvik autonomous drill rigs and driverless haul trucks over a subterranean fibre-optic network, Syama achieved a 30% reduction in overall mining costs. Autonomous machines require no shift handovers and are immune to the toxic blast gases that force ventilation pauses for human crews, machine utilisation reached 22 hours a day. That is Mali, not Western Australia. At the Kibali mine in the DRC, the continent’s largest gold mine, remote stope bogging allows Congolese operators to control heavy underground loaders from a surface control room, removing the worker physically from the hazard while maintaining extraction velocity. In early 2025 Kibali recorded a 15% quarter-on-quarter production increase operating primarily on automated underground haulage. These are not pilots. They are operating systems in the field, on this continent.

    Predictive maintenance is the clearest commercial wedge. Acoustic, vibration, and thermal sensors mounted directly on crusher bearings and mill motors allow machine learning algorithms to detect the microscopic degradation patterns that precede catastrophic failure, enabling pre-emptive replacement during scheduled maintenance windows rather than emergency stoppages. Peer-reviewed underground deployments show AI-driven predictive maintenance cutting unplanned downtime by up to 70%. The return on that investment, against a baseline of $180,000 per incident and maintenance budgets consuming half of operating expenditure, is immediate and quantifiable.

    Tailings management is where the stakes of getting this wrong are highest and most visible. In September 2022, the Jagersfontein tailings dam in South Africa collapsed. Three people died. Forty were injured. Toxic slurry destroyed local infrastructure across a wide area. Following the even larger Brumadinho disaster in Brazil, the Global Industry Standard on Tailings Management now mandates continuous digital telemetry and automated early warning systems for tailings facilities. African operators are no longer choosing whether to deploy physical sensing on their dams. Regulation has made it compulsory.

    One honest qualification belongs here. Mining procurement cycles are among the most conservative in any industry, years of vendor vetting, massive liability insurance requirements, and a strong default toward bundled solutions from established OEMs like Caterpillar and Sandvik. African startups targeting the majors risk spending years in pilot programmes without reaching commercial scale. The smarter path for early-stage founders is the mid-tier: agile producers with faster decision-making, real operational pain, and less entrenched incumbent relationships.

    African mining is at the centre of the global energy transition whether it wants to be or not. The minerals are here. The demand is coming. The only question is whether the intelligence layer that makes extraction safer, more efficient, and more sustainable is built by people who understand these operations from the inside or imported from elsewhere by people who do not.

    Agriculture: information is not enough

    In 2023, 298.4 million people in Africa experienced hunger. The prevalence of moderate to severe food insecurity across the continent reached 58%, nearly double the global average. This is not a drought statistic or a conflict statistic. It is a productivity statistic. African cereal yields average 1,250 kilograms per hectare against a global average of 4,100 kilograms per hectare. Less than 4% of arable land is irrigated, against a global average of roughly 20%. Rainfed agriculture supports the vast majority of the continent’s food production while being entirely dependent on rainfall that is becoming less predictable every year. Africa has been feeding more people by farming more land, not by farming better. Between 1980 and 2018, cereal output tripled because land under cultivation expanded from 48 million to 112 million hectares. Yields improved by only 30% over the same period. That approach has a physical limit, and it is approaching.

    The technology community has been trying to solve this for a decade. The dominant strategy has been information: SMS agronomic advice, digital market linkage platforms, weather forecasting apps, crop insurance schemes. A comprehensive meta-analysis of digital interventions in Sub-Saharan agriculture found that while they improved fertiliser adoption, their actual impact on crop yields averaged just 6%. Knowing that a crop requires water does not irrigate the field. Receiving an SMS alert that Fall Armyworm has been detected in the region does not apply targeted biopesticides to a specific hectare. Information is not enough. The failures of African agriculture are physical, biological, and thermodynamic, and they require physical intervention to fix.

    The Fall Armyworm example deserves its own paragraph because it illustrates the gap precisely. Ninety-five percent of Africa’s maize crop is grown in areas climatically suitable for year-round Fall Armyworm infestation. The pest has established a permanent, continent-wide foothold. The current response is reactive: an agronomist or an SMS system alerts a farmer that the pest has been detected, often after significant crop damage has already occurred, and the farmer applies pesticide broadly. Autonomous acoustic and visual sensors deployed in fields can detect the specific frequencies and morphological signatures of invasive species long before human scouting is possible, enabling targeted spraying of affected areas rather than blanket chemical application. Drones equipped with multispectral cameras can map infestations at field level and deploy biopesticides over terrain that tractors cannot reach without compacting fragile soils or requiring passable roads. That is not a future technology. It is a current one, waiting for founders who will build it for African field conditions at African price points.

    Irrigation is the other structural physical gap. Less than 4% of cultivated land is equipped for irrigation. Solar-powered drip systems paired with IoT soil moisture sensors that calculate precise evapotranspiration rates and deliver exact water volumes to the root zone, without manual labour, without guesswork, without waste, are deployable now on smallholder plots using pay-as-you-go financing aligned with harvest cycles. The farmer captures the full return on efficiency gains. The incentive structure is clean. SunCulture in Kenya has demonstrated solar irrigation on smallholder farms at commercially viable price points. The technology works. The question is how fast it scales.

    Post-harvest loss is where the physical failure compounds everything that came before. Up to 37% of total food produced in Sub-Saharan Africa is lost before reaching a consumer, with maize farmers reporting losses of 20 to 27% of harvests to rot, pests, and mould from lack of hermetic storage or temperature-controlled transit. IoT sensor integration in perishable supply chains reduces spoilage-related losses by 24 to 38%. Cold chain robotics and monitoring that can operate in low-infrastructure environments, off-grid, ruggedised, priced for cash-constrained operators, directly recapture economic value from food that is already grown and already lost. That is not yield improvement. That is recovery of output that already exists.

    African farmers are not elderly. Research drawing on nationally representative surveys across six African countries finds that the average age of people primarily engaged in farming is between 32 and 39 years. The continent has a median age of around 19. The problem is not a retirement crisis. The problem is that farming is too unproductive, too unrewarding, and too physically brutal to hold the next generation in it. Young people are leaving agriculture not to retirement but to cities and screens. Physical AI that makes farming less physically punishing, more productive, and more commercially viable is not a labour-replacement argument. It is the argument for making farming worth staying in.

    The opportunity is not to build another advisory platform that floats above a broken physical system. It is to rebuild the physical system itself, and to make farming productive enough, precise enough, and modern enough that the next generation sees a future in it rather than a reason to leave. Almost none of the technology builders with the skills to do this are working on it.

    Construction: where physical failures have physical causes

    To fast-track Africa’s structural transformation to meet its development needs, the continent requires an estimated $402.2 billion in annual infrastructure investment through 2030. It is not getting anything close to that. The resulting gap is not abstract, it is the road that does not exist, the hospital that was never built, the housing deficit that now exceeds 50 million units across the continent. African cities are absorbing millions of new residents into informal settlements because the formal construction sector cannot build fast enough, safely enough, or cheaply enough to keep pace.

    The reason it cannot is largely physical. Software cannot prevent a building from collapsing if the concrete was improperly cured. A project management dashboard cannot detect a missing safety harness before a worker falls. These are not process failures or governance failures. They are physical execution failures, and they are killing people and destroying capital at a scale that is simply not being addressed by the digital tools that dominate the African technology conversation.

    The safety numbers are the starkest evidence. Sub-Saharan Africa records 21 construction fatalities per 100,000 workers and 16,012 injuries per 100,000 workers. In Kenya, the rate reaches 64 per 100,000, against 0.44 in the United Kingdom. Surveys across the region find that 94.8% of construction workers do not use proper protective equipment, not because it is unavailable in every case, but because manual enforcement has comprehensively and repeatedly failed. A supervisor cannot watch every worker on a large, fragmented site simultaneously. A computer vision system running on a standard camera can, continuously, and can alert supervisors to violations in real time before accidents occur.

    The structural collapse data is even more instructive about where physical AI belongs. Analysis of building collapses in Nigeria over a decade found that 68.5% were caused directly by substandard materials or poor concrete curing, not design errors, not earthquakes, not age. Concrete poured incorrectly, or allowed to cure under the wrong thermal conditions, becomes a structural liability that no inspection regime can reliably catch after the fact. IoT sensors embedded directly into concrete before a pour transmit real-time thermal and strength maturity data as it cures, alerting contractors to deviations before the slab is load-bearing. That is physical AI intervening at the precise failure point that kills people. South Africa generates 20.2 million tonnes of construction and demolition waste annually, much of it from rework, tearing down and rebuilding structures that failed quality checks after the fact rather than preventing the failure during execution.

    The proof points are on this continent. In Kenya, a single COBOD 3D printer produced 10 houses in 10 weeks in 2023, a pace that manual methods cannot approach for affordable housing at that price point. Drone surveying replaces weeks of manual topographic work with hours of autonomous flight, producing accurate cut-and-fill calculations that prevent the earthwork errors that generate much of the waste and rework. Predictive maintenance systems retrofitted onto generators and heavy plant flag failures before they cause the site-wide halts that cost heavy industrial operations an average of $125,000 per hour. Early energy optimisation deployments in South Africa have cut diesel consumption on grid-unstable sites by 10 to 15%.

    The African founder advantage here is not about building the most sophisticated robotics. Capital-intensive autonomous bricklaying machines face genuine economic friction in markets where manual labour is abundant and cheap. The opportunity is in hardware-lite, ruggedised, retrofitted intelligence: computer vision running on existing site cameras, concrete sensors that cost a fraction of a laboratory test, drone survey services that replace surveyors on fragmented sites, predictive maintenance modules that bolt onto the heavy plant already on site. Solutions designed for intermittent power, low bandwidth, extreme dust, and the constant threat of theft, designed, in other words, for conditions that a founder who has worked on these sites understands intuitively.

    Africa needs to build more infrastructure in the next 25 years than it has built in its entire history. The tools that make that buildout safer, faster, and less wasteful must be physical, and they must be built by people who know what these sites are like.

    Energy: the capital is there, but it cannot see in the dark

    Sub-Saharan Africa accounts for eight out of every ten people globally who lack electricity, around 600 million people. That number captures the access problem. The operational problem is different, and in some ways more instructive. African enterprises experience an average of 50 hours of power interruptions per month, roughly 25 lost working days a year. Firms lose approximately 5% of annual sales directly to outages. To survive, 70% of businesses in Nigeria and nearly as many across the continent run diesel backup generators at an estimated $0.47 per kilowatt-hour, three times the cost of grid power where grid power exists. Research from Benin found that 44% of grid-connected firms’ total electricity spending goes not toward production but toward defensive measures: generators, voltage stabilizers, backup systems. When those hidden costs are counted, 60% of African businesses are severely electricity-burdened. Africa is not just paying for energy. It is paying twice, once for a grid that fails, and again for the hardware to survive the failure.

    The physical infrastructure underpinning these failures is degrading without being watched. In West Africa, technical and commercial losses on national grids frequently exceed 39% of total output. Transformers fail from thermal degradation and moisture ingress that continuous IoT monitoring would catch weeks in advance. Distribution lines sag into vegetation and cause faults that maintenance crews spend days locating through manual sectionalization, guessing where the problem is rather than knowing. Phase imbalances from unmonitored illegal connections cause equipment overheating and cascading failures along entire distribution feeders. Nigeria’s national grid collapsed 222 times between 2010 and 2022. South Africa experienced load-shedding on 332 days in 2023. These are not governance failures dressed up as engineering problems. They are physical assets failing in the dark because nobody embedded the intelligence to see them.

    The decentralised energy transition is already happening, and it is creating an urgent need for physical intelligence that the market is not yet supplying. Global battery pack prices have collapsed 93% since 2010, reaching $192 per kilowatt-hour for utility-scale systems in 2024. Solar’s share of global mini-grid capacity surged from 14% to 59% between 2018 and 2024, while diesel dropped from 42% to 29%. Africa imported 15 gigawatts of solar panels in 2025, a 60% increase on the previous year. The physical generation hardware is arriving at scale. What is not arriving with it is the intelligence layer that makes distributed generation reliable: battery management systems that optimise charge and discharge cycles in 40-degree ambient temperatures to prevent thermal degradation of expensive storage assets; swarm intelligence controllers that allow mini-grid nodes to reconfigure autonomously when components are added or fail, eliminating single points of collapse; predictive maintenance systems that flag transformer anomalies before a blackout, not after.

    The most revealing number in African energy is not about generation or access. It is this: $9 billion has been committed to the African mini-grid sector by donors and private equity. Actual disbursement stands at 14%. The capital exists. It is not flowing because financiers cannot adequately monitor, trust, or verify the physical performance of remote assets once deployed. Physical AI is not just an operational tool in this context; it is the verification layer that unlocks capital that is already committed and sitting idle. A peer-reviewed study in Nigeria found that a 1% increase in smart meter installation yields a proportionate 0.8% decrease in aggregate technical, commercial, and collection losses. The connection between embedded hardware and recovered revenue is empirically established. The problem is not knowledge. It is deployment.

    The honest qualification runs through this section as it does through water and logistics. Governance failures, non-cost-reflective tariffs, political interference, utility insolvency, are not engineering problems. Providing a diagnostic tool to a utility that cannot afford the replacement transformer the AI has correctly predicted will fail in two weeks yields no cure. The physical AI case is strongest where the end user has both the incentive and the operational capacity to act: commercial and industrial operators paying $0.47 per kilowatt-hour for diesel who will immediately benefit from reliable solar-plus-storage; mini-grid operators whose business model depends on uptime they can verify and report to lenders; mining, manufacturing, and telecom operations where energy reliability has a direct and quantifiable cost. One additional risk worth naming for founders specifically: importing sensors and batteries priced in dollars while collecting revenue in Nigerian naira or Ghanaian cedi, both of which have experienced severe recent devaluations, can destroy margins even with strong customers. Hardware ventures in African energy need dollar-denominated revenue pathways or strong FX hedging from the start.

    The solar panels are arriving. The mini grids are being built. The capital is committed. What is missing is the intelligence layer that makes distributed African energy verifiable, reliable, and bankable. That layer does not yet exist at anything like the scale the market requires. Almost nobody is building it seriously.

    Water: knowing where the thesis holds

    Poor water infrastructure costs Sub-Saharan Africa an estimated 5% of GDP annually, roughly $170 billion. That is not a humanitarian statistic. It is a macroeconomic one, and it reframes water as one of the continent’s most significant structural drags on growth. It sits alongside the energy deficit, the logistics gap, and the construction shortfall as a physical constraint on everything else Africa is trying to build.

    The physical nature of the failure is clearest in urban distribution networks. South Africa’s 2023 data show national non-revenue water at 47.4%, and critically, approximately 70% of that loss is pure physical leakage: burst pipes, corroded joints, overflowing reservoirs. Not billing errors. Not illegal connections. Pipes failing in the ground, undetected, because nobody is listening for them. A sophisticated billing platform cannot hear a leaking pipe. Acoustic sensors embedded in the network can and do. The case for physical AI in urban water infrastructure is direct and commercially clear: the utility is losing treated, pressurised, chemically dosed water at a rate that is destroying its financial viability, and the fix is physical intelligence embedded in the asset, not software layered above it.

    Agricultural irrigation is the second strong case, and it connects directly to the agriculture argument already made in this essay. Only 4% of Africa’s cultivated land is irrigated. The continent is overwhelmingly rain-dependent, which means climate volatility translates immediately into food insecurity. IoT soil moisture sensors combined with autonomous irrigation scheduling, applying water precisely when crop evapotranspiration thresholds are crossed, not on fixed human-determined schedules, directly improve yield while conserving a resource that is becoming scarcer. The farmer captures the full return on that efficiency. The incentive structure is clean. The technology exists and is deployable now.

    In rural handpump networks, the picture is more complicated and deserves honesty. Two hundred million rural Africans depend on roughly 700,000 handpumps, between 16 and 58% of which are non-functional at any given time. Smart accelerometer sensors on pump handles can measure the mechanics of each stroke, deduce aquifer depth, detect cylinder degradation, and flag maintenance needs before failure. In Kenya, a peer-reviewed study combining smart telemetry with professional maintenance response reduced pump downtime from 27 days to 2.6 days. That result matters. Notice what produced it. Not the sensor alone. The sensor plus the operational capacity to act on what the sensor found. In contexts where municipalities have no excavation vehicles, no replacement parts, and no repair budget, the intelligence generated by physical AI is economically inert. The technology is not the binding constraint. The operational system behind it is.

    This is the honest qualification the water section requires. The global public budget execution rate in the water sector is only 72%, meaning that even where money exists, a quarter of it fails to translate into functioning infrastructure. A smart sensor cannot cure municipal insolvency. Physical AI in water is genuinely transformative where the end user has both the incentive and the operational capacity to act on what the technology reveals: urban utilities fighting to recover revenue from physical losses, agricultural operations where water efficiency translates directly into yield, industrial and mining users for whom water reliability has a clear and immediate cost. In heavily subsidised, donor-dependent, or institutionally hollowed-out rural networks, the technology arrives ahead of the system that would make it useful.

    The founders who will matter most in African water are not the ones building dashboards that visualise a broken network from the outside. They are the ones embedding intelligence into the pipes, the pumps, and the soil, in environments where the operator has both the means and the motive to act on what the hardware finds. That is a large and largely unclaimed space. It is just not every space.

    Logistics: intelligence in the vehicle, not just the platform

    Logistics costs in Southern Africa run at between 20 and 60% of the landed value of goods. In OECD countries the equivalent figure is 8 to 10%. That gap, sustained across every sector, from food to mining to manufactured goods, is one of the most significant structural disadvantages the continent carries. It inflates the price of everything, erodes competitiveness, and hits the poorest hardest. Despite having a population comparable to India’s, Africa’s paved road network is only 10% the size. Only 34% of rural Africans live within two kilometres of an all-season road. Roads carry 80% of the continent’s goods and more than half of them are unpaved.

    African logistics attracts a lot of technology investment, and most of it goes into the coordination layer: platforms that match trucks to loads, aggregate freight, and build visibility into supply chains. These are useful products. They are also software sitting on top of a physical system that remains deeply broken. The research is unambiguous on where the real friction sits on the Johannesburg–Lusaka corridor, trucks spend 81% of total journey time stationary at regulatory stoppages. A significant share of that is a governance problem, involving customs procedures, border bureaucracy, and paper-based clearance, that no algorithm can fix from the outside. Yet underneath that, and inseparable from it, is a physical layer that is failing on its own terms.

    Vehicles break down because maintenance is reactive rather than predictive. Cold chains fail because the physical refrigeration is unreliable, not because nobody booked the right truck. Africa loses up to 40% of perishable food in transit, the same post-harvest loss problem that devastates agriculture, now repeating itself in the supply chain. Two working systems on this continent show what embedded physical intelligence looks like at scale. Takealot launched its first robotics deployment in South Africa in March 2025, six months from conception to live, processing 50,000 parcels a day at 99.9% accuracy. Zipline’s drone network has delivered over 1.5 million medical supplies to 4,800 facilities across Africa, cutting delivery times from hours to under an hour and, in Ghana alone, reducing medical stockouts by 60% and maternal mortality by over 50%. A software platform matching a nurse to a courier could never achieve that if the courier was stalled by a washed-out road. The drone is a physical solution to a physical constraint.

    The opportunity for African founders is in the physical layer that the platform companies are not building predictive maintenance systems for fleets operating on unpaved roads in extreme heat; cold chain sensing and management designed for environments where power is intermittent and infrastructure is sparse; drone logistics for last-mile delivery in places where roads make ground transport uneconomical. An African founder who has driven those roads, who knows what a truck goes through, who understands the local constraints around power and connectivity and theft, that founder has knowledge no imported product carries.

    The platform layer is becoming crowded. The physical layer is nearly empty. That is where the next generation of African logistics companies will be built.

    Fisheries: where the thesis is strong, and where it is not

    Software cannot freeze a deteriorating fish. An algorithm cannot oxygenate a hypoxic tilapia pond. These are not rhetorical points; they describe the precise reason why the long history of digital interventions in African fisheries has produced so little. Mobile marketplaces connecting fishers to buyers. SMS-based supply chain trackers. Blockchain traceability platforms. All of them coordinate above a physical reality that is simply broken, and none of them change the weight of anything. Africa holds less than 1% of global cold storage capacity. Between 30 and 40% of the fish caught in Sub-Saharan Africa spoils before it reaches a market. The problem is physical. The solution must be too.

    The strongest case for physical AI in African fisheries sits in inland aquaculture, and the reason is incentive structure. A fish farmer operating a controlled pond captures the entirety of whatever efficiency a physical system delivers. Feed accounts for up to 70% of aquaculture operating costs, and most of it is wasted through manual guesswork, feed sinking to the bottom of ponds because nobody measured whether the fish were hungry. IoT sensors monitoring dissolved oxygen, pH, and temperature continuously; machine vision cameras reading fish behaviour to trigger automated feeders only when appetite is genuine; solar-powered aerators preventing the overnight hypoxia events that wipe out an entire harvest in hours. AquaRech in Kenya has demonstrated that IoT pond monitoring compresses production cycles from 13 months to 8 to 10 months. M-KOPA’s solar aerators, available on a $5-per-month lease-to-own model, deliver yield increases of 30 to 40%. These are working systems, built for African constraints, priced for African incomes.

    The mid-stream cold chain is the second clear opportunity, and it connects fisheries directly to the logistics and energy arguments already made in this essay. Whoever builds off-grid, solar-powered, IoT-monitored cold storage for fish landing sites simultaneously solves the cold chain problem for high-value agriculture and pharmaceuticals. Machine vision sorting and grading systems in processing plants address the fact that 55% of Africa’s edible aquatic food loss occurs not at sea but on land, during processing. This is not a remote or futuristic opportunity. It is a current, large, and almost entirely unclaimed one.

    The honest qualification is that the physical AI thesis is considerably weaker when applied to open-access marine fisheries. Africa loses $11.2 billion annually to illegal, unreported, and unregulated fishing by foreign industrial fleets. Satellite AI can identify those vessels with precision even when they disable their tracking systems. Identifying an illegal trawler is not the same as stopping one. If the navy lacks fuel to intercept it, or if political relationships make enforcement inconvenient, the technology produces high-definition documentation of the theft rather than prevention of it. In open-access environments, the binding constraint is governance, not sensing capability.

    The technology that will secure the future of Africa’s blue economy will not live in the cloud. It will be ruggedised, solar-powered, and in the water, deployed where the incentives align perfectly with the hardware, in the ponds and cold chains where a founder who understands African constraints can build something that compounds quietly and durably over time.

    Wildlife and conservation: collapsing assets and an emerging market

    Wildlife is not a romantic sideshow to Africa’s economic story. In Botswana, wildlife tourism contributed $2.4 billion to GDP in 2023, 12.1% of the entire national economy, and supported over 30,000 jobs. Across the continent, 36.3% of all direct tourism GDP is attributable to wildlife. These are not ecosystem statistics. They are macroeconomic ones. The assets producing that value are collapsing.

    African lions have declined from roughly half a million in 1950 to fewer than 20,000 today, losing 94% of their historical range. Elephant populations at monitored sites have fallen by around 77% since 1964. These declines are happening because the systems protecting these animals are structurally overwhelmed. Sixty-five percent of ranger employers report their staffing is inadequate to secure their territory. Forty percent of rangers lack adequate communications tools. Only 51% have reliable internet access at their workplace. The people doing this work are being asked to protect areas the size of small countries with almost nothing.

    Physical AI in this context is the only credible path from reactive, forensic conservation to proactive, real-time interdiction. Sensor networks now cover 2.8 million hectares of African conservation land. AI-enabled camera traps process images at the edge and transmit only kilobyte-sized alerts when they detect a human presence, making early warning viable in areas where half the staff have no internet. Acoustic sensors triangulate the sound of a rifle shot and alert rangers to its exact location within 60 seconds. EarthRanger deployments combining GPS collars, camera traps, and ranger reports into a unified tactical map have removed more than 50,000 snares and reduced wildlife-conflict deaths by over 91% in some reserves. In parks managed by African Parks Network, sustained physical infrastructure investment has produced a 90% drop in elephant poaching.

    There is a funding angle most people in the technology community are not yet tracking. African Parks Network sites alone store an estimated 11 billion tonnes of CO2. Emerging biodiversity credit and carbon markets, projected to reach $1 billion to $2 billion by 2030, require continuous, verifiable, high-integrity data to prove that a unit of biodiversity has been protected. Manual counting methods cannot satisfy the auditing requirements of corporate buyers and global financial institutions. Physical sensing can. The founder building conservation monitoring infrastructure is not just building a conservation tool. They are building the data layer for an entirely new financial market.

    Terra Industries, a Nigerian defence-technology startup, recently raised $34 million from global investors to manufacture ruggedised autonomous drones and sentry towers locally, designed for African conditions, sourced on the continent. That is the model.

    Two honest qualifications: technology cannot compensate for absent enforcement capacity, a real-time detection alert is useless without vehicles, fuel, and institutional backing to respond. Surveillance deployed without community trust can replicate the historical injustices that drive poverty-induced poaching. Physical AI must augment community partnership, not replace it.

    Building in conservation is not a consolation prize for founders who could not crack fintech. The asset is collapsing. A financial market is emerging that needs the measurement layer only physical sensors can provide. Almost nobody is building it seriously.

    The real prize

    None of this is an argument against building software. Software matters. Digital infrastructure matters. The coordination layer of African economies needs to be built, and it is right that people are building it.

    Antaeus did not become invincible by floating above the earth. He became invincible by staying connected to it. Technology that changes the weight of things, that reduces the number of people who go to work and do not come home, that saves the harvest after it is grown, that keeps the water running and the lights on, that protects the wildlife that belongs to no one and everyone, that technology has to be physical. It must be embedded in the real world. It must touch the ground.

    The sectors that define the physical reality of African life are enormous in scale, poorly served by technology, and deeply important to the people who depend on them. They are also sectors where an African founder has genuine advantages that no outsider can easily replicate proximity to the problem, understanding of the environment, relationships with operators, and insight into constraints that someone building from London or San Francisco will never fully grasp.

    Physical AI is harder to build than software. It requires understanding of hardware, of real environments, of industries that move slowly and carry serious consequences for failure. That difficulty is also protection. The companies that build real physical intelligence into African mining, farming, construction, energy, water, logistics, fisheries, and conservation will not be displaced by the next software update from a distant competitor. They will be embedded in the physical reality of the continent in a way that compounds over time.

    The question for the next generation of African founders, students, and researchers is how to build. It is where to look.

    Put your feet on the ground. That is where the power is. That is where it has always been.