Diagelo
The Quarterly · Q2 2026 Subscribe
The Diagelo Quarterly · Q2 2026

Halfway Through 2026

What the first six months of 2026 reveal about the world we are actually building.

13 minutes Foresight · Intelligence · Ecosystem Published July 2026
The Continent, First Half of 2026
Six layers of the same year. Toggle any layer to read the quarter differently.
Cover
Northern CorridorCentral CorridorAbidjan to LagosMaputo CorridorTrans KalahariSuez capital seeking a landing SudanCentral SahelEastern DRCSomalia Uganda · JanCongo · MarBenin · AprDjibouti · AprCabo Verde · MayEthiopia · JunSouth Africa · local, later 2026 JohannesburgCape TownNairobiLagosCairoCasablancaAccraKigaliDakar
everything here has an address. Diagelo Intelligence, drawing on AfDB, IEA, SEMI, SIPRI and national electoral calendars, July 2026.
Editor's Letter

The Return of Where

For thirty years, location was treated as a solved problem. The first six months of 2026 were an extended argument against that assumption.

The idea had a founding text: in 1997 Frances Cairncross gave the era its slogan, The Death of Distance. She was half right, the most expensive way to be wrong. Distance died for the bit, the email crossing the planet overnight. It never died for the electricity, the land and the water those bits demand. The first half of 2026 was the invoice, arriving thirty years late.

The half-year’s biggest numbers all point one way. The IEA expects global electricity demand to grow 3.6 percent a year through 2030, adding roughly 1,100 terawatt hours annually, all landing somewhere specific: a grid, a substation, a permit office. Chip-tool sales are heading for a record 165.9 billion dollars in 2026, sited for reasons unrelated to cost. SIPRI counted 2,887 billion dollars of military spending in 2025; defence is the most literal case of the instinct, money to hold ground.

It all flows from that. AI turns out to be a physical industry in a software costume: it needs power, cooling, land, fibre, silicon and people, all with addresses. Picture a large data centre not as a computer but as an aluminium smelter that happens to think, turning cheap power into something you can sell, planted where the electricity is, not the people. Industrial policy became an operating system: a state that believes location matters will pay to own the locations that count.

Africa’s quarter carried the same signature. The African Development Bank released its 2026 outlook in Brazzaville in May, projecting 4.2 percent continental growth. The headline conceals more than it reveals: twenty-two economies grew above 5 percent in 2025, and twelve of the world’s twenty fastest-growing economies were African. The continental average describes almost none of them.

Washington said the same from the other side. AGOA lapsed in September 2025, was restored by law in February 2026 with retroactive effect, and now expires again on 31 December. The preference survived; the predictability did not. South African vehicle exports to the United States collapsed by roughly three quarters, from 25,544 units to 6,530. A one-year renewal tells a plant manager near Gqeberha that access to America’s market is now an annual political call.

The binding constraint was gloriously unglamorous: not the model, not the chip, not the engineer shortage everyone feared, but the interconnection queue, the years-long wait to plug new load into the grid. The most futuristic industry of the decade is rationed by one of its most nineteenth-century bottlenecks: the wire. Ask me where the boom will be settled, and I have stopped pointing at laboratories and started pointing at substations.

One version of this curdles into fatalism: geography as destiny, the map already decided. It is wrong, and the evidence is in the same data. South African automakers lost three quarters of one market and still closed 2025 with record total exports of 414,268 vehicles, because they found new buyers faster than most expected. Geography constrains which routes are open; it does not decide which one an operator takes.

Terrain raises the cost of being wrong about place. For three decades an organisation could hold a mediocre view of place and survive on margins; that grace period has closed. The institutions that came through in shape knew which grid, which port, which regulator, which election. Which brings us to the question this issue is built around: not where the world is going, but where you actually are. The first is a forecast. The second is a matter of record, and few institutions can state theirs precisely.

Founder, Diagelo · Johannesburg · July 2026
The Half-Year in Brief

Everything has an address again

Clear the noise from the last six months and one thing is left standing: place matters again. Power, chips, land, the permit, the fibre in the ground, the engineers who keep the models fed. For thirty years we were told location was a detail you sorted out later. This year the detail sent a bill.

back to maps, not slogans.

The AI boom turned out to be a building project, not a coding one. Money crowded into the few sites a government has promised to defend, and paid up to sit there. And that tidy 4.2 percent everyone quotes for Africa fits almost no country you could actually book a flight to. Geography did not go quiet. It started charging rent.

place sends a bill.
first question: where, exactly?
The question we keep coming back to where the world is going, a forecast where you actually are, a matter of record
Every stage of the new economy stands on ground
someone owns. Height here is where value collects.
Five Signals That Matter

The quarter, read against the grain

Five readings of the first half of 2026. Each earns its place by changing what you do next, not by being interesting at a dinner.

not dinner-party trivia.
01

Artificial Intelligence Has Geography

The race is no longer decided in model architecture alone. It is decided by electricity, silicon, land and people, and every one of those inputs has a location and a landlord.

Data centres consumed roughly 485 terawatt hours in 2025 and the IEA expects that to reach about 950 by 2030, around 3 percent of global electricity. In the United States, data centre expansion is expected to account for about half of all demand growth to 2030. The scarce input is no longer the chip but the substation feeding it, and the questions worth asking now are physical ones. Where is the capacity being built? Who owns the compute once it is built? Who controls the electrons that feed it, and under whose law?

Africa holds under 1 percent of global data centre capacity while the Africa Data Centres Association puts near term requirement at a minimum of 1,000 megawatts across roughly 700 facilities. Kenya is tracking a projected 30 percent compound growth in IT load through 2028. The bottleneck sits upstream of demand.

Compute is rationed by grid connection now, and grid connection is a government's to grant.
ImplicationArtificial intelligence is becoming physical.
How much power will a thought cost by 2030?
48595020252030fTWh
Global data centre electricity consumption. Source: IEA, Energy and AI, 2026.
02

Industrial Policy Is No Longer Temporary

Subsidies, tariffs, local manufacturing requirements and export controls were once framed as crisis measures with a sunset attached. They now behave like permanent architecture, and firms have begun to budget for them as fixed cost.

SEMI attributes the current build cycle to two forces working together: demand for AI silicon, and a commitment across every major region to domestic manufacturing capacity through localised supply chains. Governments are underwriting fabs, grids and ports because they have concluded that owning the physical layer is worth the fiscal price.

AGOA is the clearest African expression of the shift. Restored by law on 3 February 2026 with retroactive effect to 30 September 2025, it now expires on 31 December 2026. A preference that ran on fifteen year horizons at inception now resets annually, which converts a trade programme into a recurring negotiation. Market access has become a policy variable that moves faster than a factory can be depreciated.

ImplicationMarkets are becoming political objects.
Fifteen years of certainty, cut to one.
200015 yr201510 yr20261 yr
Length of each AGOA authorisation at the point it was granted. Source: USTR, Congressional Research Service, 2026.
03

Capital Is Becoming More Selective

Money has narrowed its aperture. Infrastructure, defence, compute and energy are absorbing capital at scale while everything outside those four lanes is finding the terms harder and the diligence longer.

SIPRI counted 2,887 billion dollars of world military expenditure in 2025, an eleventh consecutive rise, taking the global military burden to 2.5 percent of GDP, its highest since 2009. Three countries accounted for 51 percent of it. SEMI expects the chip equipment market to set a record 165.9 billion dollars in 2026 and to reach 229.5 billion by 2028. Concentration is the point: capital is flowing toward assets that governments have promised to protect.

The corollary lands hardest where the AfDB set its marker in May. A continental financing gap above 1.3 trillion dollars a year sits alongside average public debt near 63 percent of GDP, with interest payments consuming close to 15 percent of government revenue in many countries. The same reallocation that concentrates capital abroad is what widens the financing gap at home.

Capital is flowing toward the assets governments have promised to defend.
ImplicationNot every market benefits equally.
Half the world’s defence budget, three addresses.
51%top three spendersof $2,887bn spent in 2025
Share of world military expenditure held by the United States, China and Russia. Source: SIPRI, April 2026.
04

Africa's Story Is Becoming More Divergent

The single continental number is now actively misleading. Africa is projected to grow at 4.2 percent in 2026, and almost no African economy will grow at 4.2 percent.

East Africa is expected to reach 5.9 percent this year, easing from 6.6 percent as energy and import costs bite, with a rebound to 6.4 percent projected for 2027. West Africa holds near 4.7 percent on agricultural output and infrastructure spending. Central Africa lifts to 3.8 percent on oil receipts. Underneath the regional averages the spread is wider still: 22 economies above 5 percent in 2025, six above 7 percent, and twelve African economies inside the world's twenty fastest growing.

Four distinct trajectories are now visible, and they demand different behaviour. Reforming economies are buying credibility with fiscal pain. Commodity exporters are riding a price cycle they do not set. Manufacturing hubs are exposed to a tariff regime that resets annually. Technology ecosystems are competing for compute and talent against Dublin and Dubai. Grouping these under one heading produces decisions that fit none of them.

Reforming economies, price takers, manufacturing hubs, technology ecosystems. One continent, four trajectories.
ImplicationThere is no single African trajectory.
One continent, or four different speeds?
East5.9West4.7Africa4.2Central3.8real GDP growth, percent, 2026 projection
Source: African Development Bank, African Economic Outlook 2026, Brazzaville, May 2026.
05

Positioning Has Overtaken Prediction

Uncertainty has not resolved, and the organisations that came through this half year intact were not the ones with the best forecasts. They were the ones whose decisions survived being wrong.

The pattern repeats across sectors. Build options before you need them. Hold more resilience than the spreadsheet justifies. Shorten the interval between noticing and deciding. None of that requires knowing what happens next. All of it requires knowing what you would do under three different versions of next, and having pre committed the first move under each.

The South African vehicle sector is the cleanest illustration of the quarter. Exports to the United States fell by roughly three quarters in 2025. Total exports still reached a record 414,268 units, because the alternative buyers had been cultivated before the tariff arrived. No one forecast the specific shock; the substitute demand was already in place when it landed.

ImplicationPrecision is worth less than optionality.
Committing the move before the path is clear.
todayhorizonprice the options, skip the path
Diagelo Foresight schematic. Each node is a decision pre committed under a defined condition.

Key Takeaways

  • The organising fact of the first half of 2026 is the return of place. Power, silicon, ports and permits all have addresses, and the addresses are now priced.
  • Artificial intelligence has become an energy and land question. Data centre consumption roughly doubles to about 950 TWh by 2030, and grid access, not model access, is the constraint.
  • Industrial policy has hardened into permanent architecture. AGOA's one year horizon shows that duration itself is now the lever.
  • Capital is concentrating into assets that states have promised to defend: infrastructure, defence, compute, energy. Everything outside those lanes pays a wider spread.
  • Africa's 4.2 percent continental projection describes almost no African economy. Four distinct trajectories now require four distinct sets of decisions.
  • Organisations that held options outperformed organisations that held forecasts. South African vehicle exports lost three quarters of the United States market and still set a record.
Twenty-two African economies grew above 5 percent in 2025. The single continental average now conceals more than it reports.
Signal Four · Diagelo Foresight
Graphic of the Quarter

The Geography of Artificial Intelligence

Five dependencies stand between an idea and a working model. Each one has an owner, a jurisdiction and a queue. The chain narrows at every step.

gatekeepers all the way down.
Diagelo Foresight

What a model is standing on

Read left to right. The width of each band is the width of the funnel it feeds.

ELECTRICITYSILICONDATA CENTRESCLOUD AND MODELSTALENT AND DEMANDSTAGE 01Electricity950 TWhdata centre demand by2030, roughly double2025 and about 3% ofworld consumptionGATEKEEPERNational grids, waterrights, interconnectionqueues, tariff regulatorsSTAGE 02Silicon$165.9bnrecord chip equipmentsales forecast for 2026,heading to $229.5bn by2028GATEKEEPERFour regions, a handfulof toolmakers, exportcontrol regimesSTAGE 03Data centresunder 1%of world data centrecapacity sits in Africa,against a stated need of1,000 MW over 700 sitesGATEKEEPERLand, permits, fibre,hyperscalers and localoperatorsSTAGE 04Cloud and models$320bncommitted to capitalspending in 2025 by fourUnited States platforms,up from $230bnGATEKEEPERPlatform balance sheets,data residency law,procurement termsSTAGE 05Talent and demand30%projected compoundgrowth in Kenyan IT loadthrough 2028, one of thefastest rates trackedGATEKEEPERUniversities, visas,wage levels, latency tothe customer
01

Every stage in this chain is owned by somebody, and none of the owners are software companies. Power, silicon, land and interconnection all answer to a government, a landlord or a queue.

02

The chain narrows fastest where Africa is thinnest. Holding under one percent of world data centre capacity is not a demand problem, it is a position four steps upstream of demand.

Every gatekeeper in this chain is a government, a landowner or a queue. Diagelo Foresight. Data: IEA on demand, IEA on platform capital spending, SEMI, Africa Data Centres Association, 2026.
Three Charts Worth Studying

Three charts we could not argue our way out of

Chart One

The sliver everyone is arguing about

World electricity consumption moves from 28,200 terawatt hours in 2025 to 33,600 by 2030, roughly 1,100 added each year against an average of 700 over the previous decade. Data centres are a small share of that total and a large share of the argument: 485 terawatt hours in 2025, about 950 by 2030.

Why it matters. The share stays near 3 percent, which makes the total the wrong object to watch. Concentration is the variable. That demand arrives in dense clusters, on specific substations, over a five year horizon, against grid assets that take a decade to build. A 3 percent national number can still be a 40 percent local one. Grid planning, not generation totals, is where this gets settled.

Where does all the new demand have to land?
010,00020,00030,000Global electricity consumption, TWh28,200202533,6002030 forecastThe bright cap is data centres485 TWh now, about 950 by 2030data-centre band shown at 6× scale+5,400 TWh by 2030Source: IEA, Electricity 2026 and Energy and AI, 2026. Volumes drawn to scale.
Chart Two

The industry that forgot how to bust

Worldwide 300mm fab equipment spending crossed 100 billion dollars for the first time in 2025 and is forecast to rise 18 percent to 133 billion in 2026, then 14 percent to 151 billion in 2027, reaching 172 billion by 2029. SEMI builds the series from 404 tracked facilities and lines.

Why it matters. Semiconductor capital expenditure has always been a boom and bust industry that punished anyone who read a peak as a plateau. A five year ramp with a shallow 2028 pause has the profile of an infrastructure programme, because a meaningful share of it is underwritten by governments buying self sufficiency at close to any price. That changes who the marginal buyer is, and buyers with fiscal motives do not exit at the trough.

Did the chip cycle forget how to fall?
050100150300mm fab equipment spending, US$ billion10720251332026f1512027f1552028f1722029f2028 pauses at 3%Five consecutive years. SEMI tracks 404 facilities and lines to build the series.Source: SEMI 300mm Fab Outlook, 1 April 2026.
Chart Three

Four percent, and the plane it has to clear

The African Development Bank projects continental growth of 4.2 percent in 2026, easing from 4.4 percent in 2025 and recovering to 4.4 percent in 2027. Population continues to expand at roughly 2.4 percent a year. The band between the two is what actually reaches a household, and the Bank measured per capita growth at 1.9 percent in 2025.

Why it matters. A 4 percent economy growing against a 2.4 percent population is a respectable performance and nowhere near enough to move poverty at speed. The margin compounds slowly in the right direction and unforgivingly in the wrong one: a single point of growth lost to a tariff, a drought or a debt service payment takes almost half of it. This is the chart that makes the 1.3 trillion dollar financing gap a structural fact.

Is four percent enough to clear the gap?
02%4%6%Real GDP growth, percent4.4%20254.2%2026f4.4%2027fpopulation growth, about 2.4% a yearEverything below the violet plane is absorbed by more people. 1.9% per head reached a household in 2025.Source: AfDB, African Economic Outlook 2026; UN population estimates. Per capita growth measured at 1.9% in 2025.
Grids are local. So are the permits, the water rights and the votes that decide where new load is allowed to land.
Chart One · Diagelo Foresight
Books, Reports and Papers

Five documents that earned their weekend

For each, a line on what it should change in how you hold the problem, not just that it is worth your time.

African Development Bank · Brazzaville · May 2026
The take

Published under a theme about mobilising African development financing at scale in a fragmented world, which is the Bank saying plainly that the external financing model it spent two decades assuming is gone. The 1.3 trillion dollar annual gap is the number everyone quotes. The more useful passage is the regional decomposition, which shows that the continental average has stopped being a decision input. Read it as a map of four Africas, and read the debt chapter twice.

International Energy Agency · 2026
The take

The pairing matters more than either document alone. Electricity 2026 establishes that demand growth is accelerating to 3.6 percent a year through 2030, half again the pace of the previous decade. Energy and AI establishes where the incremental load lands. Together they retire the idea that energy policy and technology policy are separate portfolios. If you own a grid, you now own an AI position whether you sought one or not.

July 2026 · Intelligence Brief · Diagelo
The take

Our own reading of the 2024 to 2025 election supercycle, when 1.6 billion people voted, Africa cast few of those ballots, and now lives with nearly all of their consequences. The settings that matter most, trade, aid and the price of money, were reset in foreign capitals inside the same window.

May 2026 · Systems Brief · Diagelo
The take

Why the continent keeps producing world-class builders while other economies capture the ownership, and where the value actually compounds. Read it against the compute story in this issue: the machines arriving in Johannesburg and Nairobi are a test of exactly this thesis.

Revisiting Our Thinking

Six Months Later

Few institutions publicly audit their own published work. We intend to do it every quarter, on the record, including the parts that did not hold.

Diagelo Intelligence · published September 2025

The Tariff War Nobody Could Ignore

Revisited July 2026, against ten months of outcomes.
What held

We argued that the tariff regime would outlast the administration that built it, because the coalition benefiting from it now spans both parties and several ministries. Nothing in the first half of 2026 contradicts that. The January vote to renew AGOA for three years was cut to one in the Senate, in line with a demand that the programme be rebuilt around reciprocity. The direction of travel was correct.

Where we were right

We wrote that the damage would arrive through uncertainty ahead of the tariff schedule itself, and that the lapse window would do more harm than the rates. TRALAC put AGOA exports down 32 percent in the year to November 2025, before the retroactive restoration. Orders cancelled during a lapse do not return when the law is backdated. That mechanism worked exactly as described.

Where we were surprised

We underestimated substitution badly. South African vehicle exports to the United States fell from 25,544 units to 6,530, close to our downside case. We then expected total exports to fall with them. They rose almost 6 percent to a record 414,268 units. Buyers were found faster than we had modelled. We priced the loss of the market correctly and missed the speed of substitution.

32%
Fall in AGOA exports, year to November 2025.
6,530
South African vehicles exported to the United States in 2025, from 25,544.
414,268
Total South African vehicle exports in 2025, a record.
1 yr
Length of the AGOA renewal signed on 3 February 2026.

Key Takeaways

  • The thesis held on direction and mechanism: tariffs became structural, and the lapse did more damage than the rate.
  • The thesis failed on adaptation speed. We priced the loss of a market and not the presence of alternatives already cultivated.
  • The correction we are carrying forward: model the second door. Any forecast of a shock that ignores existing optionality will overstate the damage.
Questions We Are Watching

Five files we cannot close

We publish the questions we cannot yet close, alongside the evidence we are watching to settle each one.

01
Will AI infrastructure create a new class of geopolitical winner?
Watching: interconnection queues, sovereign compute announcements, where the second wave of campuses lands once the obvious sites are full.
02
Can South Africa hold its reform momentum through a local election year?
Watching: the reform pipeline against the electoral calendar, grid availability, and whether the record export year proves repeatable.
03
Will tariffs become a permanent feature of trade, or a bargaining posture?
Watching: what happens to AGOA before 31 December 2026, and whether reciprocity talks with African capitals are actually convened.
04
Where does global capital go once defence, compute and energy are fully bid?
Watching: spreads on African sovereign issuance, domestic pension allocation, and whether the 1.3 trillion gap attracts anything at scale.
05
How fast can energy systems actually adapt to concentrated AI demand?
Watching: time from application to energisation, behind the meter generation, and the first jurisdiction to refuse a campus outright.
Backdating the law repaid the tariff, not the payroll.
Revisiting Our Thinking · on the AGOA lapse
Looking Ahead

What we are building next

Systems

Artificial Intelligence Has Geography

The full treatment of the chain sketched in this issue. Where compute is being built and why there, who owns the electrons, which jurisdictions can compel what, and what an African position in that chain would have to cost before it could be called real.

Next up · Systems brief
Foresight

South Africa Before Tomorrow

A full study of the reform decade as it is actually being lived: what has moved, what was announced and did not, and what the country looks like on the grid, the ports, the fiscus and the demographic clock as the decisive years arrive.

Next up · Foresight study
The Half-Year, in One Picture

A computer with an electricity bill

Strip off the software and a data centre is an aluminium smelter that happens to think: a machine for turning cheap, dependable power into something you can sell, planted where the electricity is. It runs on a grid it does not own, on land someone permits, over fibre someone leases. Hover each layer to meet its landlord.

THE GRID POWERthe grid operator LAND & PERMITSa municipality FIBRE & PORTSa carrier SILICONa foundry HOVER A LAYER · MEET ITS LANDLORD WHERE YOU ACTUALLY ARE THE CLOUD has an electricity bill it thinks. and it still pays rent. held up byone wire.
Power, land, fibre and silicon each answer to a grid operator, a municipality, a carrier or a foundry. Distance did not die; it started sending an invoice. Diagelo Foresight
From the desk

A last note before you close the file.

The number I keep coming back to appears on none of these charts. It is the two years a grid connection takes. The most futuristic industry of the decade, held up by a wire, a queue and a permit clerk who has never heard of your model. I find that oddly steadying. It means the edge still belongs to whoever pays attention early: to the substation, the port, the regulator, the second buyer found before the tariff. Distance did not die. It just started sending an invoice, and the people who kept their footing had already paid it.

The EditorJohannesburg

The future rarely arrives all at once.

It appears through incentives changing, institutions adapting and capital moving, usually some time before anyone announces it. By the time a shift has a name, most of the decisions that produced it have already been made, in permit offices and budget committees and board meetings that nobody covered.

That is what this half year has been. No single event explains it. A thousand small relocations of belief, all pointing the same direction, all agreeing that where a thing happens has consequences again.

None of it was announced in advance. What separated the institutions that kept their footing from the ones that did not was earlier commitment, the grid connection applied for two years ago, the second buyer cultivated before the tariff, the permit already sitting in the queue. Geography had re-priced the cost of holding a mediocre view of place. The advantage went to whoever had bought their options before that price moved.

Write to us

The Diagelo Quarterly is published as a public good. Disagreement, correction and better data are all welcome, and the revisiting section exists to carry them.

hello@diagelo.com Subscribe to the Quarterly