Compute Accounts for 91 Percent of Disclosed AI Commitments Tracked in Emerging Markets
Capital Flows | Accendo Signals
Signed AI capital across five emerging-market regions, 1 August to 25 September 2026.
Compute infrastructure accounts for 91 percent of the disclosed value of AI commitments signed in emerging markets between 1 August and 25 September 2026. Of USD 3.79 billion committed under executed agreements across five regions, USD 3.45 billion funds compute. The Accendo Signals Capital Flows Map records 25 signed deals in the period. Applications and services closed 15 of them, worth USD 169 million.
The headline rests on one transaction. Zankore, an AI cloud provider in Indonesia, signed a senior term loan facility of up to USD 3.1 billion with Citi, ING, Natixis, Qatar National Bank and UOB. That facility is 82 percent of all signed value. Without it, signed value falls to USD 690 million and the compute share falls to 51 percent. The period therefore holds two findings: an exceptional concentration in one infrastructure financing, and a broader market of many smaller application deals.
A signed deal here means an executed agreement for a specific, dated transaction. The figures are commitments at signing. A loan is recorded at its full facility amount, which is a ceiling on borrowing and says nothing about how much has been drawn. Ten of the 25 signed deals did not disclose an amount, so every value in this note is disclosed value. For the full dataset from January 2023, see Necessary, Not Sufficient.
The Zankore facility funds the next phase of GPU deployment for an NVIDIA-powered AI cloud platform. The second largest signed deal is a USD 300 million equity investment in WIOCC, the African digital infrastructure company, by the Africa Finance Corporation and Vision Invest. Together the two account for about 90 percent of signed value. Debt carries 86 percent of signed value across two loans, which reflects the size of the Zankore facility more than a general preference for debt.
Compute stays the largest layer once Zankore is set aside, at USD 351 million of USD 690 million. The share drops from nine dollars in ten to about one in two. Five of the eight compute deals did not disclose an amount, so the disclosed figures likely understate compute rather than overstate it. The concentration finding holds, and it holds at a scale much closer to the rest of the market.
The full record since January 2023 is far less sensitive to any single deal. Across 206 signed transactions worth USD 90.1 billion in disclosed value, compute holds 97 percent. Setting aside the largest, the USD 25 billion Stargate data center in Patagonia, Argentina, compute still holds 95 percent. The two months covered here are more fragile than the long-run pattern, and the long-run pattern points the same way.
Applications Closed the Most Deals with the Smallest Tickets
Applications and services closed 15 of the 25 signed deals, mostly early-stage rounds in Latin America and strategic investments in Saudi Arabia. The largest is USD 125 million for Kapital Grupo Financiero in Mexico. Early-stage venture capital closed eight deals worth USD 32 million between them, and grants closed three worth USD 11.5 million. Human capacity drew one signed commitment, a USD 170 million loan from South Korea’s Economic Development Cooperation Fund for an AI and digital training institute in Tanzania. Policy drew one signed deal with no amount disclosed.
We identified no separately classified signed data-foundations transaction in the tracked entries. Data work may sit inside application, research or infrastructure budgets, so the absence of a data-foundations row does not mean no money went to data.
Capital intensity differs across layers. A data center costs more than a software company at seed stage, so a low value share for applications does not by itself show that applications are underfunded. The stronger question is whether applications, skills and data receive enough financing to turn new infrastructure into services that people and businesses use.
The Evidence Raises a Different Question for Each Type of Investor
The signed figure shows where someone has taken on risk, on what terms and with whom. The shares above describe where signed money is committed. They do not prove where money is missing. They do point each group of investors to a specific line of inquiry.
Demand behind the capacity
Announced capacity is large. The diligence questions are demand, utilization, power availability and the financing terms behind each facility.
Where catalytic capital changes the model
Local data, implementation skills and adoption support barely register in signed deals. It is worth testing where these lack a viable funding model, and where blended or catalytic capital would change that.
From activity to repeatable revenue
Fifteen small deals signal activity. What matters next is whether these companies reach repeatable revenue and raise follow-on rounds.
| Region | Deals signed | Signed value | Not signed | What drives it |
|---|---|---|---|---|
| Southeast Asia | 3 | USD 3.1 bn | USD 34.5 bn | Zankore loan; Philippines AI+ Infrastructure Masterplan (USD 34.4 bn target to 2033) |
| South Asia | 4 | USD 12 m | USD 17.7 bn | Three Indian data center campuses (two in Telangana, one in Puri, Odisha) and a proposed National Frontier AI and Compute Fund |
| Latin America and the Caribbean | 9 | USD 206 m | USD 12.0 bn | Kapital Grupo Financiero; CAF digital transformation agenda |
| Sub-Saharan Africa | 3 | USD 470 m | USD 1.7 bn | WIOCC equity and Tanzania training loan; Mombasa data center and LNG plant |
| Middle East and North Africa | 6 | USD 1.5 m | USD 1.0 bn | Five undisclosed strategic investments; Egypt sovereign AI data center target |
Across Sub-Saharan Africa and Egypt, recorded signed value is USD 470 million against about USD 2.7 billion unsigned, a gap of about six to one against 18 to one overall. The signed and unsigned entries are different projects on different timelines, so this ratio is not a conversion rate. Stratos Lab’s GPU cloud deployment in South Africa, with ECOBLOX and Digital Parks Africa, is the third signed African deal and did not disclose an amount.
More than half of the unsigned African value pairs data centers with their own power supply: a USD 1.5 billion data center and LNG plant in Mombasa backed by AMACO Energy Group and GE Vernova, and a USD 75 million data center and power plant in Douala by Cybastion. The U.S. International Development Finance Corporation signed a letter of intent for USD 155 million in WIOCC, alongside the Africa Finance Corporation investment.
In the Middle East and North Africa, five of six signed deals are strategic investments by HUMAIN and White & Case with no amount disclosed. Traced value in that region understates activity.
Philanthropy Committed USD 160 Million to AI for Smallholder Farmers
Two philanthropic commitments in the period target agriculture. Google and the Gates Foundation committed USD 100 million to extend AI tools for smallholder farmers across Sub-Saharan Africa and South Asia. The OpenAI Foundation committed USD 60 million to AI-powered weather and crop disease forecasting through six partners, with work across South Asia, Southeast Asia and East Africa. The USD 10 million grant to AIM for Scale is part of that USD 60 million and is not additional.
Both larger commitments are recorded as committed envelopes: the money is committed but not yet tied to one transaction. They sit in the USD 980 million envelope figure and outside the signed total. The USD 10 million AIM for Scale grant is the first signed award under the OpenAI program and is counted in the signed total. Signed and envelope figures are never added together, so the overlap does not inflate either one.
Eight Plans Make Up Almost All of the USD 67 Billion Not Yet Signed
Announced plans point mostly toward compute. Of USD 67 billion in targets and letters of intent, about USD 55 billion (82 percent) is for data centers, compute funds and the power to run them. The largest single plan is the Philippines AI+ Infrastructure Masterplan at USD 34.4 billion. A further USD 12 billion is the digital transformation agenda of CAF, the development bank of Latin America and the Caribbean, which covers AI, data centers, high-performance computing and cybersecurity. CAF has not stated the AI or compute share of that agenda, so we record the full amount and treat its compute share as an upper bound.
Adding every category together would put these two months at about USD 72 billion, against USD 3.79 billion committed to specific transactions. A target can be revised, delayed or dropped without anyone breaking an agreement, and headline amounts are often ceilings spread across many years.
Eight entries make up USD 66.6 billion of the unsigned total. Seven are targets and one is a letter of intent. All eight came in through the September sweep and are still under review, so the total may move once duplicates and restatements are resolved.
Many unsigned plans are multiyear programs announced recently, and they may be proceeding on schedule. Where a plan stalls, the cause is often a missing financing structure, power supply, offtaker or guarantee. That is where blended and development finance can help turn an announcement into a transaction.
The question for the next two quarters is conversion. We will track the same entries and report how many of the eight large plans reach signed financing.
Conversion, Disclosure and the Layers Above Compute Deserve Attention
Which plans reach signed financing
We follow the same entries over time. A signed loan or equity close moves an entry into the signed total, and only then do we report it as converted.
Amounts behind strategic deals
Ten of 25 signed deals carry no amount, most of them strategic investments in Saudi Arabia. We will add amounts as they are published.
The layers above compute
Compute is being financed. We will track whether the skills, data and adoption support needed to use it attract signed capital in the coming months.
Compute is necessary for AI in emerging markets, and it is being financed at scale. The next test is whether the layers that turn compute into useful services follow. We invite investors, funders and founders to use the Capital Flows Map, now available in English, Spanish and French, and to tell us about signed deals we have missed.
Every entry in this note, with its source, is on the Capital Flows Map. Filter by region, stack layer, investment type and sector, or read the full breakdown on the analysis page.
How the Numbers Were Counted
An executed agreement
Commitment status Committed, record type Deal. An executed agreement exists for a specific, dated transaction; its terms may not be public. The 25 signed deals are 8 in compute, 15 in applications and services, 1 in human capacity and 1 in policy.
Committed, destination open
Funds, corporate country programs and national allocations that are committed but not tied to one transaction (USD 980 million across 14 entries). Excluded from the signed total.
Intention without obligation
Targets, letters of intent and memoranda of understanding (USD 67 billion across 14 entries). A letter of intent or memorandum is classed by the evidence of a financing obligation in its text, not by its title. Excluded from the signed total and reported separately.
- Scope. Entries with an announcement date from 1 August 2026, recorded on the map as of 25 September 2026. This is a September-to-date update; entries dated after 25 September will appear in the next edition. Rows held as duplicates are excluded.
- Reconciliation. The period holds 61 entries: 25 signed deals, 14 committed envelopes, 14 announced but not signed, and 8 child entries. The child entries (USD 761 million) are already represented within a parent entry and are not counted again. The USD 761 million is not missing financing.
- Values. Commitments at signing, converted to US dollars on the exchange-rate basis recorded for each entry. Loans are recorded at the full signed facility amount, not the amount drawn. Shares use disclosed value only.
- Status. 36 entries are published and 25 are under review. Thirteen of the 14 unsigned entries, and USD 66.8 billion of the USD 67 billion, are under review, so the unsigned figures are provisional.
- Rounding. Figures are rounded for reading. Percentages are calculated on unrounded values.
