Global AI investment in 2026 will reach approximately $1.019 trillion — roughly $200 billion more than the hyperscaler capital expenditure figure the market routinely cites — according to new Goldman Sachs research that argues the standard metric is structurally flawed in four separate ways.
The recalibration matters for African readers for a specific reason. The commonly repeated finding, from the Lawyers Hub and Agence Française de Développement report launched at the Africa Forward Summit in Nairobi, is that Africa captures an estimated 2.5% of global AI investment. That percentage was calculated against a smaller denominator. If Goldman is correct that global AI investment is $1.019 trillion rather than the widely cited $794 billion, Africa’s actual share of global AI capital is meaningfully lower than 2.5% — and the gap between continental ambition and continental capital is wider than the headline figure suggests.
Four flaws in the standard metric
Goldman economists Joseph Briggs and Sarah Dong argue in a Global Economics Analyst report that the roughly $794 billion hyperscaler capex figure simultaneously understates global AI investment by about $200 billion and overstates US domestic investment by a similar amount.
The report identifies four structural problems with using hyperscaler capex as a proxy for AI investment. First, it ignores investment from US private companies — which Goldman’s credit team data suggests matters considerably, with hyperscalers directly accounting for only 40% of AI-related supply in 2026 — as well as capex from other listed companies. Second, it omits non-US companies entirely, particularly in China and elsewhere in Asia. Third, hyperscaler capex already exceeded $150 billion before the AI boom began, meaning a portion of current spending is unrelated to AI. Fourth, US hyperscalers operate globally, and a significant share of their capex is deployed outside the United States.
After adjusting for all four, Goldman puts US domestic AI investment in 2026 at approximately $581 billion, with the global total at $1.019 trillion.
Three methods, consistent answers
Goldman ran three independent estimation methods and found them converging. The primary enhanced-capex approach yields $1.019 trillion. A second method, tracking revisions in gross margin forecasts for AI-related listed companies against a 2022 baseline, produces approximately $1.06 trillion. A third, built from official national accounts and global trade data, gives approximately $1.002 trillion.
That third method is worth noting for its granularity: US national accounts show annualised AI-related hardware investment reaching approximately $463 billion as of May 2026, with roughly $100 billion in AI-related R&D and intellectual property investment on top, bringing the US annualised total close to $600 billion. Averaging the three methods, Goldman estimates cumulative global AI investment from 2022 through end-2026 at $1.8 trillion.
On geographic distribution, Goldman allocates hyperscaler capex based on announced project locations, estimating roughly 70% flows to US domestic projects, 15% to Asia and 9% to Europe. The remaining 6% covers everywhere else — a category that includes Africa, and which underlines how concentrated global AI infrastructure investment remains.
Rising share of GDP, with limits
Goldman projects US AI capex rising from 1.8% of GDP in 2026 to 2.5% in 2027 and 2.8% in 2028. Globally the figures are 0.9%, 1.3% and 1.4%. Those levels sit within the 2% to 5% peak investment shocks observed during historical general-purpose technology build-out cycles — meaning that even with substantial upward revisions to 2027 forecasts, AI investment would remain within historical range rather than constituting an unprecedented bubble.
The report also flags two cautions. Cost inflation is eroding real gains: 8% of the nominal increase in US AI-related hardware spending so far in 2026 is attributable to inflation rather than real investment expansion. And AI investment’s contribution to US GDP is structurally constrained by measurement conventions — national accounts don’t count semiconductor purchases as investment goods, and the high import content of AI hardware is netted out of GDP calculations.
Goldman recommends tracking a dashboard of leading indicators to spot deceleration, including semiconductor manufacturing equipment imports from Taiwan and South Korea, related PMI sub-indices, import prices, and memory procurement and GPU rental prices. All currently sit in elevated ranges not seen since 2022, though Taiwanese and South Korean trade data points to possible moderate slowdown in June and July.
What it means for the continent
The African relevance runs beyond the share calculation. Every projection iAfrica has covered about AI’s potential contribution to African economies — the African Development Bank’s $1 trillion in additional GDP by 2035, the GSMA’s $2.9 trillion by 2030, the IMF’s finding that faster adoption could lift African GDP by 4% — assumes a level of AI infrastructure the continent does not currently have. Africa accounts for roughly 1% of the world’s data centres. McKinsey projects continental capacity needs to rise from 0.4 gigawatts to as much as 2.2 gigawatts by 2030, requiring $10 billion to $20 billion in new investment.
Against a $1.019 trillion global investment pool, that $10-20 billion requirement is between 1% and 2% of a single year’s global AI capex. The constraint has never been that the money doesn’t exist globally. It’s that almost none of it currently flows to Africa.
