By Boakye Stephen, Kumasi, Ghana | Reporting for Ghanaian News Canada
Mr Simons, Vice-President and Head of Research at the IMANI Centre for Policy and Education, raised the concerns following a discussion on Joy News’ Digital Economy Forum on AI and the Ghana Economy.
In a post on X, he pointed to figures in the strategy that reportedly list the country’s computing-capacity targets as “1021 FLOPS” for 2027 and “1025 FLOPS” for 2035.
He argued that the figures may have been intended to represent 10^{21} and 10^{25} floating-point operations per second (FLOPS), rather than the literal numbers printed in the document.
FLOPS measure how many floating-point calculations a computing system can perform each second. The distinction between an ordinary number and a number expressed as a power of ten is substantial, making accurate notation important when setting national technology targets.
Mr Simons said the apparent discrepancy raised questions about the technical review conducted before the strategy was approved and publicly launched.
Cost estimates raise feasibility questions
Using his interpretation of the 2027 target as 10^{21} FLOPS, Mr Simons estimated that achieving the proposed computing capacity could require approximately 200,000 advanced Nvidia chips, around 500 megawatts of electricity and an investment of roughly US$24 billion.
He also argued that the US$250 million artificial intelligence data center being procured would account for only a small fraction of the computing target.
These figures are estimates presented by Mr Simons. Their accuracy depends on technical assumptions, including the type of hardware, computing performance, infrastructure design, power efficiency and the precise definition of the government’s target. They would require independent technical and financial verification.
The analyst also questioned wording in the strategy concerning an annual energy-capacity increase of “0.5 to 1 OOM”. He interpreted OOM as referring to orders of magnitude and argued that the language could imply an exceptionally large expansion in electricity output.
The government would need to clarify the intended meaning of the expression before the energy target’s implications could be assessed reliably.
Mr Simons criticised what he described as insufficient critical scrutiny of the strategy during its preparation, Cabinet consideration and public launch. He linked the issue to his broader concerns about government projects proceeding without adequate technical and financial examination.
GhanaChat and the national AI agenda
The strategy includes plans for GhanaChat, a proposed AI platform intended to serve Ghanaian users and address local needs. The initiative is expected to draw on information from 48 public-sector registries and be trained or hosted through a new AI data centre.
The proposed data centre and an accompanying AI Fund have been estimated at approximately US$270 million, with additional expenditure anticipated for registries, applications and related services.
The programme is being pursued by the Ministry of Communication, Digital Technology and Innovations under Minister Sam George as part of Ghana’s ambition to become a significant artificial intelligence player in Africa.
However, questions remain about the appropriate scale of the initial investment, the country’s computing infrastructure and how the proposed system would be financed and maintained over time.
Alternative approach proposed
Rather than immediately developing a large foundational AI model from scratch, Mr Simons suggested that Ghana could begin with an AI assistant designed for the civil service.
He estimated that such a system could operate with approximately 144 graphics processing units and cost about US$13 million to establish, with annual operating expenses of around US$6.7 million.
He also proposed using existing open-weight AI models as a way to develop useful public-sector applications while retaining greater control over government data.
Open-weight models are AI systems whose model parameters are made available for others to use or adapt, subject to their respective licences. Their use does not automatically guarantee data security or technological independence; those outcomes also depend on deployment, governance and cybersecurity arrangements.
The competing approaches raise questions about whether Ghana should prioritise a large national computing infrastructure immediately or build capacity incrementally around defined public-sector needs.
COMMENTARY
Ghana’s AI ambitions deserve serious consideration, but national prestige should not replace technical scrutiny. Computing targets must be correctly expressed, electricity requirements realistically assessed, and projected costs independently evaluated. The government should publish clear performance measures, funding assumptions and implementation milestones. Starting with practical applications that solve identifiable problems could deliver early public benefits while the country builds the infrastructure needed for larger ambitions. The ultimate measure of success should be useful, secure and affordable technology-not merely the scale of the announcement.
