Vice-President and head of Research at the IMANI Centre for Policy and Education, Bright Simons has proposed a cheaper alternative to Ghana’s planned US$270 million GhanaChat project, arguing that a sovereign artificial intelligence (AI) assistant for the civil service could be a more practical starting point for the country’s AI ambitions.

In a post on X, Mr Simons questioned the rationale for developing a national AI platform to compete with global services such as ChatGPT, Gemini and Claude when existing models could potentially support government operations at a lower cost.

Mr Simons said if the government’s main concern was protecting sensitive public-sector information from being uploaded to external AI platforms, it could begin by developing a secure AI assistant for civil servants.

He estimated that such a system would require about 144 graphics processing units (GPUs), delivering approximately (7 \times 10^{17}) operations per second at eight-bit precision.

He put the initial cost at US$13 million, with annual operating expenses of about US$6.7 million.

“That is under a thousandth of the 2027 target, and a far more realistic initial-term strategy,” he wrote.

Mr Simons also questioned the need to train a foundational AI model from scratch, arguing that systems capable of coordinating multiple existing open-weight models could provide an alternative.

He suggested that such an approach could allow government institutions to retain control over their data while drawing on the capabilities of different AI models.

The government’s AI strategy includes plans to develop GhanaChat, a national platform intended to compete with international AI services.

The proposed system would be trained and hosted in a new AI data centre, with 48 public-sector registries expected to contribute data to improve its understanding of Ghanaian institutions and local needs.

The data centre and an accompanying AI Fund are estimated to cost about US$270 million, with additional expenditure expected for registries and related applications.

Mr Simons has questioned whether the proposed investment is justified, particularly given the cost of the infrastructure and the resources required to operate it.

Mr Simons also challenged the strategy’s computing capacity target for 2027, which he interpreted as (10^{21}) floating-point operations per second (FLOPS).

He estimated that meeting the target would require about 200,000 Nvidia chips, 500 megawatts of power and investment of approximately US$24 billion.

He said the US$250 million data centre being procured would provide only about one per cent of the stated target.

“Clearly, no serious critique of the plan would have okayed such a vision,” he wrote.

The estimates are Mr Simons’ assessment of the infrastructure needed to achieve the target. Their assumptions and the precise relationship between the target and the proposed data centre would require independent technical verification.

Mr Simons described the apparent shortcomings in the strategy as “classic KATANOMICS”, a term he uses to criticise projects that proceed without sufficient independent scrutiny.

He drew a comparison with a sugar factory he says was designed to fail, arguing that inadequate critical review has contributed to the failure of other government programmes.

“Throughout the Ministry's own review, cabinet approval, and a high-profile media launch, it is absolutely clear that not one critical reviewer was engaged,” he wrote.