Question 1

Can FirstNet's EngineAI team fine-tune an AI model to our industry terminology?

Accepted Answer

Yes. EngineAI, the build team of FirstNet's AI business unit FirstCoreAI, offers fine-tuning and model optimisation that adapt models to your domain, your terminology and your own data.

Fine-tuning helps when general-purpose models handle sector jargon, product names or internal abbreviations poorly. EngineAI can combine it with its other capabilities:
- RAG and retrieval, so answers draw on your documents, contracts and internal knowledge.
- Document and call pipelines for intake, extraction, transcription and structured output.
- Integration into Microsoft 365, your CRM and other line-of-business systems.

Most EngineAI projects run on InfraAI compute in FirstNet's South African data centre, so models, data and pipelines stay in the country. Open-weight models are the default, and because their weights are available they can be adapted and hosted locally rather than through a foreign platform.

FirstCoreAI scopes each problem first and will say plainly if AI is not the right tool. A scoping call or demo with FirstCoreAI is the place to start.

Question 2

Can FirstNet's FirstCoreAI AI adoption service be combined with EngineAI custom AI work?

Accepted Answer

Yes. Many FirstCoreAI clients combine AI Adoption & Enablement with EngineAI builds, such as integrations into Microsoft 365 and line-of-business systems. Both are services of FirstCoreAI, FirstNet's AI business unit, so the people side and the build side can be planned together.

Adoption covers private AI assistants, role-based training, AI governance and change leadership. EngineAI is FirstCoreAI's build team, taking a use case from proof of concept to production. Its work includes:
- RAG and retrieval that make documents, contracts and internal knowledge searchable in plain language.
- Document and call pipelines for intake, extraction, transcription and structured output.
- Autonomous agents for multi-step workflows that approve, route, tag and report.
- Fine-tuning that adapts models to your domain, terminology and data.
- Integration into Microsoft 365, your CRM and other line-of-business systems.

EngineAI solutions are deployed on the AI Factory in FirstNet's South African data centre or inside your own environment. FirstCoreAI's four solution lines each work on their own, but most organisations combine two or three.

A scoping call with FirstCoreAI can cover both the adoption programme and the custom build.

Question 3

Can FirstNet's FirstCoreAI process call recordings with AI without sending them offshore?

Accepted Answer

Yes. For a national insurer, EngineAI, the build team of FirstNet's AI business unit FirstCoreAI, processed 965 call recordings into structured, searchable data on sovereign infrastructure, so no recording left South Africa.

The pipeline covered every stage from audio to usable data:
- Transcription and correction of every recording.
- Name resolution across calls.
- Structured JSON output for downstream systems.
- Processing entirely on local infrastructure.

Keeping call recordings local matters because sending them to an offshore AI API is a cross-border transfer, with POPIA consent and breach questions attached. The FirstCoreAI AI Factory in FirstNet's South African data centre includes a speech-to-text model exposed through an OpenAI-compatible audio transcriptions endpoint.

Document and call pipelines are a core EngineAI capability, covering intake, extraction, transcription and structured output for large volumes of files and recordings. FirstCoreAI names call archives as a typical use case in financial services and insurance, and as a good single workload to start with.

FirstCoreAI can scope your own archive on a 30-minute scoping call.

Question 4

Can FirstNet's FirstCoreAI train our teams to use AI?

Accepted Answer

Yes. FirstCoreAI, FirstNet's AI business unit, provides practical, role-based training, so people learn to use AI on their own work, safely and consistently. It is part of the AI Adoption & Enablement service and aims to make teams productive with AI, not merely licensed.

How the training fits in:
- It is role-based and built around real, day-to-day tasks.
- It sits alongside AI governance, where usage policies, data-handling rules and guardrails tell staff what is safe to share.
- In FirstCoreAI's adoption path, training is in place before wider rollout, after a pilot group has used AI on real work with clear success measures.
- As AI scales across teams, usage and outcomes keep being measured.

When the assistants people train on run on the Sovereign AI Factory in FirstNet's South African data centre, the prompts and documents they use stay in the country.

Training can be combined with executive and management briefings, or with EngineAI integrations into Microsoft 365 and line-of-business systems. A 30-minute scoping call with FirstCoreAI identifies the first use cases and plans a pilot.

Question 5

Can I get dedicated GPU capacity in South Africa from FirstNet's InfraAI service?

Accepted Answer

Yes. InfraAI, the GPU and inference service from FirstNet's AI business unit FirstCoreAI, offers both dedicated and shared GPU capacity on the AI Factory, sized to the workload and billed in rand.

The AI Factory runs current-generation NVIDIA Blackwell-class GPUs with AMD EPYC processors and more than a terabyte of system RAM per node, in FirstNet's South African data centre. Run.ai handles GPU-aware scheduling and vLLM serves inference.

What InfraAI capacity includes:
- Dedicated or shared GPU options, sized to your workload.
- Dedicated capacity instead of a shared public queue.
- Private inference endpoints hosted locally, with every inference inside the South African perimeter.
- Support from the engineers who run the Factory.
- Rand-denominated, forecastable commercials with no FX surprises.

InfraAI suits high-volume document and call processing, RAG and retrieval back-ends, internal assistants, classification and extraction. On the production cluster, FirstCoreAI measured 271 concurrent requests sustained at peak with zero errors across its stress tests.

To size capacity for your workload, book a demo or scoping call with FirstCoreAI.

Question 6

Can I still use overseas frontier AI models with FirstNet's FirstCoreAI sovereign AI platform?

Accepted Answer

Yes. Sovereign does not mean isolated: FirstCoreAI, FirstNet's AI business unit, runs open-weight models in South Africa by default, and frontier APIs from overseas labs can be used where a workload justifies it. FirstCoreAI tells you plainly when a request would leave the country.

Most workloads run well on the open-weight models hosted on the AI Factory in FirstNet's South African data centre. The live catalogue includes an 80B mixture-of-experts coding model, a 31B general chat model, vision-language, speech-to-text, multilingual embeddings and a retrieval reranker.

How the choice is made:
- Every model on the AI Factory is benchmarked on the same hardware against frontier APIs for accuracy, latency, concurrency and cost per unit of work, so you see the evidence before committing.
- Frontier routing is optional, and you get notice when a request goes offshore.
- The decision is made per workload, so sensitive data such as customer records or call recordings can stay on local models while frontier APIs are reserved for workloads that justify them.

This lets you keep sensitive processing inside South Africa, with rand-denominated costs, without giving up access to frontier models where they genuinely add value.

Question 7

Can I tour FirstNet's FirstCoreAI AI Factory GPU cluster?

Accepted Answer

Yes. FirstCoreAI, FirstNet's AI business unit, runs AI Factory tours of 60 to 90 minutes that cover the hardware, the model catalogue and a live workload. Tours are booked directly with FirstCoreAI.

A tour is one of two ways to start an engagement, alongside a 30-minute scoping call. It shows:
- The hardware: the NVIDIA Blackwell-class GPU cluster that FirstCoreAI owns and operates in FirstNet's South African data centre.
- The model catalogue: six live endpoints covering coding, chat, vision-language, speech-to-text, multilingual embeddings and reranking.
- A live workload running on the cluster through OpenAI-compatible APIs.

FirstCoreAI stress-tested the production cluster under real serving conditions, measuring a median time to first token of about 0.6 seconds on chat and coding models and 271 concurrent requests sustained at peak with zero errors.

To get the most from a tour, bring one workload you want to see running sovereign, such as a document pipeline, call archive, chatbot or a model you cannot deploy today.

Question 8

Can approval workflows be built into FirstNet's StudioAI content production process?

Accepted Answer

Yes. StudioAI, the creative and content AI service from FirstNet's AI business unit FirstCoreAI, automates campaign workflows and approval flows, so content moves through review and publishing in one connected system.

StudioAI connects the stages of content production so scripting, production, approvals and publishing run through one system, plugged into the creative tools your team already uses.

How approvals fit in:
- Campaign workflows are automated, so content moves from draft to publish faster.
- Approval flows sit in the same connected system as production and publishing.
- On-brand templates and asset libraries keep output aligned to your guidelines.
- Your existing creative tools stay in place; StudioAI connects them rather than replacing them.

This matters when personalised video, voiceover, copy and design are produced at a volume manual production cannot match. FirstCoreAI can walk you through a complete campaign, from script to publish, in a single demo session.

Question 9

Can we see a demo of FirstNet's StudioAI creative content service?

Accepted Answer

Yes. FirstCoreAI, FirstNet's AI business unit, demonstrates a complete, worked StudioAI campaign from script to publish in a single session. Demos are booked directly with FirstCoreAI.

A StudioAI campaign runs through connected steps in one system:
- Scripting and production of personalised video, voiceover, copy and design.
- On-brand output through templates and asset libraries.
- Automated campaign workflows and approval flows.
- Publishing through the same connected workflow.

A demo is a useful way to see how StudioAI would plug into the creative tools your team already uses, since it integrates with them rather than replacing them. Many organisations look at StudioAI after infrastructure or engineering conversations with FirstCoreAI, because content often delivers a quick, visible result.

To arrange one, book a demo with FirstCoreAI, or talk to FirstNet on 0861 989 896 or sales@firstnet.co.za.

Question 10

Do I need to rewrite my application to use FirstNet's InfraAI GPU and inference service?

Accepted Answer

No. Every model on InfraAI, the GPU and inference service from FirstNet's AI business unit FirstCoreAI, is exposed through an OpenAI-compatible API, so any SDK that already works with OpenAI works with InfraAI. You only swap in the endpoint URLs and credentials issued at onboarding.

InfraAI runs on the AI Factory in FirstNet's South African data centre. Applications call standard endpoints:
- /v1/chat/completions for chat and coding
- /v1/embeddings for search and retrieval
- /v1/audio/transcriptions for speech-to-text
- /v1/rerank for retrieval reranking

Your application authenticates with OAuth2 client credentials and receives a short-lived bearer token. TLS terminates at the cluster edge and each request is validated. Your token endpoint and URLs are issued at onboarding, and existing OpenAI SDKs work without code changes.

For an application already built on an OpenAI SDK, moving to InfraAI is a configuration change rather than a rebuild. The difference is where it runs: inference then happens in South Africa, on dedicated or shared GPU capacity sized to the workload, with rand-denominated billing.

Question 11

Do our prompts and documents leave South Africa when we use FirstNet's FirstCoreAI assistants?

Accepted Answer

No. FirstCoreAI's private AI assistants run on the Sovereign AI Factory, and inference runs on GPUs in FirstNet's South African data centre, operated by FirstCoreAI. Your prompts, documents and outputs are processed and stay in the country.

The assistants are secure chat and document tools that work with your own knowledge, with nothing sent offshore. By contrast, sending customer records or call recordings to an offshore AI API is a cross-border transfer, with POPIA consent and breach questions attached.

Details worth knowing:
- FirstCoreAI owns and operates the GPU cluster, and it runs on networks FirstNet operates.
- Models are open-weight and hosted locally, so you are not locked into a foreign platform.
- Access uses OAuth2 client credentials and short-lived bearer tokens, and each request is validated.
- Frontier models from overseas labs are optional. If a workload is routed to one, FirstCoreAI tells you plainly that the request would leave the country.

This is what makes the approach POPIA-aligned by design, and it removes foreign-jurisdiction and data-export concerns before rollout starts.

Question 12

Do we have to replace our current creative tools to use FirstNet's StudioAI?

Accepted Answer

No. StudioAI, the creative and content AI service from FirstNet's AI business unit FirstCoreAI, integrates with the creative tools your team already uses and connects them into one workflow, rather than asking you to rebuild your marketing stack.

StudioAI plugs proven creative tools into your existing stack and connects the steps between them, so scripting, production, approvals and publishing run through one system.

What that means in practice:
- Your existing creative stack stays in place, and StudioAI integrates into it.
- Approvals and publishing flow through one system.
- On-brand templates and asset libraries keep output aligned to your guidelines.
- Personalised video, voiceover, copy and design can be produced at a volume manual production cannot match.

Because nothing has to be replaced, StudioAI often delivers a quick, visible win, and many clients add it after starting with FirstCoreAI's infrastructure or engineering services. A FirstCoreAI demo walks through a complete campaign from script to publish in a single session.

Question 13

Does FirstNet's FirstCoreAI help with AI policy and governance?

Accepted Answer

Yes. As part of its AI Adoption & Enablement service, FirstCoreAI, FirstNet's AI business unit, helps you define AI usage policies, data-handling rules and guardrails aligned to POPIA, so people know what is safe to share and AI use stays compliant.

Most AI projects stall between the pilot and everyday use: people are unsure what they may share, data ends up in foreign services and nobody owns the outcome. Governance answers those questions before AI use spreads.

What the governance work covers:
- Usage policies that set out how AI may be used in your organisation.
- Clear data-handling rules on what data may be used where.
- Guardrails that help keep everyday use within policy.
- Timing: in FirstCoreAI's adoption path, policies, guardrails and training are in place before wider rollout, after a controlled pilot.

It also helps that the assistants themselves run on the Sovereign AI Factory in FirstNet's South African data centre: prompts and documents are processed in the country rather than sent to foreign services, which removes the biggest data risk first.

FirstCoreAI can scope governance work alongside a pilot on a 30-minute scoping call.

Question 14

Does sovereign AI from FirstNet's FirstCoreAI help with POPIA compliance?

Accepted Answer

Yes, it helps. FirstCoreAI, FirstNet's AI business unit, designs its endpoints, retention and access controls to align with POPIA obligations, and because inference runs in FirstNet's South African data centre, workloads that stay on the AI Factory involve no cross-border transfer to reconcile.

Sending customer records, claims or call recordings to an offshore AI API is a cross-border transfer, with consent and breach questions attached. Data held by foreign providers can also fall under foreign law, such as the US CLOUD Act. Local, sovereign hosting means only the people you have authorised can reach your information.

How FirstCoreAI supports a POPIA-aligned approach:
- Prompts, documents and outputs are processed in South Africa on infrastructure FirstCoreAI owns and operates.
- Access uses OAuth2 client credentials and short-lived bearer tokens, and each request is validated.
- AI governance work sets usage policies, data-handling rules and guardrails aligned to POPIA, so staff know what is safe to share.
- If a workload is routed to an overseas frontier model, FirstCoreAI tells you plainly that the request would leave the country.

Local hosting supports compliance but does not replace your organisation's own POPIA obligations. A FirstCoreAI scoping call can identify which workloads should stay on the AI Factory.

Question 15

How do we get started with sovereign AI from FirstNet's FirstCoreAI?

Accepted Answer

Start with one workload. Bring FirstCoreAI, FirstNet's AI business unit, a single use case, such as a document pipeline, call archive, chatbot or a model you cannot deploy, and book a 30-minute scoping call or an AI Factory tour. FirstCoreAI will show it running sovereign.

A typical path looks like this:
- Scoping: a 30-minute call that starts with the business problem, not the technology, and defines what a working solution looks like. FirstCoreAI will say plainly if AI is not the answer.
- See it running: a demo, or a 60 to 90 minute AI Factory tour covering the hardware, the model catalogue and a live workload.
- Pilot: a defined group uses the solution on real work, with clear success measures.
- Scale: roll out once policies, guardrails and training are in place, and keep measuring usage and outcomes.

Good first workloads are ones where data is sensitive and volumes are high, for example claims, call archives and document extraction in financial services and insurance, or personalisation and campaign automation in retail and FMCG.

Demos, scoping calls and tours are booked directly with FirstCoreAI. You can also talk to FirstNet on 0861 989 896 or sales@firstnet.co.za.

Question 16

How does FirstCoreAI, the AI business unit, relate to FirstNet and the First Technology Group?

Accepted Answer

FirstCoreAI is a business unit of FirstNet, the data-centre, connectivity and ISP arm of the First Technology Group. It is FirstNet's AI business, and its AI Factory GPU cluster runs inside FirstNet's South African data centre on networks FirstNet operates.

That structure means the team that runs your connectivity also runs the platform your AI lives on. FirstCoreAI owns and operates the physical compute, and network tuning, rack scaling and cross-connects are handled by one team.

FirstNet was founded in 2006 as a wholly owned subsidiary of the First Technology Group, which has over three decades in South African IT.

FirstCoreAI delivers four solution lines from that platform:
- GPU & Inference (InfraAI): private inference endpoints and GPU capacity, hosted locally and billed in rand.
- Custom AI Engineering (EngineAI): RAG systems, AI agents, document and call pipelines, and fine-tuned models.
- Creative & Content AI (StudioAI): personalised video, voice, copy and design produced at scale.
- AI Adoption & Enablement: private AI assistants, training and governance.

Each line works on its own, but most organisations combine two or three. Demos, scoping calls and AI Factory tours are booked directly with FirstCoreAI.

Question 17

How does FirstNet's FirstCoreAI get leadership on board with AI adoption?

Accepted Answer

FirstCoreAI, FirstNet's AI business unit, runs executive and management briefings as part of its AI Adoption & Enablement service. The briefings start from wherever the scepticism sits and turn it into genuine sponsorship, so leaders actively back AI use.

Leadership matters because most AI projects stall between the pilot and everyday use, often because nobody owns the outcome. FirstCoreAI builds executive sponsorship from the start, as part of a structured change programme, rather than leaving uptake to chance.

The rest of the adoption path then gives leaders evidence to back:
- Discover: identify the use cases and data that will deliver value first.
- Pilot: a defined group uses a sovereign assistant on real work, with clear success measures.
- Govern: policies, guardrails and training are in place before wider rollout.
- Scale: roll out across teams and keep measuring usage and outcomes.

Running assistants on the Sovereign AI Factory also answers a common executive concern about data. Prompts, documents and outputs are processed in South Africa on infrastructure FirstCoreAI owns and operates, which removes foreign-jurisdiction and data-export concerns before you start.

Question 18

How does StudioAI relate to FirstNet's other FirstCoreAI sovereign AI services?

Accepted Answer

StudioAI is one application of the wider FirstCoreAI platform, the AI business unit of FirstNet. Many clients start with infrastructure (InfraAI) or engineering (EngineAI) and add StudioAI as a quick, visible win.

FirstCoreAI offers four solution lines. Each works on its own, but most organisations combine two or three:
- GPU & Inference (InfraAI): private inference endpoints and GPU capacity from the AI Factory, hosted locally and billed in rand.
- Custom AI Engineering (EngineAI): RAG systems, AI agents, document and call pipelines, and fine-tuned models.
- Creative & Content AI (StudioAI): personalised video, voice, copy and design at scale, kept on-brand.
- AI Adoption & Enablement: private AI assistants, training and governance.

Most clients reach StudioAI after infrastructure and engineering conversations. Content is seldom the first sovereignty concern, but it often delivers a quick, visible result. One FMCG group's AI programme, for example, spans content, reporting and hybrid hosting.

All four lines are delivered from South Africa, and a FirstCoreAI scoping call can map which combination fits your organisation.

Question 19

How is FirstNet's FirstCoreAI sovereign AI service billed?

Accepted Answer

FirstCoreAI, FirstNet's AI business unit, bills sovereign AI in South African rand on a workload-based model, so costs are predictable and not exposed to exchange-rate movements. Pricing is discussed during scoping with FirstCoreAI.

Offshore AI services are typically metered per token in dollars, so the bill moves with both usage and the exchange rate. Running open-weight models on local hardware in FirstNet's South African data centre turns that into predictable, rand-denominated spend that finance teams can actually forecast.

Because commercials are workload-based, the starting point is the workload itself: a document pipeline, call archive, chatbot, internal assistant or a model you want to host. On InfraAI, FirstCoreAI's GPU and inference service, you can choose dedicated or shared GPU capacity, sized to that workload and billed in rand.

Every model on the AI Factory is also benchmarked against frontier APIs for cost per unit of work, alongside accuracy, latency and concurrency, so you can compare the economics with evidence before committing.

To get a rand figure for your own workload, book a scoping call with FirstCoreAI, or talk to FirstNet on 0861 989 896.

Question 20

How is access to FirstNet's InfraAI inference endpoints on the FirstCoreAI AI Factory secured?

Accepted Answer

Clients authenticate to InfraAI with OAuth2 client credentials and receive short-lived bearer tokens. TLS is terminated at the edge of FirstCoreAI's AI Factory cluster in FirstNet's South African data centre, and OAuth2 proxies validate each request.

How the security layers fit together:
- Authentication: the OAuth2 client credentials flow is designed for application-to-application access, and short-lived tokens limit how long a leaked token stays useful.
- Encryption in transit: TLS terminates at the cluster edge, where HAProxy and a Knative gateway handle incoming traffic.
- Request validation: OAuth2 proxies check every request before it is served.
- Management isolation: out-of-band management is segregated from the in-band fabric between nodes.
- Onboarding: your token endpoint, endpoint URLs and credentials are issued to you at onboarding.

Data protection goes beyond access control. Inference runs on GPUs that FirstCoreAI, FirstNet's AI business unit, owns and operates in South Africa, and FirstCoreAI designs its endpoints, retention and access controls to align with POPIA. Requests stay in the country unless a workload uses optional frontier routing, in which case FirstCoreAI gives notice.

Question 21

What can FirstNet's StudioAI creative and content AI service produce?

Accepted Answer

StudioAI produces hyper-personalised video, voiceover, copy and design at scale, kept on-brand through templates and asset libraries. It is the creative and content AI offering from FirstCoreAI, FirstNet's AI business unit, for organisations where content itself is the workload.

What it covers:
- Personalised content: video, voice and copy tailored to audiences at a volume manual production cannot match.
- Brand consistency: on-brand templates and asset libraries keep every output aligned to your guidelines.
- Workflow automation: campaign workflows and approval flows are automated, so content moves from draft to publish faster.
- Tool integration: StudioAI works with the creative tools your team already uses rather than replacing them.

StudioAI connects scripting, production, approvals and publishing into one system, plugged into your existing creative stack.

It fits retail and FMCG teams working on personalisation and campaign automation, and it often delivers a quick, visible result for organisations already working with FirstCoreAI on infrastructure or engineering. FirstCoreAI can walk you through a complete campaign, from script to publish, in a single demo session.

Question 22

What does FirstNet's FirstCoreAI AI Adoption & Enablement service cover?

Accepted Answer

FirstCoreAI's AI Adoption & Enablement service covers four things: private AI assistant rollout, role-based training, AI governance (usage policies, data-handling rules and guardrails) and change leadership for executives and managers. FirstCoreAI is FirstNet's AI business unit, and the aim is that people use AI well, safely and consistently.

What each part includes:
- Private AI assistants: secure chat and document assistants that run on the Sovereign AI Factory and work with your own knowledge, with nothing sent offshore.
- Training and enablement: practical, role-based training that makes teams productive with AI, not merely licensed.
- AI governance: usage policies, data-handling rules and guardrails aligned to POPIA, so people know what is safe to share.
- Change leadership: executive and management briefings that turn scepticism into genuine sponsorship.

Every engagement follows a measured path. Discover the use cases and data that deliver value first, pilot with a defined group and clear success measures, put policies, guardrails and training in place before wider rollout, then scale across teams while measuring usage and outcomes.

The service suits organisations that already have AI tools but are not yet seeing the value, and those that want to get adoption right the first time.

Question 23

What happens if FirstNet's FirstCoreAI finds AI isn't the right fit for our problem?

Accepted Answer

FirstCoreAI, FirstNet's AI business unit, scopes each problem first and will tell you honestly if AI is not the right tool. You start with the business problem, not the technology, and the scoping shows what a working solution would look like.

How scoping works:
- It starts with a 30-minute scoping call or a demo, booked directly with FirstCoreAI.
- The focus is the business problem and the data involved, not a particular model or product.
- FirstCoreAI says so plainly if AI is not the answer.
- Where AI does fit, open-weight models are the default, with frontier APIs only where justified.

When AI is a fit, FirstCoreAI's EngineAI team takes the use case from proof of concept to production, building RAG systems, document and call pipelines, AI agents, fine-tuned models or integrations. Solutions run on the AI Factory in FirstNet's South African data centre or in your own environment.

Adoption work follows the same principle: every engagement starts small, with a pilot and clear success measures, and earns its way to scale.

Question 24

What is sovereign AI, and how does FirstNet's FirstCoreAI deliver it in South Africa?

Accepted Answer

Sovereign AI is AI that runs on infrastructure in your own country, under your own laws and priced in your own currency. FirstNet delivers it through FirstCoreAI, its AI business unit, which runs inference on GPUs in FirstNet's South African data centre so your data stays in the country.

For South African organisations, sovereign AI addresses three pressures that offshore AI services leave open:
- Data residency: sending customer records, claims or call recordings to an offshore API is a cross-border transfer, with POPIA consent and breach questions attached.
- Jurisdiction: data held by foreign providers can fall under foreign law, such as the US CLOUD Act.
- Cost: offshore AI is typically metered per token in dollars, while FirstCoreAI bills in rand.

The FirstCoreAI AI Factory is a Blackwell-class GPU cluster that FirstCoreAI owns and operates, serving open-weight models through OpenAI-compatible APIs. Sovereign does not mean isolated: frontier APIs can still be used where a workload justifies it, and FirstCoreAI tells you plainly when a request would leave the country.

Sovereign AI matters most where data is sensitive and volumes are high, such as financial services and insurance, retail and FMCG, and public-sector workloads where residency is a legal requirement.

Question 25

What kind of custom AI solutions does FirstNet's EngineAI team build?

Accepted Answer

FirstNet's EngineAI team builds RAG search, document and call pipelines, AI agents, fine-tuned models and links into Microsoft 365, CRMs and line-of-business systems. It takes each use case from proof of concept to live production.

EngineAI is the build team of FirstCoreAI, FirstNet's AI business unit. What it builds:
- RAG and search: RAG (retrieval-augmented generation) makes documents, contracts and in-house knowledge searchable, with answers in plain language.
- Document and call pipelines: intake, extraction, transcription and structured output for large volumes of files and recordings.
- Autonomous agents: multi-step workflows that approve, route, tag and report without a person at every step.
- Fine-tuning and optimisation: models adapted to your field, your terms and your own data.
- Integration: hooks into Microsoft 365, your CRM and other line-of-business systems.

Most EngineAI projects run on InfraAI compute in FirstNet's South African data centre. That way, models, data and pipelines stay in the country. Or EngineAI can set up the solution on your own systems instead. Open-weight models are the default, with frontier APIs used where the case justifies them.

In one project, EngineAI turned 965 call recordings for a national insurer into structured data you can search. Not one recording left South Africa.

Question 26

Where are FirstNet's EngineAI custom AI solutions deployed?

Accepted Answer

EngineAI solutions are deployed either on the FirstCoreAI AI Factory in FirstNet's South African data centre or into your own environment, depending on what suits the workload. EngineAI is the build team of FirstCoreAI, FirstNet's AI business unit.

Most EngineAI projects run on InfraAI compute on the AI Factory, so models, data and pipelines stay in the country. Where it makes more sense, solutions can be deployed into your own environment instead.

Deployment principles:
- Build on the AI Factory or in your environment.
- Open-weight models by default, with frontier APIs only where a workload justifies them.
- Notice from FirstCoreAI whenever a request would leave South Africa.
- On the AI Factory, OpenAI-compatible APIs with OAuth2 client credentials and short-lived bearer tokens.

Running on the AI Factory keeps every inference inside the South African perimeter, on infrastructure FirstCoreAI owns and operates, with rand-denominated billing. That suits workloads where data sensitivity or cost rules out a foreign API.

A scoping call with FirstCoreAI is the place to work out which option suits your workload.

Question 27

Where is FirstNet's FirstCoreAI AI Factory hosted, and does inference stay in South Africa?

Accepted Answer

The FirstCoreAI AI Factory is hosted in FirstNet's data centre in South Africa. FirstCoreAI, FirstNet's AI business unit, owns and operates the GPU cluster, and inference on the AI Factory happens inside South Africa.

The AI Factory is physical compute, tuned for production throughput and concurrency rather than benchmark leaderboards, and it runs on networks FirstNet operates. Network tuning, rack scaling and cross-connects are handled by one team.

What is inside:
- Current-generation NVIDIA Blackwell-class GPUs with AMD EPYC processors and more than a terabyte of system RAM per node.
- A multi-terabyte NVMe model cache shared over NFS, a high-throughput in-band fabric between nodes and segregated out-of-band management.
- Kubernetes with Run.ai for GPU-aware scheduling, vLLM for inference, HAProxy and a Knative gateway at the edge, and Grafana and Prometheus for monitoring.
- Six live production model endpoints exposed through an OpenAI-compatible API.

Because processing stays local, workloads on the AI Factory involve no cross-border transfer of data. FirstCoreAI runs 60 to 90 minute tours covering the hardware, the model catalogue and a live workload, booked directly with FirstCoreAI.

Question 28

Which AI models are available on FirstNet's InfraAI service and FirstCoreAI AI Factory?

Accepted Answer

Six production model endpoints are live on the FirstCoreAI AI Factory, which InfraAI delivers as a service: an 80B mixture-of-experts coding model, a 31B general chat model, vision-language, speech-to-text, multilingual embeddings and a retrieval reranker. All are open-weight models hosted in FirstNet's South African data centre.

The catalogue at a glance:
- Coding: an 80B mixture-of-experts model with a 131K context window.
- General chat: a 31B model with a 128K context window.
- Vision-language: for tasks that combine images and text.
- Speech-to-text: transcription through the audio transcriptions endpoint.
- Multilingual embeddings and a retrieval reranker: the building blocks for search and RAG.

Every model is exposed through an OpenAI-compatible API and is benchmarked on the same hardware against frontier APIs for accuracy, latency, concurrency and cost per unit of work. On the production cluster, FirstCoreAI measured a median time to first token of about 0.6 seconds on the chat and coding models.

Where a workload genuinely needs an overseas frontier model, requests can be routed to a frontier API, and FirstCoreAI gives notice when a request leaves the country.

Question 29

Why adopt AI on sovereign South African infrastructure with FirstNet's FirstCoreAI?

Accepted Answer

Because the biggest barrier to AI adoption is trust in what happens to your data. When assistants run on the Sovereign AI Factory from FirstCoreAI, FirstNet's AI business unit, prompts, documents and outputs are processed in South Africa on infrastructure FirstCoreAI owns and operates, which removes foreign-jurisdiction and data-export concerns before you start.

Most AI projects stall between the pilot and everyday use: people are unsure what they may share, data ends up in foreign services and nobody owns the outcome. Adopting on sovereign infrastructure removes the biggest risk first, so the work can focus on making AI genuinely useful.

What sovereign hosting adds:
- Inference runs on GPUs in FirstNet's South African data centre.
- POPIA-aligned by design, because prompts and documents are processed in the country.
- Data held by foreign providers can fall under foreign law, such as the US CLOUD Act; local hosting keeps access with the people you authorise.
- Open-weight models, so you are not locked into a foreign platform.
- Rand-denominated commercials with no exchange-rate surprises.

FirstCoreAI pairs this with role-based training, AI governance and change leadership, so private assistants move from pilot to everyday use.

Sovereign AI questions & answers