AI Infrastructure Is Becoming a Business Planning Issue
You may not run a data centre, build an AI model, or manage advanced cloud infrastructure. Even so, the growing demand for AI computing can affect how you choose software, plan operations, protect business data, and serve customers.
For many Malaysian SME owners, the immediate question is not whether to build AI infrastructure. It is much more practical: how do you adopt useful AI without creating unreliable processes, unexpected dependency, or unnecessary complexity?
A recent report says British AI infrastructure company Nscale is discussing up to $3.5 billion in additional pre-IPO financing. The proposed package reportedly includes $1.5 billion in convertible notes and $2 billion from Nvidia. Source: TechCrunch
TL;DR: AI computing is becoming a major infrastructure industry, but you do not need to buy infrastructure to benefit. Focus on dependable workflows, clear data rules, integration with your existing systems, and measurable business outcomes.
Before committing to an AI tool, ask what happens when usage grows, data changes, staff make mistakes, or the provider changes its terms. Those questions matter more than simply choosing the tool with the most impressive demonstration.
What This Means
AI computing refers to the processing power needed to train and operate AI systems. Large AI models require specialised chips, data centres, cooling systems, storage, networking, and software. Providers such as Nscale supply this infrastructure to companies developing or running AI applications.
Nscale was founded about two years before the report and had previously raised $155 million in its Series A round in December 2024. Source: TechCrunch Its Series B financing in March 2026 was reported at $1.1 billion and described by the company as the largest Series B in European history. Source: TechCrunch
The important point is not the fundraising headline itself. It is the scale of the infrastructure being built behind AI services. When major investors and chip companies support computing providers, it signals that AI usage is expected to require substantial capacity for years.
The article also reports that Nscale signed a deal with Anthropic valued at approximately $45 billion. It says the company reportedly told potential investors that the agreement represented approximately $103 billion in revenue, although this was a projection based on signed customer leases rather than current sales. Source: TechCrunch
For you, this distinction is essential. A projected contract value is not the same as cash already received, completed work, or guaranteed future performance. The same principle applies when an AI software vendor presents impressive future capabilities: assess what works now, what is committed in writing, and what still depends on assumptions.
How This Applies to Malaysian SMEs
First, treat AI as an operating capability rather than a standalone app. A trading company may use AI to classify enquiries, draft quotations, summarise supplier emails, or identify slow-moving items. A clinic may use it to organise appointment requests, while a professional services firm may use it to prepare meeting summaries and follow-up tasks. These uses create value only when they connect to your existing workflow. If staff still copy information between WhatsApp, spreadsheets, email, and accounting software, the AI tool may add another layer of work.
Second, review your data before connecting any system. Malaysian SMEs commonly handle customer contact details, identification documents, invoices, employee information, supplier records, and confidential contracts. You should know where that data is stored, who can access it, whether it is used to train a public model, and how it can be deleted or exported. Start with low-risk information, such as internal procedures or general product descriptions, before allowing an AI system to process sensitive records.
Third, plan for reliability instead of assuming continuous perfection. AI services depend on external infrastructure. A provider may experience an outage, change an application interface, limit usage, or alter how a feature works. You need a simple fallback process for important activities such as order confirmation, payroll approval, customer complaints, and regulatory documentation. If your business cannot operate when one AI feature is unavailable, that feature has become a hidden single point of failure.
Fourth, consider local operating realities. Your customers may communicate in Bahasa Malaysia, English, Mandarin, Tamil, or a mixture of languages. Staff may work from branches with different internet quality. Certain approvals may still require a human decision. Test AI using real examples from your business, including abbreviations, local names, addresses, product codes, and mixed-language messages. A tool that performs well in a polished demonstration may behave differently with everyday Malaysian business data.
Finally, measure time and error reduction. Do not evaluate an AI project only by how modern it sounds. Record how long a task takes today, how many corrections are required, and where delays occur. After a trial, compare the results. For example, measure the time needed to prepare a weekly sales report, the percentage of customer enquiries routed correctly, or the number of invoice fields requiring manual checking.
A Simple AI Readiness Scorecard
| Area | Question to ask | Practical target |
|---|---|---|
| Workflow | Does the tool remove a repeated manual step? | Choose one clearly defined process for the first trial |
| Data | What information will the system receive? | Classify data into public, internal, confidential, and restricted |
| Accuracy | Who checks the output before action? | Assign one named reviewer for important outputs |
| Continuity | What happens if the service is unavailable? | Document a manual backup process |
| Results | How will you know the trial worked? | Track time, error rate, response speed, or completion rate |
Practical Takeaways
- Start with one repetitive workflow instead of introducing AI across the whole company.
- Keep a human approval step for financial, legal, employment, medical, and customer-impacting decisions.
- Ask vendors where your data is processed, how it is retained, and whether it is used for model training.
- Check whether the system can export your data and connect with the tools you already use.
- Test Bahasa Malaysia, English, mixed-language messages, local addresses, and common staff abbreviations.
- Create a written fallback procedure for every AI-supported process that affects daily operations.
- Review access permissions when staff join, change roles, or leave the company.
- Compare results against a baseline before deciding whether to expand the trial.
The infrastructure may be enormous, but your best AI decision can begin with one small, well-measured workflow.
Questions to Ask an AI Vendor
When speaking with a vendor, ask questions that reveal how the service works in practice. Ask whether your company’s prompts and documents are retained, whether data is isolated between customers, and whether administrators can control user access. Ask what happens when the provider changes its model or platform.
You should also ask about accuracy testing. Can the vendor show examples from a business similar to yours? How does the system handle uncertain answers? Can it identify when a human should review the result? If the answer is always “the model is improving,” you still need a process for handling errors today.
For business continuity, ask about service availability, incident communication, backup options, export formats, and integration limits. You do not need to understand every technical detail, but you do need clear answers about what your team can do if the service is interrupted.
The Bigger Picture
The Nscale financing report highlights a broader development: AI depends on a large physical and financial infrastructure layer. The market may continue to produce powerful tools, but businesses will still need sound processes, accurate data, responsible access controls, and staff who understand when to trust an automated result.
For Malaysian SMEs, this creates an advantage for disciplined adopters. You do not have to chase every new AI announcement. You can choose practical applications that reduce repeated work, improve response consistency, and help your team find information faster. The strongest results will usually come from connecting AI to a clean, well-managed business process.
Make your next step simple. Select one workflow, document how it currently operates, remove unnecessary data from the trial, appoint a reviewer, and define one success measure. After the test, keep, improve, or stop the process based on evidence.
That approach lets you benefit from the growing AI ecosystem without allowing infrastructure trends or vendor promises to dictate how your business operates.
Ready to Streamline Your Operations?
Your business should run itself. AutoRunBiz deploys AI agents to automate your daily operations — WhatsApp orders, invoicing, customer follow-ups, and accounting. Book a free 15-min ops audit to see where automation fits your business →