AI in Malaysia · Energy

Petronas is putting AI agents on its upstream data. Its contractors should start with the paperwork

Malaysia's energy giants are signing AI agreements and planning for a grid reshaped by data centres. For the more than 1,800 service companies behind them, the quicker return is in permits, inspection reports and tender packs.

Pikabyte Engineering 6 min read

1,855

oil and gas services companies counted by MPRC for FY2023. The top 100 took 76% of the revenue

On 8 July 2026 Petronas Carigali signed its third joint development agreement under TriCipta AI, its AI programme for upstream operations, with IBM Malaysia and Tridiagonal.AI. This one is about surface equipment: production and maintenance decisions. On 4 September 2026 Petronas, through Malaysia Petroleum Management, signed with Iraya Energies to add AI agents to myPROdata, the platform investors use to study Malaysian exploration blocks and fields. The stated aim is to cut the time from discovery to first hydrocarbons from 100 months to 50.

The electricity side is moving as fast. In a written parliamentary answer in July 2026, the energy ministry said Tenaga Nasional was supplying 36 operating data centres with a planned capacity of around 4.5 gigawatts, with 23 more under construction. It projected peak demand rising from 21.3 gigawatts in 2026 to 33.5 gigawatts by 2035.

Those are the giants. Behind them stands a long tail. The Malaysia Petroleum Resources Corporation’s OGSE100 report for the 2023 financial year counted 1,855 oil and gas services and equipment companies with combined revenue of RM83.9 billion. The top 100 accounted for 76% of it. The other 1,755 shared the remaining quarter.

This article is for those companies, and for their counterparts who build and maintain the grid.

More activity means more records

The Petronas Activity Outlook for 2026 to 2028, published on 30 January 2026, guides development and exploration wells up from 79 in 2025 to 91 in 2026 and 100 by 2028. Well plugging and abandonment rises from 63 to 70 and then 80. Analysts quoted by The Star read the outlook as favouring hook-up, commissioning and plant maintenance providers. Petronas itself described the plan as “creating sustained demand for technical services and innovative solutions from OGSE partners”.

On the power side, the government has approved RM43 billion of capital expenditure for TNB for the 2025 to 2027 regulatory period. The ministry expects data centres to account for up to 31% of electricity demand in 2035, against 7% this year, while the generation mix moves towards a target of 70% clean energy capacity by 2050.

Every one of those jobs is carried out by a contractor, and every one produces paper. Permits to work. Job hazard analyses. Inspection and test records. Calibration and equipment certificates. Daily reports. Completion dossiers. The work is physical. Getting mobilised, getting accepted and getting paid depend on the documents.

Do not copy the operator

Petronas’s AI programmes sit on decades of structured subsurface and equipment data, and on teams whose job is to act on what a model says. A service company with 60 staff has neither. Its history is in PDF reports, scanned forms, spreadsheets and WhatsApp groups.

That is not a reason to wait. It is a reason to start somewhere else. Predictive maintenance needs a clean maintenance history before it can predict anything. The first projects are the ones that create that history as a by-product of ordinary work.

Five paperwork jobs worth automating

1. Tender and prequalification packs. A tender arrives as hundreds of pages. A model can read it and draft the compliance matrix: each requirement, where your answer lives, and what is missing. It can assemble the standard company documents and flag the certificate that expires before the closing date. Your commercial team decides what to bid and at what price.

2. Inspection and test reports. Field forms, gauge photos and handwritten remarks can be read into structured records for an inspector to confirm. We built a quality control platform for a manufacturer of water reticulation products. Every pressure test is recorded per attempt, only products that passed can be put on a delivery order, and each certificate carries a QR code a purchaser can scan to verify it. The same shape fits hydrotests, valve tests and site acceptance tests.

3. Maintenance records. Technicians write what they did in free text. A model can turn “tukar seal, pump trip lagi” into an asset, a failure type and an action, in whichever language the technician used. After a year you can ask what was done the last three times that pump tripped. That is the data set a prediction project needs.

4. HSE paperwork. Toolbox talk records, near-miss reports and permit forms are filled in by hand, in volume. Software can check that a form is complete before the job starts, sort incident reports by type and location, and show the trend a monthly safety meeting should be looking at.

5. Certificates and competency. Calibration certificates, lifting equipment inspections, medical and offshore training records all carry an expiry date. Extract the date once and let ordinary code send the reminder. A crew member turned back at the heliport because a certificate lapsed is an expensive way to find out.

We run the reading part of this pattern in production for other industries. Reading and structuring documents is the job current models do well.

Where AI should not decide

A permit is signed by a competent person. Software can check that every field is filled and every attachment is present. It does not authorise work on live plant.

Pass or fail comes from the instrument and the rule. A pressure held for the required time is a comparison of numbers. Let code do it, the same way every time, and let the inspector sign.

Engineering calculations stay deterministic. A model that is usually right is not acceptable for a load, a setting or an isolation.

Client data is confidential by contract. Drawings, well data and plant information belong to the operator. Know which documents are sent to a model, to which provider, and in which country it runs. If an AI service holds your client’s documents, how it authenticates matters as well.

The division we recommend is the one we use ourselves: AI reads, code checks, a person signs. Designing that review step is most of the work.

What this means if you run an energy services business

  1. Pick the document that delays money or mobilisation. For some it is the tender pack. For others it is the completion dossier that holds up the final invoice.
  2. Capture at the source. A phone form with photos, filled in at the worksite, beats a report typed up at the end of the hitch.
  3. Keep every record against the asset and the job. Then the handover dossier is a query, not a fortnight of compiling.
  4. Put a named person on every AI output. Record who confirmed it.
  5. Only then talk about prediction, with a year of clean history to feed it.

The same order of work applies in manufacturing and construction. The operators will keep announcing AI agreements. The contractor who can produce any inspection record for any job in a minute is solving a smaller problem, and getting paid sooner for it.

Sources

Written by

Pikabyte Engineering

Pikabyte builds custom business systems, applied AI and secure cloud infrastructure from Malaysia. Talk to us.

The work behind this article

Platform · Manufacturing · Water reticulation

Every product traced from the factory floor to the purchaser, with a certificate anyone can verify.

A quality control platform for a water reticulation products manufacturer. Workers scan a QR code, record the pressure test, and passed products flow into delivery orders and test certificates that a purchaser or authority officer can verify by scanning.

20,000+
products and delivery orders traced
< 1 s
to generate a test certificate
9.5 weeks
from kickoff to production
94%
backend test coverage
Tour IQC Soon Chuan Hong Hardware Sdn. Bhd. · Malaysia

Start a conversation

Want this applied to your business?

Tell us what is slowing the business down. You will talk to an engineer, not a salesperson, and get a straight answer on whether software is the right fix.