by Agmo Studio
Today's digest covers a data centre capacity announcement for Malaysia and three global developments: new agent security controls from Microsoft, open models from Nvidia, and a chip software tool from Amazon.
Malaysia
Aolani signs letters of intent for up to 500MW of AI data centre capacity in Malaysia
Aolani, a neocloud founded in Singapore, said it has signed letters of intent for up to 500MW of AI data centre capacity across several Malaysian sites, with capacity targeted from the second half of 2027. The company said the sites will be designed to Tier III standards. The letters are non-binding and depend on definitive agreements, and the sites, counterparties and investment amounts were not named. The details come from Aolani's own press release.
Why it matters: Planned capacity is not the same as built capacity. Businesses weighing local AI hosting or sovereign compute options should treat this as an early signal and watch for confirmed agreements.
Source: Media OutReach Newswire (Aolani press release)
Global
Microsoft adds controls to monitor AI agents as insider-style risks
According to TechCrunch, Microsoft introduced security features that treat AI agents as a potential insider threat. They include Guardian Agents to monitor the agents a company runs, a Defender capability that can detect up to 60 types of agents staff bring to work, and access controls that adjust permissions by task and require human approval for risky decisions. A Microsoft security executive said frontier models cannot be fully trusted because they are unpredictable.
Why it matters: Staff in Malaysian firms may already run AI agents on work data. Inventorying which agents are in use and what they can access is a sensible first step, whichever vendor's tools are chosen.
Source: TechCrunch
Nvidia releases open models for enterprise agents and local PCs
TechCrunch reports that Nvidia launched new open-weight models. Nemotron 3.5 Super VL is a 120-billion-parameter model aimed at enterprise agents, which organisations can post-train on their own data. The Bolt models are designed to run locally on AI PCs and workstations. Nvidia said Super VL and the larger Bolt model are available now, and the smaller Bolt model is due later this month. Speed and accuracy comparisons are Nvidia's own claims.
Why it matters: Open models that can run on a company's own infrastructure or local machines may help Malaysian businesses with data residency concerns, though they should test performance themselves rather than rely on vendor benchmarks.
Source: TechCrunch
Amazon releases TorchNeuron to move AI apps onto its Trainium chip
Amazon Web Services released TorchNeuron, a tool built on PyTorch that helps developers move AI applications to its Trainium chip. According to Amazon, PyTorch apps can run on Trainium with a one-line configuration change. The company also said voice AI provider Deepgram now processes 70 per cent more audio per dollar after moving. TechCrunch framed the release as a challenge to Nvidia's software advantage and noted that other chips already offer similar porting.
Why it matters: More chip options could eventually affect cloud pricing for AI workloads. Malaysian teams building AI products may want to compare costs across hardware, treating vendor savings claims as unverified.
Source: TechCrunch
AI Daily News Corner is compiled with AI assistance from the sources linked above and summarised in our own words. Details can change, so check the original reports. More daily AI news · AI consulting from Agmo.
