Megaport weighs A$827 million raising to fund Nvidia GPU pool

Megaport is reportedly considering an A$827 million entitlement offer, fully underwritten, to finance a shift from linking clouds and data centres to supplying the computing power that AI workloads need. The company has proposed putting about A$350 million into a distributed pool of Nvidia GPUs. The Australian Financial Review reported that the plan involves sizeable US contracts in AI infrastructure, and that Megaport has reportedly signed four of them, together worth A$458.9 million.

Should it go ahead, the raising would rank among the biggest follow-on equity offerings in Australia in 2026. NextDC recently finished an A$1.1 billion raise, which puts Megaport's reported figure just below it. Megaport's shares have gained roughly a third this year and the company is valued at $3.8 billion.

Megaport operates a software-defined network that reaches more than 1,100 data centres, and customers use it to join cloud services, facilities and private infrastructure. It now wants to site distributed GPU capacity near those connected environments. A GPU pool tied to the virtual network would give customers AI capacity at several locations, and could spare them from buying hardware and organising connectivity separately. The design is also intended to send more traffic over Megaport's current infrastructure.

The strategy brings risks that sit outside the core connectivity business. Nvidia accelerators are costly and power-hungry, and they can leave an owner exposed if demand, pricing or hardware generations move faster than expected. Megaport would need to handle procurement, deployment and utilisation while keeping its existing network reliable. Other variables include depreciation, supply dependencies, limits on energy availability and uncertain resale values for specialised parts. Headline contract values point to early demand but say nothing about when revenue arrives, how much capital delivery needs or what the GPUs will eventually return.

Gartner forecasts global data-centre systems spending of $653.4 billion in 2026, up 31.7 per cent year on year, with server spending expected to climb 36.9 per cent on AI demand. Training and inference workloads shift large datasets between storage, GPU clusters and cloud regions, and 400G and 800G Ethernet is becoming standard. The TMC Insight analysis adds that remaining purely a connectivity provider carries strategic risk as value moves toward integrated AI infrastructure.

The raising would show whether investors regard distributed GPU capacity as a natural extension of the platform or as a capital-heavy break from the model that built its valuation.