Moving from connecting data centers to owning the computing power itself
In a moment when artificial intelligence has moved from laboratory curiosity to industrial necessity, Megaport — an Australian company long content to lay the pipes of the internet — has stepped forward to own the engines running through them. The company secured four contracts with major U.S. technology firms worth roughly A$458.9 million and announced a A$827.3 million capital raise to build a globally distributed inference cloud spanning more than 1,100 data centers across 31 countries. This is not merely a business expansion; it is a declaration that the next frontier of the AI economy belongs not to those who train the models, but to those who can run them — reliably, quickly, and close to where the world actually works.
- Enterprise demand for AI inference — running models at scale for real users — has surged past available GPU supply, creating a scramble among infrastructure providers to fill the gap.
- Megaport is issuing new shares at a 14 percent discount to raise A$827.3 million, a bold and dilutive move that signals how urgently the company believes it must act before competitors claim the same ground.
- Four contracts with unnamed U.S. tech giants, valued at A$458.9 million and beginning in H1 2027, give Megaport a concrete revenue anchor — but also a hard deadline requiring A$369.5 million in capital expenditure, mostly on scarce Nvidia GPUs.
- The company is pivoting from connectivity middleman to compute operator, a fundamental shift in identity that analysts are framing as a 'picks-and-shovels' play in the AI infrastructure boom.
- Trading in Megaport shares was halted until June 5, leaving the market to weigh whether sustained enterprise AI adoption and GPU availability will validate a bet of this scale.
Megaport, the Australian infrastructure company best known for connecting data centers and cloud platforms, announced Wednesday that it is raising A$827.3 million and has secured four contracts with major U.S. technology firms worth A$458.9 million — a twin move designed to plant it firmly at the center of the AI compute economy.
The contracts are inference deals, not training ones. Where training involves building AI models from scratch at enormous cost, inference is the ongoing work of running those models for real customers at speed and scale — the difference, as one framing goes, between constructing a factory and operating it at full capacity. The agreements begin in the first half of 2027 and will require roughly A$369.5 million in capital expenditure, the bulk of it going toward high-performance Nvidia GPUs and the networking infrastructure to support them.
Megaport's strategic logic rests on its existing footprint: more than 1,100 data centers across 31 countries. Rather than forcing enterprises to route AI workloads to distant cloud regions, the company intends to place GPU capacity close to where the work actually happens — reducing latency, easing power constraints, and offering an alternative to the bottlenecked global chip supply. A dedicated A$250 million GPU pool will be available to enterprise customers through both fixed contracts and pay-as-you-go pricing.
To fund the expansion, the company is issuing new shares at A$10.25 each, a discount of roughly 14 percent from its June 1 close. Shares were halted from trading until June 5. Megaport also narrowed its 2026 revenue forecast to between A$220 million and A$226 million, a modest tightening that suggests growing internal confidence.
For years, Megaport profited by being the plumbing of the internet. Now it is moving up the value chain, owning and operating the hardware enterprises increasingly depend on. Whether that bet pays off hinges on two things it cannot fully control: the pace at which companies move from AI experimentation to full production deployment, and whether GPU supply remains accessible enough to build on.
Megaport, the Australian infrastructure company, announced Wednesday that it has locked in four contracts with major U.S. technology firms and will raise $594 million to build what it calls an inference cloud—a globally distributed network designed to deliver AI computing power closer to the companies that need it.
The four deals are worth roughly $329 million in Australian dollars and represent a significant bet on where the AI market is heading. These aren't training contracts, where companies build massive models from scratch. These are inference deals—the work of running those models at scale, quickly, for real customers. It's the difference between building a factory and running it at full capacity. The contracts are set to begin in the first half of 2027 and will require Megaport to spend about $265 million on equipment, mostly high-performance Nvidia GPUs, plus the networking and storage infrastructure to support them.
Megaport operates a network of more than 1,100 data centers spread across 31 countries. The company's pitch is straightforward: instead of routing AI workloads across the globe to distant cloud regions, enterprises can access GPU capacity from a pool that sits much closer to where the work actually happens. This matters because latency—the delay in getting data to the processor and back—can be the difference between a useful AI application and a sluggish one. Power availability and access to scarce GPU chips are also bottlenecks that Megaport believes it can help solve.
To fund this expansion, Megaport is issuing new shares at $10.25 each, a discount of roughly 14 percent from where the stock closed on June 1. The company will raise about $594 million this way. It's also setting aside $250 million specifically for the GPU pool itself, which will be offered to enterprise customers through a mix of fixed contracts and pay-as-you-go pricing.
This move signals a shift in Megaport's business model. For years, the company has made its money by connecting data centers and cloud platforms—essentially acting as a middleman in the plumbing of the internet. Now it's moving up the value chain, actually owning and operating the computing hardware that enterprises need. Hebe Chen, an analyst at Vantage Markets, described Megaport as a "picks-and-shovels player in the AI gold rush," borrowing the old metaphor for companies that profit from infrastructure during booms rather than from the boom itself.
The timing reflects a broader scramble among infrastructure providers to capture a piece of the AI compute market. Demand for GPU capacity has far outpaced supply, and companies across industries are desperate for access. Megaport's existing network gives it an advantage—it already has relationships with data center operators and cloud providers worldwide. The question now is whether it can execute the buildout quickly enough and whether demand for inference workloads will sustain at the levels the company is betting on.
Megaport also tightened its revenue forecast for 2026, now expecting between $220 million and $226 million, up from its previous range of $216 million to $227 million. The company's shares were halted from trading until June 5, when the market would digest the capital raise and the new contracts. What happens next depends partly on how quickly enterprises shift from experimenting with AI to deploying it in production—and whether Megaport can deliver the speed and reliability they'll demand.
Citações Notáveis
Megaport is positioning itself as a picks-and-shovels player in the AI gold rush—benefiting from the next wave of AI infrastructure buildout.— Hebe Chen, market analyst at Vantage Markets
The GPU pool could push Megaport further up the value chain, from simply connecting data centres and cloud platforms to helping enterprises access the computing power needed to deploy AI at scale.— Hebe Chen, market analyst at Vantage Markets
A Conversa do Hearth Outra perspectiva sobre a história
Why does it matter that these are inference contracts rather than training contracts?
Training is the expensive, one-time work of building a model. Inference is the ongoing work of using it. If you're Megaport, inference is where the recurring revenue lives—it's the difference between a single big project and a steady stream of customers running workloads every day.
So Megaport is betting that enterprises will actually deploy AI at scale, not just experiment with it.
Exactly. Right now a lot of companies are still in the pilot phase. But these four contracts suggest that at least some major U.S. tech firms believe the market is ready to move from "let's try this" to "we're running this in production." Megaport is betting that belief is correct.
What's the advantage of having data centers in 31 countries?
Latency. If your AI model is running in a data center on the other side of the world, there's a delay every time data travels to it and back. For some applications, that delay is acceptable. For others—real-time recommendations, fraud detection, interactive systems—it kills the user experience. Megaport's argument is that by having GPU capacity distributed globally, they can put compute closer to where it's actually needed.
Is this risky? What if enterprise AI adoption doesn't accelerate the way they're expecting?
Very risky. They're committing $265 million in capital expenditure on the assumption that demand will be there. If enterprises stay in pilot mode longer than expected, or if GPU supply loosens and prices drop, Megaport could find itself with expensive hardware and fewer customers willing to pay premium prices for it.
Why the 14 percent discount on the share offering?
It's a sweetener to make the deal attractive to investors. The company needs the capital quickly to start building out the infrastructure before the contracts begin in early 2027. A discount encourages existing shareholders to buy more shares and attracts new investors who might otherwise wait.
What does "picks and shovels" really mean in this context?
During gold rushes, the people who got rich weren't always the miners—they were the people selling picks and shovels to the miners. Megaport isn't trying to build AI models or run AI applications. It's selling the infrastructure that other companies need to do those things. If AI adoption explodes, Megaport profits regardless of which companies win or lose.