A new pastime has taken hold in Silicon Valley. Techies now compete to see who can burn through the most AI tokens, a game they have coined “tokenmaxxing.” It sounds trivial but is not. The craze is a symptom of something deeply structural: demand for artificial intelligence is accelerating faster than the infrastructure required to support it can be built.
That gap is one of the most important investment dynamics of our time. And it is why Megaport (ASX: MP1), a long-time holding in the Flagship Investments portfolio, has become one of the more compelling positions we hold.
The Bottleneck Is Real and It Is Getting Worse
The scale of investment flowing into AI is genuinely extraordinary. The five major hyperscalers: Alphabet, Amazon, Meta, Microsoft, and Oracle, have tripled their combined capital expenditure to more than US$750 billion, with Anthropic alone announcing a US$100 billion partnership with Amazon to secure up to five gigawatts of server capacity. Yet the hardware suppliers filling those data centres, chipmakers, networking equipment providers, cooling systems manufacturers, have increased their own investment by only around half that rate, and will spend less than a third as much this year.
The consequence is a deepening supply crunch across the entire AI stack. Nvidia’s H100 GPUs are commanding rental premiums of around 30% above November 2025 levels. High-bandwidth memory, the specific chip architecture that large AI models depend on, is effectively sold out across all three major producers for the entirety of 2026. CPU shortages are emerging as a new constraint, driven by the rise of agentic AI systems, which require roughly one CPU for every GPU, compared to a ratio of one-to-twelve for traditional chatbot applications.
Meanwhile, building the physical infrastructure to run all of this is facing headwinds that capital cannot easily overcome. Over US$156 billion worth of data centre projects were blocked or delayed in America last year alone due to community opposition and litigation. Power grid constraints are tightening across the US, Europe, and Asia. And as The Economist recently observed, the fundamental asymmetry is stark: improving software takes months, whereas expanding supply chains takes years.
The bottleneck, in other words, is structural. And structural bottlenecks create enduring value for those who are already on the right side of them.
Why Megaport Sits on the Right Side
Megaport spent the better part of a decade building a global Network-as-a-Service (NaaS) platform: software-defined, on-demand cloud connectivity spanning more than 1,100 enabled locations worldwide. For years, the network was the quiet foundation of the business. What was always clear to us was that any serious AI infrastructure platform would require network quality as a precondition, not as an afterthought.
To understand why Megaport’s model matters, it helps to appreciate what Jared Pohl of ECP Asset Management aptly calls the “Hyperscaler Trap.” The major cloud providers (AWS, Google Cloud, Microsoft Azure) have built ecosystems where moving data in is straightforward but moving it out or between providers is deliberately expensive. Megaport inverts that logic by controlling the transit layer across more than a thousand data centres and over 100 cloud regions globally, it sits at the centre of the architecture rather than inside any one walled garden.
In November 2025, Megaport extended that logic further by acquiring Latitude.sh, an automated bare-metal compute platform. The combination was incisive. High-performance GPU, CPU, and storage capabilities provisioned directly onto Megaport’s private network fabric. For enterprises running latency-sensitive AI inference workloads, that integration represents a meaningful step forward in how global-scale compute can be deployed.
Crucially, Megaport’s distributed model achieves this by stitching together smaller capacity points across existing data centres worldwide; bypassing the multi-year power grid approvals and real estate constraints that are stalling traditional hyperscale construction. As Jared Pohl concluded in his recent analysis, Megaport has quietly built “a faster, cheaper, and more secure alternative to the cloud itself.”
A Step-Change in Contracted Revenue
The market has since confirmed what the strategic logic suggested. In May 2026, Latitude.sh secured three binding contracts worth US$182.9 million (approximately A$254 million) with two US-based AI technology companies, spanning GPU, CPU, network, and storage infrastructure. Around 90% of that value is tied to initial 36-month terms, providing high-quality, recurring revenue with genuine visibility.
Then, in early June, Megaport announced four further AI infrastructure contracts with a combined value of approximately A$458.9 million, alongside a fully underwritten equity raising of A$827.3 million to fund contracted infrastructure and a globally distributed, on-demand AI inference cloud. Together, these announcements represent over A$710 million in contracted AI infrastructure revenue. The Compute division’s pro forma Annual Recurring Revenue has reached A$385.2 million. Now the majority of total Group ARR of A$662.9 million.
The Foundation Remains Strong
What matters alongside the headline contract wins is the condition of the business beneath them. Megaport’s network operation continues to compound with quiet consistency. Network ARR reached A$277.7 million in April 2026, up 25% year-on-year in constant currency terms, with Net Revenue Retention rising to 113%. FY26 group revenue guidance has been tightened to A$307–315 million. The company is raising capital to press an advantage in the market not just to fill a gap.
The Long View
We are careful not to conflate a good business with an inevitable one. The inference cloud ambition, where Megaport deploys a speculative GPU pool beyond its contracted revenue floor, carries real execution risk. Capital committed today funds revenue that largely begins in 2027, in a market moving fast enough that 18 months is a meaningful horizon.
But that is precisely where long-term, research-driven investing earns its returns. The companies that built the right platforms before the demand was obvious are now in a position to earn that foresight. Megaport is one of them. We have held it because we believed in the quality of the platform and the discipline of the management team. Megaport has deliberately built itself into the industry, proving that the infrastructure constraints underpinning AI demand are structured and not just a temporary roadblock.
