Trip report: Fastly Xcelerate Sydney 2026

Fastly's Xcelerate conference occurred in Sydney on 20 August 2026.

Why was Virtified there?

You might wonder why we would attend a Fastly conference. The answer is simple: we're always looking for significant technology trends and developments.

1. Distributed infrastructure: What trends might influence this 'mid-cloud' distributed architecture'? We look for trends in the space between centralised hyperscalers and 'the edge' (be that a device, an enterprise data centre or anything in between). Although the necessity is clear—for gaming, online retail, streaming and other content-heavy workloads—this architecture is yet to find a problem needing to be solved for most enterprises (examples: financial services and public sector).
   

2. WebAssembly: Is anything changing with Wasm? We continue to track the potential of WebAssembly (Wasm). This includes Fastly's pioneering and continuing use. In broad strokes, Wasm is a technology that could rival containers whilst also abstracting the application runtime. But Wasm has limited adoption beyond the browser.

Not an official mascot, but these cute plushies were offered to those attending the conference in Sydney's Sofitel hotel.

Who Fastly serves today

The event included speakers from Stan, New Relic, freelancer.com and amazee.io. These are solid examples within Fastly's area of core competence (and, admittedly, not our area of focus).

Where Fastly could go

Once we get through the hype, the future is AI. In that light, Fastly co-founder Tyler McMullen clearly differentiated what Fastly wasn't planning. Quote:

"We're not doing GPUs at the edge".

Reasons that we agree with include that GPUs are too expensive, are in high demand, are in short supply and they're not necessary in that mid-cloud architecture between frontier models and on-premises inference. He didn't discount that this plan may change; we assert that it will (and before 2030). What Fastly and its competitors have is fast connectivity to hyperscalers and distributed points of presence.

In an anywhere-to-anywhere architecture, organisations need a control point

The similarities to zero trust architecture (ZTA) are readily apparent (verify, don't trust; segment; assume the worst among other principles). But these are, by design, concepts related to secure networking. Determining intent and enforcing AI policy will be key to charting the path for inference beyond the hyperscale data centre. Call it an AI gateway, a policy agent. Other drivers include:

• The need for predictable AI costs through routing of queries originating outside the perimeter to appropriate models^.

• Targeted use cases that don't need frontier models. Small and micro models show the other end of today's AI world. Models such as Qwen and Phi. A simple demo showed the potential for mid-cloud-architecture to use natural language processing for language translation, for efficiently filtering inappropriate queries and for applying business logic.

• Physical AI wherein planet-scale deployment would demand a multi-tiered approach.

A non-update on WebAssembly

Virtified did not find any major changes in how Fastly uses Wasm. It remains intrinsic to their platform and formed part of the demo mentioned above, showing how Fastly could play the role of AI gateway.

^ It's arguable that public cloud suffered the same challenge yet found ways to sufficiently sidestep the issue through discounted bulk-buying. Perhaps experience with FinOps and cloud expense management will expand to cover the variable portion of AI expenditure. That would make sense...in theory.