The scramble to build artificial intelligence has produced an awkward, expensive secret inside many corporate data centres: the graphics processing units that cost a fortune to buy and power frequently sit idle, waiting for work. Nutanix, the American hybrid-cloud and infrastructure company with a sizeable footprint across Australia and New Zealand, is betting that fixing that inefficiency is where the next phase of enterprise AI will be won.
The company has acquired Ryax, a French software startup whose technology orchestrates data and AI pipelines with an eye to keeping GPUs busy. According to CRN Australia, the deal is designed to lift GPU utilisation and to sharpen Nutanix’s push into agentic AI, the category of software in which autonomous agents chain together tasks, call tools and act with a degree of independence rather than simply answering a single prompt.
Why idle silicon is the problem worth solving
To understand the logic of the acquisition, it helps to understand the waste it targets. Training and running AI models relies on accelerators from the likes of Nvidia, and those chips are both scarce and dear. Yet in practice a GPU is rarely pinned at full tilt. Jobs queue unevenly, data has to be shuffled into place, pipelines stall waiting on upstream steps, and whole clusters can idle between workloads. Every one of those gaps is depreciation and electricity spent on nothing.
Ryax’s pitch sits squarely in that gap. Its platform orchestrates the messy middle of AI work, the movement of data, the scheduling of jobs and the stitching together of pipelines, so that the expensive hardware spends more of its time actually computing. Folding that capability into Nutanix’s stack lets the larger company tell customers a tidy story: buy fewer GPUs, or get more out of the ones you already have, while standing up the agentic applications that boards are now demanding.
That story dovetails with Nutanix Enterprise AI, the company’s platform for running models and AI services across on-premises kit, the edge and public clouds. Nutanix has spent the past two years repositioning itself from a virtualisation and hyperconverged infrastructure vendor into an AI-ready infrastructure player, a shift given extra momentum by the uncertainty that followed Broadcom’s takeover of VMware and the licensing upheaval that came with it. Acquiring orchestration smarts is a way of moving up the value chain, from running the plumbing to managing the workloads that flow through it.
Two ways to read the deal
The optimistic reading is that this is exactly the kind of tuck-in acquisition the market rewards. GPU efficiency is a genuine pain point, the technology is complementary rather than overlapping, and agentic AI is the term every enterprise software vendor wants on its roadmap. By owning more of the orchestration layer, Nutanix reduces its customers’ reliance on cobbling together open-source tools and cloud-native services, and it gives its channel partners something concrete to sell beyond raw infrastructure.
The more sceptical reading is that “boosting GPU utilisation” has become one of the most crowded claims in the industry. Nvidia itself, the big hyperscalers, Kubernetes-based schedulers and a long list of startups all promise to carve waste out of accelerator fleets. Ryax is a small European company, and integrating its software into Nutanix’s platform, then convincing conservative enterprise buyers to route their AI pipelines through it, is a multi-year job rather than a quarterly win. Acquisitions of this size succeed or fail on execution and integration, not on the press release. There is also the perennial question of whether efficiency gains get reinvested in more ambitious models rather than banked as savings, a pattern that has played out repeatedly as computing gets cheaper.
What it means for Australia
For Australian organisations, the economics here are not abstract. Local enterprises and government agencies are building AI capability against a backdrop of constrained access to high-end GPUs, elevated power prices and a national conversation about sovereign AI infrastructure that has run through FluentSea’s coverage all year, from sovereign inferencing ventures to the data-centre buildouts reshaping Sydney, Melbourne and the regions. When accelerators are both expensive and hard to get, software that extracts more useful work from each chip is effectively a way of adding capacity without pouring more concrete or drawing more megawatts.
Nutanix sells heavily into Australian enterprise and the public sector, often through the channel partners and managed service providers that CRN covers, so a capability that lands in Nutanix Enterprise AI tends to reach Australian data centres on a relatively short lag. For a chief information officer in Brisbane or Canberra weighing whether to keep AI workloads on local, sovereign infrastructure rather than defaulting to an offshore hyperscaler, better utilisation changes the sums. It narrows the cost gap between running your own accelerators and renting someone else’s, which is one of the quiet variables in every sovereignty debate.
There is a sustainability angle too. Australian data centres are under growing scrutiny over their energy and water appetite, and regulators and boards alike are asking harder questions about the carbon cost of AI. Getting more computation out of each watt is not just a financial argument; it is part of how the sector answers its critics. Any tooling that lifts the ratio of useful work to energy consumed will find a receptive audience among operators trying to square AI ambitions with net-zero commitments.
What happens next
Financial terms of the acquisition were not disclosed, which is typical for a deal of this scale. The real test will come as Nutanix integrates Ryax’s technology into its platform and lays out how it will be packaged and priced for customers, likely at the company’s next major product milestones. Watch for whether the GPU-utilisation claims are backed by independent benchmarks rather than vendor figures, and whether Australian partners start bundling the capability into sovereign AI and private-cloud offers.
The broader signal is clear enough. The AI infrastructure race is shifting from a simple contest over who can amass the most silicon to a subtler fight over who can use that silicon most efficiently. For a country where every GPU is precious and every megawatt contested, that is a shift worth watching closely.
Sources: CRN Australia.
Comments 1