The Scale Is Staggering
xAI operates the Colossus supercomputer with over 200,000 NVIDIA GPUs, expanding to 1.5 million. OpenAI has committed to deploying more than 10 gigawatts of NVIDIA systems. Google, Microsoft, and Amazon are all building AI-specific data centres at unprecedented scale.
Training xAI's Grok 4 alone produced over 72,000 tonnes of CO2, equivalent to driving 17,000 cars for a year.
Why Should You Care
Every time you use ChatGPT, Claude, or any AI tool, your request is processed in one of these data centres. The cost, speed, and availability of AI tools are directly tied to data centre capacity.
When NVIDIA releases new chips that deliver 35x lower cost per token (as they just did with the GB200), that translates into cheaper AI tools, faster responses, and more capable models. The infrastructure improvements don't stay abstract. They reach your business within months.
The Cost Trajectory
AI compute costs are dropping 10x roughly every 18 months. This means systems that cost £10,000 to run monthly in 2024 now cost £1,000, and will cost £100 in 2027. The implication: AI automation that was only viable for enterprise companies two years ago is now accessible to SMEs.
The Energy Question
AI data centres consume massive amounts of power. Anthropic has committed to paying 100% of grid infrastructure costs for upgrades needed for their data centres. The industry is investing in water-efficient cooling and renewable energy. But energy consumption is a legitimate concern, and "Green AI" (smaller, more efficient models) is becoming a competitive advantage.
What This Means for Your AI Strategy
The cost of AI will keep falling. The capabilities will keep rising. Building AI systems now means you're positioned to benefit from every infrastructure improvement. Waiting means catching up later at greater cost.
Build Now, Benefit Forever
OrcaScale builds AI systems that get cheaper and more capable as the infrastructure improves. Every model upgrade and cost reduction benefits your business automatically.