NVIDIA’s next-generation GB300 AI servers are slated to begin shipping in the second half of this year, according to sources deep within the supply chain. 
These new servers, built on the powerful Blackwell architecture, are expected to dominate the AI infrastructure landscape – even overshadowing the buzz surrounding Apple’s next iPhone.
With up to 72 Blackwell AI GPUs in its highest-end configuration, the GB300 platform aims to be the most powerful AI server setup on the planet. Leading Taiwanese manufacturers, including Foxconn, Quanta, Wiwynn, and Inventec, are already in full swing to fulfill NVIDIA’s massive production demands. Foxconn, in particular, has secured the bulk of these orders, continuing its reign as NVIDIA’s key assembly partner.
Foxconn expects a smooth transition from GB200 to GB300 production, thanks to architectural similarities between the two systems. The company anticipates AI server revenue to surpass 50% of its total server income – a major shift driven by the growing hunger for high-performance computing solutions in the AI arms race.
Meanwhile, Quanta Computer has already begun GB200 server shipments and is currently testing the GB300 with enterprise clients. The firm expects to commence deliveries by September. However, Quanta’s management acknowledges that despite GB300’s superior processing power, many AI firms might not wait due to urgent model training timelines.
Other manufacturers like Wiwynn and Inventec are also targeting a September rollout. Shipment volumes are expected to surge in H2 2025, possibly enabling these firms to grab larger chunks of the global AI hardware market.
Amid NVIDIA’s growing dominance, tech giants are still experimenting with alternatives to escape the bottleneck of high costs and limited supply. Amazon and Alphabet are developing their own AI chips, while Broadcom and Marvell continue to receive custom silicon orders. Still, CEO Jensen Huang remains unfazed, claiming NVIDIA’s offerings outperform all rivals in real-world workloads.
That said, reports suggest even OpenAI is starting to lean on Google’s TPUs, seeking relief from NVIDIA’s steep pricing. While the GB300’s arrival is sure to shake the market, the broader AI hardware ecosystem is clearly evolving into a multi-lane race.