The AI revolution has the face of Jensen Huang and the NVIDIA logo, but behind the scenes of the media hype surrounding the launch of the Blackwell platform, a much quieter but equally fierce war is being waged. It is a battle for the billions of dollars of contracts to build servers that will become the foundation of the new economy. While the overall server market, according to the latest TrendForce research, has caught breath and stabilised, its segment dedicated to artificial intelligence is experiencing an unprecedented boom.
In this new reality, the old contract manufacturing (ODM) giants – companies such as Quanta, Supermicro, Foxconn and Wiwynn – must redefine their strategies. Victory no longer depends on production scale alone, but on strategic alliances, the ability to integrate ultra-complex systems and the trust of a handful of cloud giants who dictate terms.
Two-speed market
The catalyst that is shuffling the cards in the industry anew is undoubtedly the NVIDIA Blackwell platform. This is not simply the next, faster generation of GPUs. Systems such as the GB200 server rack or the HGX B200 platform are integrated, powerful supercomputers, the assembly and deployment of which require entirely new competencies for manufacturers. The complexity of these solutions means that the barrier to entry into the top end of the AI server market has risen sharply.
The scale of this transformation is huge. Analysts at TrendForce predict that the Blackwell family will account for more than 80% of NVIDIA’s high-end GPU shipments as early as 2025. For ODM manufacturers, this means one thing: whoever fails to master the new platform will be left behind. The question is, who is in the best starting position in this race?
The power of long-term relationships
Quanta Computer, a veteran of the ODM market, is playing the game by building on its greatest strength: decades of trust and deep integration with Silicon Valley’s biggest players. As a long-standing partner of Meta, AWS and Google, the company has seamlessly entered the Blackwell era, securing orders for GB200 and GB300 cabinets from its regular customers. What’s more, Quanta has also recently managed to win additional contracts from Oracle, demonstrating its flexibility and confirming its status as a leader to be reckoned with. Its strategy is proof that in a business based on the latest technologies, old acquaintances are still priceless currency.
Supermicro and Foxconn
Supermicro and Foxconn have adopted a completely different model. Their recent success is inextricably linked to one powerful customer – Oracle. Oracle’s aggressive expansion of AI data centres by Oracle in North America has become a powerful wheel for them. For Supermicro, as analysts point out, the AI server business has even become a ‘major growth engine’, as evidenced by the recent securing of several large projects based precisely on GB200 cabinets. It’s a high-growth, but also high-focus strategy of closely aligning its fortunes with the investment plans of a single, but extremely determined giant.
Wiwynn: An alternative route through ASICs
In this race, Wiwynn presents the most different and perhaps the most forward-looking strategy. Instead of competing solely in a field dominated by NVIDIA, the company is consciously diversifying its portfolio. Deepening its partnerships with Meta and Microsoft, Wiwynn is focusing on servers based on custom ASICs (Application-Specific Integrated Circuit). This is best exemplified by its key role as a systems provider for AWS Trainium – AI training chips designed by Amazon itself.
It’s a strategic move that allows Wiwynn to satisfy the cloud giants’ growing appetite for their own task-optimised and potentially cheaper-to-maintain silicon solutions. It also sends a signal to the market as a whole: the future of AI infrastructure does not have to belong solely to universal GPUs.
The end of universal suppliers
The landscape of the server market is changing before our eyes. The era in which ODM manufacturers were able to compete by offering universal platforms to a wide audience is coming to an end in the high-end AI segment. It is being replaced by a model based on specialisation, unique competences and almost exclusive partnerships.
In this new hand, it is the cloud giants – Oracle, Meta, AWS, Google – who have become the real ‘kingmakers’. Their strategic decisions regarding the choice of technology and production partners directly create the winners and losers in this race.
