Microsoft’s ambitious in-house AI chip program, a critical part of its strategy to reduce a costly dependence on NVIDIA, has suffered a major setback. The company’s next-generation chip, codenamed ‘Braga,’ has been delayed by at least six months, pushing mass production into 2026, according to The Information. More critically, sources familiar with the project claim the chip is now expected to “fall well short of the performance” of NVIDIA’s current-generation Blackwell processors, which are already shipping.
The delay is a significant blow to Microsoft, which has invested billions in its partnership with OpenAI and is racing to build out the massive infrastructure required to power the AI revolution. The internal project has reportedly been hampered by unanticipated design changes requested by OpenAI, staffing constraints, and high turnover. This stumble not only amplifies Microsoft’s reliance on NVIDIA’s expensive and often supply-constrained hardware but also puts it on the back foot as rivals accelerate their own custom silicon plans.
This project has evolved significantly since early reports in 2023 pointed to a chip codenamed ‘Athena’. While Microsoft did unveil its first-generation Maia 100 chip later that year, it was largely designed for image processing and has seen limited use in powering the company’s generative AI services. The delayed Braga chip was meant to be the true answer to NVIDIA, but it now risks being obsolete before it even reaches data centers.
An Industry-Wide Push For Silicon Independence
Microsoft’s struggle highlights the immense difficulty of designing cutting-edge chips, yet its effort is part of a crucial, industry-wide trend. Tech giants are pouring billions into developing proprietary AI accelerators to gain control over their technological destiny and escape NVIDIA’s market dominance. Amazon Web Services (AWS) has been a key player, unveiling its second-generation Trainium2 chips and partnering with AI firm Anthropic to build massive compute clusters.
Similarly, Apple is significantly broadening its custom silicon horizons with its ‘Baltra’ server chip project, an initiative aimed at powering its own data centers. Meta continues to develop its own MTIA processors and even attempted an $800 million acquisition of South Korean chipmaker FuriosaAI, which was ultimately rejected. T
his strategic alignment is also evident in the formation of the Ultra Accelerator Link (UALink) Promoter Group, an industry alliance including Microsoft, AMD, Google, and Meta that aims to create an open standard to challenge NVIDIA’s proprietary interconnects.
NVIDIA’s Unrelenting Pace and Rival Advances
While its would-be rivals work to build alternatives, NVIDIA is only picking up speed. The company is ramping up production of its Blackwell AI server racks after resolving early technical issues. CEO Jensen Huang recently described the rollout as “Blackwell is ‘the fastest product ramp in our company’s history, unprecedented in its speed and scale.”
This momentum is reflected in the market, with partners like Dell reporting a surge in orders for its AI-optimized servers that use NVIDIA, AMD, and Intel chips. Dell COO Jeff Clarke confirmed the company is seeing “strong momentum” with a significant pipeline of interest from a diverse set of customers.
NVIDIA’s aggressive roadmap, which includes the ‘Vera Rubin’ architecture planned for 2026, means that by the time Microsoft’s Braga chip is available, NVIDIA will likely be two generations ahead.
The situation perfectly captures a sentiment Huang has previously expressed about the futility of such efforts: “What’s the point of building an ASIC if it’s not going to be better than the one you can buy?” The competitive pressure on Microsoft is not just coming from NVIDIA.
Rival AMD recently launched its most direct assault yet with its Instinct MI350 series. At the launch, AMD CEO Lisa Su declared, “This is the beginning, not the end of the AI race.” The market is further fragmented by specialized, well-funded startups like Groq, which recently launched an aggressive campaign to offer its high-speed inference chips directly to developers.
Geopolitics and the Fractured Supply Chain
These internal and competitive pressures are compounded by a volatile geopolitical landscape that has fractured the global semiconductor supply chain. After the U.S. government blocked exports of NVIDIA’s H20 chip to China, Huawei seized the opportunity to announce its next-generation Ascend 920 processor, a move that led one analyst to predict that “Chinese companies are just going to switch to Huawei.”
Recent analysis of Huawei’s latest notebook confirms that its domestic manufacturing partner, SMIC, is producing chips on a 7nm process, several generations behind industry leader TSMC. According to TechInsights, the situation shows that “US-imposed technology controls are likely continuing to impact SMIC’s ability to catch up to current foundry leaders.”
This gap illustrates the tangible impact of US export controls, which may be tightened further, with officials reportedly considering revoking licenses that allow foreign firms to supply their existing fabs in China with American technology. This complex environment underscores why access to cutting-edge fabrication from foundries like TSMC remains a critical, and politically charged, component of the AI arms race.
Microsoft’s internal stumbles with the Braga chip are therefore happening at the worst possible time. As the entire industry accelerates its push for custom silicon, the competitive landscape becomes fiercer by the month, and the global supply chain grows more complex. For a company betting its future on AI leadership, falling further behind in the foundational hardware race is a costly and dangerous proposition.
