A fierce struggle for computational self-sufficiency is reorganizing the global chip architecture as tech conglomerates attempt to decouple from Nvidia’s overwhelming 95% market monopoly. Rather than waiting on external supply chains, hyperscalers are leveraging massive internal workloads to fund proprietary silicon lifelines.
The chip market has subsequently evolved into a highly paradoxical two-front war. Hyperscalers are simultaneously committing multi-billion-dollar infrastructure outlays to survive the immediate generative AI race, all while trying to create the long-term, custom-silicon pipelines intended to eventually make those exact same Nvidia investments obsolete.
Big tech has been relying on high performing chips for several gadgets. Last May, Apple introduced the M4 chip in the latest iPad Pro, marking a significant performance and efficiency upgrade. That’s when big tech companies started struggling for self-sufficiency with regards to the chip.
Hence, the pivot to manufacturing their own chip.
Independence from Nvidia: Pivot to Custom Silicon
In July, OpenAI CEO Sam Altman was talking to chip designers like Broadcom about a new chip venture to reduce OpenAI’s dependence on Nvidia. OpenAI would have access to much more computing power with their own chip, which it will need. OpenAI has been pushing ahead on its plan to reduce its reliance on Nvidia for its chip supply by developing its first generation of in-house AI silicon.
American companies have been building up to match Nvidia’s chips. Last September, Intel’s foundry, or contract manufacturing business, signed up Amazon’s cloud services unit as a customer for making custom AI chips.
June last year, AMD unveiled its latest AI processors and detailed its plan to develop AI chips over the next two years in a bid to challenge Nvidia.
Meta is actively deploying four generations of its custom MTIA processors through 2027. Developed with Broadcom and fabricated by TSMC, this initiative aims to diversify Meta’s hardware supply and optimize power efficiency for its heavy internal data center workloads. The company began testing its first in-house chip for training AI systems in March, a key milestone as it moves to design more of its own custom silicon and reduce reliance on external suppliers like Nvidia.
Apple has been bypassing Nvidia entirely to train foundational software infrastructure on Google-designed silicon. Apple relied on chips designed by Google rather than industry leader Nvidia to build two key components of its AI software infrastructure for its forthcoming suite of AI tools and features, an Apple research paper showed.
Google has been working on a new initiative to make its AI chips better at running PyTorch, the world’s most widely used AI software framework. The move is aimed at weakening Nvidia’s longstanding dominance of the AI computing market, according to Reuters.
OpenAI is aggressively co-developing in-house AI hardware alongside Broadcom. Anthropic has been weighing building its own AI chips. AWS introduced Graviton5, the company’s most powerful and efficient CPU.
This year, Samsung revealed plans to start production of its next-generation high-bandwidth memory chips, or HBM4, and supply them to Nvidia. Gates Frontier, the venture arm of Microsoft, invested US$10 M in US-based startup Neurophos’ optical processing unit, or OPU chip.
This aggressive pivot to custom silicon, symbolized by Amazon Web Services introducing its ultra-efficient Graviton5 server processors and anchoring a strategic agentic-AI framework partnership with Meta, are designed to systematically dilute Nvidia’s pricing leverage and mitigate the existential risks of a single-vendor hardware dependency.
Meanwhile, last year, startups like Tenstorrent, Cerebras systems, D-Matrix, Groq, SiMa.ai, and Ola, took on Nvidia’s dominance in AI hardware by creating highly efficient AI chips that claimed superior performance, lower power consumption, and cost effectiveness competing directly with the GPU giant in data centers, edge computing, and AI inference tasks. As per AIM, NVIDIA should be worried.
Tying Up to Defy the Chip Dominance
Hyperscalers are even coming together in their bid to diminish the Nvidia monopoly. Meta signed an agreement with AWS to power agentic AI on AWS Graviton Chips. Meta is now one of the largest Graviton customers in the world.
Intel is joining Elon Musk’s Terafab AI chip complex project with SpaceX and Tesla to make processors powering the tech billionaire’s robotics and data center ambitions. SpaceX and Tesla have plans to ?build two advanced chip factories at a sprawling facility in Austin, Texas – one to power cars and humanoid robots, and another designed for artificial intelligence data centers in space, CEO Elon Musk said.
In January, in a reflection of the AI industry’s growing hunger for computing power to run applications like ChatGPT, Microsoft signed a $9.7 billion cloud services contract with data center owner and operator IREN that gives Microsoft access to Nvidia’s GB300 processors over a five-year period.
But Will They Succeed?
While Google, Microsoft, AWS, Apple, and Meta, alongside Intel and AMD, are investing heavily in developing custom chips to run AI workloads, experts believe they will never be able to catch up with Nvidia, as it controls a whopping 95% of the AI chip market. ?
Read more: Chip In: How Beijing Turned Western Restrictions into Domestic Market Gold
This push for hardware independence is widespread but do these companies really possess structural soundness of foundational manufacturing and software prowess? While startups like Cerebras and tech titans like Intel build cheaper hardware alternatives, or try to bypass Nvidia by optimizing open-source frameworks like PyTorch, Nvidia continues to dominate the market by moving on to next-generation component pipelines, such as securing Samsung’s highly coveted HBM4 memory allocation.
The market has subsequently evolved into a highly paradoxical two-front war. Hyperscalers are simultaneously committing multi-billion-dollar infrastructure outlays, such as Microsoft’s massive $9.7 billion server contract to access Nvidia’s elite GB300 processors, to survive the immediate generative AI race, all while trying to create the long-term, custom-silicon pipelines intended to eventually make those exact same Nvidia investments obsolete.
