Why everyone from OpenAI to SpaceX is building their own chips (and turning up the heat on Nvidia)
TechCrunch
•Fri, 26 Jun 2026 17:43:22 +0000
📰 What Happened
A growing wave of major technology companies, including OpenAI, Google, Apple, and SpaceX, are investing heavily in custom chip development, signaling a significant shift away from total dependence on Nvidia for AI hardware. OpenAI announced plans for a custom inference chip codenamed "Jalapeño," built in collaboration with Broadcom, designed specifically to run AI models more efficiently. The trend mirrors Apple's successful transition away from Intel chips to its own Apple Silicon, which gave the company greater control over performance and power efficiency. On TechCrunch's Equity podcast, hosts Kirsten Korosec, Anthony Ha, and Sean O'Kane analyzed what this industry-wide custom silicon push means for the competitive landscape.
The core motivation behind this shift is what industry observers call "single-supplier risk" — the vulnerability that comes from relying on one company for a critical component. Nvidia currently commands an estimated 80-90% of the AI chip market, with its H100 and B200 GPUs becoming practically essential infrastructure for training and running large language models. By developing their own chips, these companies gain more control over their hardware roadmap, can optimize silicon specifically for their unique workloads, and potentially reduce costs at scale. The move represents less a clean break from Nvidia and more a strategic hedge, as companies seek to diversify their supply chains while maintaining access to Nvidia's industry-leading hardware.
🔍 The Backstory
Nvidia's dominance in AI chips is rooted in an unexpected convergence. The company's graphics processing units (GPUs) were originally designed for rendering video games, but researchers discovered in the early 2010s that their massively parallel architecture was ideally suited for the matrix calculations underlying deep learning. This insight sparked a revolution: as AI models grew from millions to trillions of parameters, demand for Nvidia's hardware exploded. The company's market capitalization surged past $3 trillion, making it one of the world's most valuable corporations. However, concerns about dependence on a single supplier have been building. Apple's successful transition from Intel to its own M-series chips proved that custom silicon could deliver superior performance, while Google's Tensor Processing Units (TPUs) demonstrated the advantages of domain-specific architecture. The recent rise of competitors like AMD and startups like Groq has added further pressure, but Nvidia's CUDA software ecosystem — a deeply entrenched platform that AI developers rely on — remains a formidable barrier to switching. OpenAI's Jalapeño chip, focused specifically on inference (running already-trained models) rather than training, represents a targeted strategy to reduce costs in the fastest-growing segment of AI compute demand.
🎯 Why It Matters
Nvidia's near-monopoly on AI chips has given it extraordinary market power and made it one of the world's most valuable companies. If major customers like OpenAI, Google, and SpaceX succeed in developing competitive custom alternatives, it could reshape the entire AI hardware landscape, reducing Nvidia's pricing power and forcing the company to innovate more aggressively. More broadly, the custom chip trend suggests the AI industry is maturing, with companies moving beyond off-the-shelf solutions toward vertically integrated hardware-software stacks optimized for their specific needs — a pattern that historically signals the consolidation of a transformative technology.
A growing wave of major technology companies, including OpenAI, Google, Apple, and SpaceX, are investing heavily in custom chip development, signaling a significant shift away from total dependence on Nvidia for AI hardware. OpenAI announced plans for a custom inference chip codenamed "Jalapeño," built in collaboration with Broadcom, designed specifically to run AI models more efficiently. The trend mirrors Apple's successful transition away from Intel chips to its own Apple Silicon, which gave the company greater control over performance and power efficiency. On TechCrunch's Equity podcast, hosts Kirsten Korosec, Anthony Ha, and Sean O'Kane analyzed what this industry-wide custom silicon push means for the competitive landscape.
The core motivation behind this shift is what industry observers call "single-supplier risk" — the vulnerability that comes from relying on one company for a critical component. Nvidia currently commands an estimated 80-90% of the AI chip market, with its H100 and B200 GPUs becoming practically essential infrastructure for training and running large language models. By developing their own chips, these companies gain more control over their hardware roadmap, can optimize silicon specifically for their unique workloads, and potentially reduce costs at scale. The move represents less a clean break from Nvidia and more a strategic hedge, as companies seek to diversify their supply chains while maintaining access to Nvidia's industry-leading hardware.
Nvidia's dominance in AI chips is rooted in an unexpected convergence. The company's graphics processing units (GPUs) were originally designed for rendering video games, but researchers discovered in the early 2010s that their massively parallel architecture was ideally suited for the matrix calculations underlying deep learning. This insight sparked a revolution: as AI models grew from millions to trillions of parameters, demand for Nvidia's hardware exploded. The company's market capitalization surged past $3 trillion, making it one of the world's most valuable corporations. However, concerns about dependence on a single supplier have been building. Apple's successful transition from Intel to its own M-series chips proved that custom silicon could deliver superior performance, while Google's Tensor Processing Units (TPUs) demonstrated the advantages of domain-specific architecture. The recent rise of competitors like AMD and startups like Groq has added further pressure, but Nvidia's CUDA software ecosystem — a deeply entrenched platform that AI developers rely on — remains a formidable barrier to switching. OpenAI's Jalapeño chip, focused specifically on inference (running already-trained models) rather than training, represents a targeted strategy to reduce costs in the fastest-growing segment of AI compute demand.
Nvidia's near-monopoly on AI chips has given it extraordinary market power and made it one of the world's most valuable companies. If major customers like OpenAI, Google, and SpaceX succeed in developing competitive custom alternatives, it could reshape the entire AI hardware landscape, reducing Nvidia's pricing power and forcing the company to innovate more aggressively. More broadly, the custom chip trend suggests the AI industry is maturing, with companies moving beyond off-the-shelf solutions toward vertically integrated hardware-software stacks optimized for their specific needs — a pattern that historically signals the consolidation of a transformative technology.