The race to become robotics’ dominant chipmaker is already underway, and the winner may not be who you expect. While Nvidia commands the AI computing landscape, two chip designers are quietly positioning themselves as the essential infrastructure for the physical AI revolution: Qualcomm and Arm Holdings. Qualcomm unveiled its ambitious full-stack robotics platform at CES 2026, while Arm created an entirely new Physical AI business unit to capture the emerging market. These moves echo Nvidia’s playbook from a decade ago—not selling the robots themselves, but the brains and building blocks that make them possible.
Figure, Kuka Robotics, Boston Dynamics, and others are already integrating these platforms, suggesting the infrastructure play in robotics is moving from speculation to execution. Qualcomm’s announcement of its IQ10 platform, designed to scale from household robots to full-size humanoids, marks the moment when robotics infrastructure became a board-level priority at major semiconductor companies. Unlike Nvidia, which dominates through raw compute horsepower, these competitors are pursuing vertical integration—offering not just chips but software, algorithms, and pre-built partnerships that compress the development timeline for robotics companies. This is quieter than Nvidia’s splash into AI, but potentially more consequential for robots actually reaching production.
Table of Contents
- Who Are the Infrastructure Providers Reshaping Robotics Hardware?
- Why the Physical AI Market is Attracting Serious Capital Investment
- Strategic Partnerships Are Building the Ecosystem Effects That Matter
- How Enterprises Are Planning Robotics Adoption in 2026 and Beyond
- The Hidden Fragmentation Risk That Could Splinter the Market
- How Chips for Robotics Differ From AI Data Center Infrastructure
- What Success Looks Like in the Next 24 Months
- Conclusion
Who Are the Infrastructure Providers Reshaping Robotics Hardware?
Qualcomm’s entry into robotics arrived fully formed rather than experimental. The company announced partnerships with eight major players simultaneously—Figure, Kuka Robotics, Advantech, APLUX, AutoCore, Booster, Robotec.ai, and VinMotion—all committing to the IQ10 platform. This wasn’t a single technology; it was an ecosystem. Advantech, a major industrial computing company, already began shipping IQ9-powered edge AI systems in 2025, including the AIR-055 sensor module and ASR-A503 robotic controller. These aren’t research projects; they’re shipping products. The comparison to nvidia‘s early CUDA strategy is apt: Nvidia didn’t force adoption through brute force superiority alone. Instead, it built software ecosystems and partnerships that made it the obvious choice. Arm’s approach differs in scope but matches the ambition.
By creating a dedicated Physical AI business unit under Drew Henry, Arm signaled that robotics isn’t an afterthought to smartphone chips anymore. Boston Dynamics, owned by Hyundai, already uses Arm chip designs in its Atlas humanoid—a vote of confidence from one of the world’s most advanced robotics companies. Caterpillar, LG Electronics, and NEURA Robotics have all committed to Arm’s ecosystem. The advantage here is design flexibility; Arm’s licensing model allows manufacturers to customize chips for specific robotic applications rather than purchasing fixed-function hardware. This is more decentralized than Qualcomm’s approach, which matters for companies building specialized robots. The limitation both face is that neither company has Nvidia’s installed base of AI developers already trained on its tools. Nvidia spent 15 years building CUDA expertise; robotics infrastructure competitors have maybe two years to prove their platforms are worth the learning curve. Early adopters like Kuka and Boston Dynamics will drive ecosystem effects, but fragmentation risk remains. If three different platforms emerge, manufacturers face the exact problem that slowed robotics adoption for decades: incompatibility and vendor lock-in.
Why the Physical AI Market is Attracting Serious Capital Investment
Market projections reveal the scale of what’s at stake. Physical AI was valued at $1.50 billion in 2026 but is projected to grow to $15.24 billion by 2032—a compound annual growth rate of 47.2%. That trajectory is comparable to cloud computing’s early boom or mobile’s rise, and it’s drawing every major semiconductor company’s attention. The growth target is clear: industrial robots, humanoids, autonomous vehicles, and home robots collectively represent a hardware addressable market worth orders of magnitude more than AI accelerators alone. Enterprise adoption is already moving beyond pilots. CIOs expect to begin with targeted deployments in factories and warehouses before broader scaling—this is the beachhead strategy. Qualcomm’s Advantech partnership exemplifies this: industrial automation is the profitable, unglamorous starting point. Factories aren’t buying robots for novelty; they’re solving labor shortages and optimizing production.
This is different from consumer AI, where adoption was driven by free software (ChatGPT) creating its own demand. Physical robots require capital investment, integration, and ongoing support. The companies that control the infrastructure layer—the chips, software frameworks, and development tools—will capture a disproportionate share of this value. However, there’s a significant constraint: the physical AI market depends on breakthroughs in robotics hardware itself, not just chips. Qualcomm and Arm can provide the brains, but if humanoid arms remain fragile, if wheeled bases can’t navigate real-world terrain consistently, or if power consumption stays prohibitive, the infrastructure market doesn’t scale. Neither company controls these bottlenecks. They’re betting that others (Figure, Tesla, Boston Dynamics, Kuka) will solve the mechanical problems while they provide the computational backbone. This dependency is a warning: infrastructure plays work only if the applications actually reach maturity. Betting on robotics infrastructure is a bet that robots will work at scale, which remains unproven for many use cases.
Strategic Partnerships Are Building the Ecosystem Effects That Matter
Qualcomm’s partnership announcement included eight companies across different robotics verticals: household robots (Figure), industrial robots (Kuka), edge computing (Advantech), autonomous systems (AutoCore), and others. This breadth mirrors Nvidia’s approach with CUDA, where initial adoption came from multiple industries using the same underlying tech. Kuka, one of the world’s largest industrial robotics manufacturers, committing to IQ10 is significant. Kuka sells to automotive assembly plants, electronics manufacturers, and logistics companies. If Kuka’s next generation of robots runs on Qualcomm’s platform, that’s potential deployment in thousands of factories worldwide. Arm’s ecosystem includes Boston Dynamics, which, while not a high-volume manufacturer, carries enormous strategic weight. Boston Dynamics’ Atlas humanoid is often considered the world’s most advanced general-purpose robot. If it runs on Arm chips, it validates Arm’s platform for humanoid applications.
LG Electronics also matters here; LG manufactures appliances and consumer electronics that increasingly include robotic components. CaterpillaR’s involvement points toward heavy equipment and autonomous vehicles. Unlike Qualcomm’s approach, which is more focused on AI and autonomous systems, Arm’s partnerships span household, industrial, and heavy equipment domains. The risk, however, is that partnerships don’t guarantee adoption across an entire industry. Kuka committing to Qualcomm IQ10 doesn’t stop it from also developing internal solutions or hedging bets on alternative platforms. Boston Dynamics using Arm chips doesn’t prevent it from customizing Arm designs for its specific needs rather than procuring off-the-shelf chips. Partnership announcements are often wins for both sides individually but don’t guarantee industry consolidation around a single infrastructure layer. The semiconductor industry has a history of multiple standards coexisting—x86 and ARM in computing, for example—and robotics may follow a similar pattern.
How Enterprises Are Planning Robotics Adoption in 2026 and Beyond
Enterprise deployment of physical AI is following the pharmaceutical industry’s model: targeted pilots before broad rollout. CIOs are starting with warehouses and manufacturing floors because the ROI is calculable. A warehouse automation project might reduce labor costs by 30% over five years—measurable, achievable, and defensible to the board. Humanoid robots in homes remain speculative; deployment in factories is pragmatic. This explains why Qualcomm’s Advantech partnership is significant—it targets industrial controllers, not consumer gadgets. The timeline matters. Factories don’t retrofit overnight. A major automotive plant that wants to integrate new robotic systems needs months of testing, integration, and worker training. The companies that provide the infrastructure—Qualcomm, Arm—are winning adoption now because by the time factories decide to upgrade, these platforms will already be proven through early deployments.
This is a classic infrastructure play: install the foundation, then let the applications layer build on it. Contrast this with consumer robots, where Figure or Boston Dynamics has to prove the entire value proposition themselves. The enterprise robotics space lets infrastructure providers derisk their bets by letting others prove the applications. However, enterprise adoption also means that infrastructure providers cede control of the final product to their partners. Qualcomm doesn’t control how Kuka designs its robots or sets its pricing. This limits the upside; infrastructure plays typically generate lower margins than product companies. Nvidia solved this by also selling software (CUDA libraries, TensorRT, etc.) and consulting services. Qualcomm and Arm will need to do the same to capture full value. Companies that attempt to stay pure chip vendors in robotics will lose to those that offer end-to-end solutions, even if the integrated offering requires deeper industry knowledge and longer sales cycles.
The Hidden Fragmentation Risk That Could Splinter the Market
The robotics industry spent decades fragmented across proprietary platforms—different robot operating systems, incompatible communication protocols, fragmented supply chains. Consolidation around shared infrastructure (Qualcomm IQ10, Arm’s Physical AI platform) could finally solve this. But fragmentation could also resurface if multiple competing platforms gain traction. If Qualcomm’s IQ10 captures the autonomous systems space while Arm dominates humanoids and custom applications, developers will need expertise in both. This mirrors the GPU market: Nvidia dominates for AI training, but specialized chips from others serve specific use cases. The real risk is that robotics companies opt to integrate vertical solutions rather than adopt modular infrastructure. Tesla, for instance, is building its own chips for humanoids rather than licensing from Qualcomm or Arm.
Figure AI, despite its partnership with Qualcomm, is also developing in-house solutions. If the largest, best-funded robotics companies go vertical, they’ll fragment the market despite infrastructure vendors’ best efforts. This is a warning for investors betting on a single infrastructure winner: robotics might remain decentralized, with multiple platforms coexisting without clear dominance. Another limitation is that neither Qualcomm nor Arm owns the software layer entirely. They can provide AI frameworks and optimization tools, but they can’t mandate how robotics companies write applications or structure their systems. Nvidia partially solved this through CUDA’s ubiquity, but that took decades. Qualcomm and Arm face compressing timelines—they need market dominance within 3-5 years to justify current investments. If they don’t achieve it, capital may move elsewhere, and robotics infrastructure will splinter further.
How Chips for Robotics Differ From AI Data Center Infrastructure
Robotics chips face constraints that data center chips don’t. A robot arm operating in a factory must respond to stimuli in milliseconds; latency isn’t merely inconvenient, it’s dangerous. A data center can tolerate some latency; a robot can’t. This drives different architectural choices. Qualcomm’s IQ10 and Arm’s Physical AI designs include real-time processing capabilities and edge AI inference—computation that happens on the robot itself, not in the cloud. This is fundamentally different from Nvidia’s data center approach, where inference often happens server-side. Power consumption is another differentiator. A data center can run chips at full power because electricity is cheap and abundant. A mobile robot runs on batteries; every watt matters.
This explains why Arm’s architecture, historically optimized for mobile efficiency, translates well to robotics. Qualcomm, coming from a smartphone and IoT background, also understands this constraint. Nvidia’s data center GPUs, by contrast, are power-hungry monsters optimized for throughput, not efficiency. This is a genuine advantage for Qualcomm and Arm in the robotics space—they’ve already solved the power problem for smartphones. Porting that expertise to robots is a natural extension. The example here is Advantech’s AIR-055 sensor module. It’s designed to run edge AI inference (computer vision, sensor fusion) on the robot itself, not relay data to a cloud server. This requires a different chip design philosophy than Nvidia’s CUDA ecosystem, which was built for cloud computing. Qualcomm and Arm are winning here not because they’re better at raw compute, but because they’ve pre-solved the engineering problems robotics companies face. That’s a material advantage in infrastructure.
What Success Looks Like in the Next 24 Months
The next phase of competition will be defined by real deployments, not announcements. By late 2027, we should see Qualcomm IQ10-powered robots in actual factory environments, handling tasks like assembly, sorting, or material handling. Similarly, Arm’s ecosystem should show production units of humanoids or specialized industrial robots shipping with Arm-designed chips. These deployments will be quiet—not headline-grabbing consumer robots, but workhorse machines in factories, warehouses, and logistics hubs. That’s where the next Nvidia emerges: not through flashy demos, but through boring, profitable adoption in unglamorous places.
The question for 2027 is whether Qualcomm and Arm can move from partnership announcements to ecosystem density. Nvidia achieved this by making CUDA so valuable that alternatives became uncompetitive. Qualcomm and Arm need similar network effects in robotics. If dozens of manufacturers adopt their platforms, software developers will write more applications, which will attract more manufacturers. Conversely, if uptake stalls and companies continue building proprietary solutions, infrastructure consolidation fails, and robotics remains fragmented for another decade. The outcome will be determined by how quickly the application layer (robots themselves) reaches maturity, which is beyond these chip companies’ control.
Conclusion
Qualcomm and Arm Holdings are positioning themselves to become robotics’ infrastructure providers, much like Nvidia became AI’s essential layer. Qualcomm’s IQ10 platform and Arm’s Physical AI business unit represent serious bets on the sector, backed by partnerships with major manufacturers like Kuka and Boston Dynamics. The market is genuinely large—projected to grow from $1.50 billion to $15.24 billion by 2032—and early enterprise adoption in factories and warehouses provides a beachhead for infrastructure consolidation. However, the outcome remains uncertain.
Fragmentation risks loom as larger robotics companies like Tesla integrate vertically. Arm and Qualcomm control only one layer of the stack; they can’t ensure that the robots themselves work reliably enough to justify massive infrastructure investment. The companies betting on infrastructure today should understand that they’re placing a leveraged bet on robotics reaching maturity—a bet that may not pay off for a decade. The quiet adoption happening now in factories is promising, but premature celebration would be exactly as premature as calling Nvidia’s GPU empire inevitable in 2008. Infrastructure plays only win if the applications actually ship.
