TCS survey: manufacturers are scaling physical AI fast, but 40% have no plan for when it fails

Physical AI is the category of artificial intelligence that drives robots, conveyor systems, autonomous forklifts and the rest of the moving machinery on a factory floor. It is the embodied side of the field: software that turns sensing into action in the real world rather than on a screen. According to a new survey from Tata Consultancy Services (TCS), the manufacturers who run those factories are preparing to spend on it fast, while the plans for what to do when it fails are still thin.

The TCS Future-ready manufacturing: physical AI readiness report 2026, carried by Computer Weekly, polled 300 senior manufacturing executives in North America and Europe, drawn from automotive, electronics and high-tech, industrial equipment, process industries and aerospace and defence. Three-quarters of those executives expect physical AI to have a significant or transformational impact on assembly and manufacturing, and 77% expect the same for warehouse operations. Seventy-two percent expect significant impact across logistics and material movement. The spending direction is uniform: 26% of the surveyed firms plan to increase their physical AI budget this year, and none plan to reduce it.

That is a faster adoption curve than the governance layer underneath it. The same survey finds that about 40% of manufacturers report they have unclear or no formal accountability structure for physical AI failures, and another 40% say they are unprepared for the emerging regulatory requirements now landing on industrial robotics in the European Union and on workplace AI in the United States. The same executives who expect the technology to transform their operations are still working out who inside the company is on the hook when an autonomous forklift puts a worker in the hospital.

That gap is the actual story. The "physical AI is now mainstream" framing is the vendor's own marketing thesis, and TCS, as the survey author, has an interest in selling that frame. The data inside the survey is real, and the adoption numbers are unambiguous, but the headline is doing the work of an analyst call that has not been made by an independent body. Reading the survey carefully turns the "mainstream" claim into something more useful: physical AI investment is rising while the operating rules around it are still being written.

The choice facing those manufacturers is not whether to deploy the technology. The investment signal already settles that. It is whether the deployment outruns the safety, liability and disclosure arrangements that turn a robot into a worker-safe tool. In practice that means three things the survey does not name but the governance gap implies: a clear ownership line for when a physical AI system causes harm, a documented failure-investigation process that survives a regulatory inspection, and a board-level reporting channel for those incidents before they become enforcement actions.

TCS frames the people-side of the deployment as a "human plus AI operating model" and roughly two-fifths of the surveyed executives expect significant workforce benefits, especially in hazardous, repetitive or complex tasks. That is a legitimate, important point and a real direction of travel in the industry, but it measures expected benefits, not measured worker outcomes, and the survey does not pair it with a parallel measurement of worker risk. The "people-first" framing is best read as the vendor's preferred outcome rather than an empirical finding.

What to watch next: the EU AI Act's high-risk provisions for industrial robotics begin to apply in earnest this year, and OSHA's enforcement posture on algorithmic management in the United States is hardening. The TCS survey establishes that about 40% of large manufacturers are entering that period without the governance scaffolding the regulators will expect. The next vendor report in this category will read very differently depending on whether that number moves.