The $31 million bet that cameras should act, not record
In October 2024 Spot AI launched the first Video AI Agents and closed a $31 million round with Qualcomm Ventures. What the chipmaker was actually buying.
The press release ran on October 29, 2024, and it did two things at once. It announced a product, Video AI Agents, and it announced money: $31 million in new equity, with Qualcomm Ventures joining Scale Venture Partners, StepStone, Redpoint, and Bessemer, taking Spot AI's total to $93 million. Forbes, Fortune's Term Sheet, Axios Pro Rata, and SiliconANGLE all carried it.
Funding rounds are usually the boring half of such a release. This one is worth reading for who wrote the cheque.
What was launched
Until then, the company said, video surveillance "was limited to passive recording, basic search, and manual collaboration." The agents were pitched as the opposite: cameras turned "from passive observers into proactive problem solvers that can autonomously identify issues and trigger responses without the need for human intervention." The examples were concrete. Alert a stakeholder. Turn on lights and sounds. Play a voice message. Stop a machine if someone is hurt. Improve load and unload times.
A week earlier, customers had received a preview letter from the CEO describing three agents by job: a safety agent for slips, falls, forklift near-misses and missing PPE; an operations agent for unattended workstations and loading bottlenecks; a security agent for loitering, tailgating, fence jumping, and catalytic-converter theft.
Eighty percent of our understanding of the world happens through eyes. That's why we're dedicated to helping our customers turn their video cameras into additional teammates.
Rish Gupta, the CEO, in the release. At the time the company served 1,000 customers across 17 industries and said it indexed more than twice the video uploaded daily to YouTube.
Why a chipmaker
Qualcomm does not invest in software companies for the software. It invests in workloads that will run on its silicon. Tushar Gupta, senior director at Qualcomm Ventures, said the quiet part plainly: the deal "closely aligns with Qualcomm's vision of implementing AI on edge to solve real-world challenges."
That sentence explains the product architecture better than any diagram. The release noted that the agents ran on an edge processing approach with "three times more computing power than traditional AI cameras," specifically so that newer foundation models could run on site rather than in a distant data centre. If cameras are going to act in under a second, the model has to live near the camera. That is a chip business.
The numbers that were published
The release listed outcomes by industry, as claimed by the company: directors of safety in manufacturing reducing injuries by 40 percent; auto-service owners quadrupling membership rates and lifting revenue 10 percent; school principals cutting time spent on incidents by 80 percent. Vendor numbers, to be read as such, but they mark where the company believed the value sat: safety first, then operations, then security.
What happened next
The agents became the product. Within three months came a pole-mounted security agent for the outdoors; within six, Iris, a way to build agents by describing them; within eighteen, a company-wide shift to selling only through partners. The October 2024 round was the point at which "AI camera company" stopped being an accurate description. The cameras were becoming the least important part.