Amazon’s robotics patent portfolio has grown 28x since its Kiva acquisition

The nation’s top two private employers, Walmart and Amazon, are racing to replace warehouse workers with robots, and the patent trail underscores the scope of commitment. While Amazon built its robotics arsenal over 20 years (3,097 international patents and counting), Walmart’s 2017 surge saw newly granted robotics patents jump 185% in a single year. With 3 million combined employees and Amazon already deploying 1 million robots while CEO Andy Jassy signaling that AI will lead to a leaner workforce, the company is increasingly becoming an automation-first company. Amazon already commands a fleet of 1 million robots at its facilities.

Amazon’s sizable fleet of robots operates as a quasi-central nervous system, as the company explained in a recent post. And now, that system is getting a significant intelligence upgrade. The retail giant recently unveiled DeepFleet, a generative AI model that acts as the brain behind the brawn. Think of it like Voltron, the classic animated series where five separate robotic lions combine to form one giant, more powerful robot. But here, instead of five robotic lions, it’s a million-strong robotic workforce. Scott Dresser, VP of Amazon Robotics, quantified the immediate impact, stating the AI will “coordinate the movement of robots across our fulfillment network, improving the travel time of our robotic fleet by 10% and enabling us to deliver packages to customers faster and at lower costs.” The vision extends far beyond a one-time efficiency gain. DeepFleet continuously learns and evolves, compounding Amazon’s advantage over time. As the company explains, “As DeepFleet learns from more data, it will continue to get smarter—driving deeper efficiencies… and reimagining what’s possible in robotic logistics.”

We will need fewer people doing some of the jobs that are being done today, and more people doing other types of jobs. —Andy Jassy

An analysis of international patent data from Google’s BigQuery public database reveals the divergent paths these retail giants took toward automation. Amazon’s journey began in 2003 with basic warehouse control systems, accelerated dramatically after acquiring Kiva Systems in 2012, and has since pivoted toward artificial intelligence. Machine learning patents alone increased 23-fold between 2012 and 2020, transforming what started as mechanical automation into an AI-powered robotic workforce.

Amazon vs. Walmart robotics patent race (2010-2024)

Year	Amazon granted robotics patents	Walmart granted robotics patents	Key events
2010	15	0	Pre-acquisition era
2012	132	0	Amazon acquires Kiva
2015	371	6	Walmart begins robotics push
2017	537	511	Walmart’s panic surge
2020	652	197	Amazon acquires Zoox
2024*	45	74	1 million robots deployed

*May represent incomplete data given patent reporting lags

But the raw patent counts only tell half the story. A closer look inside Amazon’s portfolio reveals a strategic pivot that explains its current aggressive moves in robotics, and Jassy’s confidence that agentic AI will reshape Amazon’s workforce. While early efforts focused on the mechanics of warehouse automation, patents for transport systems (B65G) and navigation controls (G05D), the years following the Kiva acquisition saw a deliberate and significant shift. The new R&D focus was no longer just on making robots move, but on making them “see,” “learn” and “think” as noted in the table that follows.

Amazon’s patent filing reveal a fierce focus on intelligence: AI/ML patents (G06N) exploded 23-fold between 2012 and 2020, while computer vision patents (G06V) tripled. The patents coincide with Amazon’s building the foundation for systems like DeepFleet, which now coordinates their million-robot fleet to improve travel time by 10%. As Jassy told employees in June 2025, “We will need fewer people doing some of the jobs that are being done today,” explicitly crediting AI and robotics for driving these efficiency gains. The company has over 1,000 AI projects in development, transforming everything from inventory placement to robot coordination.

Amazon’s technology pivot moves from warehouse movers to AI ‘brains’ (2012 vs. 2020)

This R&D playbook is also evident in Amazon’s acquisitions. The 2012 purchase of Kiva Systems for $775 million was a strategic absorption of technology. Kiva’s own patent filings ceased as its technology was fully integrated into Amazon’s operations. This integration was solidified in 2015 when Kiva was recast as Amazon Robotics. The next flurry acquisitions have followed a similar pattern. In 2019, Amazon acquired Canvas Technology, which brought in expertise in spatial AI for robot navigation. More recently, in 2024, a deal with Covariant added further AI capabilities through a talent and technology license acquisition. Amazon’s sustained interest in acquiring complementary technologies was also the focus of an attempted, but failed, $1.7 billion acquisition of iRobot in 2024.

Conversely, the 2020 Zoox acquisition represents a different strategy. Zoox continues to operate independently and file its own patents in autonomous vehicles. This dual approach allows Amazon to both flex its core warehouse automation muscle while placing strategic bets on last-mile delivery and future automation frontiers. Other companies are making similar moves. Tesla, for one, is targeting 50,000 humanoid robots by 2026, Figure AI is now valued at $2.6 billion and competitors like Agility Robotics have raised hundreds of millions.

Methodology note on the 28x number
Here is what went into the equation to provide that figure. To provide the query, I usd the number of patents filed in the post-Kiva era and compared them to the number filed in the pre-Kiva era:

• Patents Before Kiva (filed < 2012): 179
• Patents After Kiva (filed >= 2012): 4,946
• Calculation: 4,946 (Post-Kiva) / 179 (Pre-Kiva) = 27.63

CPC code	Category	Description
B25J	Manipulators	Robot arms, end effectors, control systems
B65G	Conveying/Handling	Conveyors, lifts, material handling
G05D	Non-electric Control Systems	Controlling physical parameters like speed
G06N	AI (Biological Models)	Neural nets, fuzzy logic, machine learning
G06V	Computer Vision	Visual recognition and scene understanding

 

The data was fetched in BigQuery from international public patent data with the following query, which also shows the CPC codes used:

SELECT 
  CASE 
    WHEN EXTRACT(YEAR FROM PARSE_DATE('%Y%m%d', CAST(p.filing_date AS STRING))) < 2012 THEN 'Pre-Kiva'
    ELSE 'Post-Kiva'
  END as era,
  COUNT(DISTINCT p.publication_number) as robotics_patents
FROM `patents-public-data.patents.publications` p
CROSS JOIN UNNEST(p.assignee_harmonized) AS a
WHERE a.name = 'AMAZON TECH INC'
  AND EXISTS (
    SELECT 1 FROM UNNEST(p.cpc) AS cpc
    WHERE cpc.code LIKE 'B25J%' OR cpc.code LIKE 'B65G%' 
       OR cpc.code LIKE 'G05D%' OR cpc.code LIKE 'G06N%' 
       OR cpc.code LIKE 'G06V%'
  )
GROUP BY era;