SK Hynix sells the most of a certain memory chip. It doesn't hold the most patents behind it. Toyota leads solid-state batteries. A different company leads batteries overall. And the AI category everyone talks about is the smallest one we found.
We start with the top five technology fields by US patent filings. Then we check the same ranking globally, to see where the US differs. Then we go field by field, into the specific areas inside each one and the real companies behind them.
The top 5 fields by filing volume
Computer technology leads by a wide margin, more than double the next field. That's not surprising. What's surprising is what happens when you compare the US to the rest of the world.
Top 5 US patent families by technology field, Jan 2024 to present
Where the US ranking diverges from the global one
Run the same classification globally, combining the US, Europe, China, and Japan, and four of these top 5 fields barely move. One does, and it tells a real story.
Field rank: United States vs. global (US + EP + CN + JP), top 5 US fields
Semiconductors ranks 5th in the US but 8th globally. Every other field in the US top 5 holds close to the same position worldwide.
Inside the fields: what the sub-topics and their filers actually look like
A field's overall growth rate and the growth rate of what's actually happening inside it are two different numbers, and they don't move together. Batteries as a whole grew a few percent last year. Solid-state batteries, one narrow slice of that field, more than doubled. AI as a whole grew modestly. LLM-driven agent patents, one slice of AI, went from single digits to thousands in the same stretch. The field-level number tells you the industry is stable. The sub-topic underneath it tells you something is actually moving fast, and most reporting never gets past the first number to find it.
Artificial intelligence
Broad AI and machine learning grew a modest 4.6% between 2024 and 2025. Patents describing autonomous, LLM-driven agents, systems that plan and carry out multi-step tasks on their own, tell a completely different story.
LLM-driven autonomous agent patent families by year
Shown on a log scale, since a normal chart would make the early years invisible.
There were 5 patents like this filed in 2021. By the partial 2026 year, there were already 2,838. It's a sub-category growing far faster than the AI field it sits inside, and it's exactly the kind of thing a category-level ranking will hide from you if you only look at the top line.
The companies actually filing these patents aren't who you'd guess. Search the US filings specifically, and Salesforce leads by a wide margin, followed by Microsoft and UiPath. Citigroup, JPMorgan Chase, and Toronto-Dominion Bank all show up too, banks don't file patents for fun, they file when they expect to license, defend, or use the IP as a differentiator on their own services. We checked the other side of this directly: OpenAI, Anthropic, and Mistral, combined, hold fewer than 10 patents that match this same search. Their foundational model work is real, but on the published patent record, it hasn't shown up here yet, likely due to the normal 18-month gap between filing and publication. If you're only watching the brand names everyone talks about, you're missing most of the field.
AI patent families by sub-focus area, 2022 to present
The LLM-agent patents from above actually fall inside "industrial AI applications" and "conversational intelligence" here. Both file more patents overall, even though agent-specific patents are growing faster.
Autonomous content generation, the thing everyone talks about when they talk about generative AI, is second-to-last out of six categories by patent volume. Conversational intelligence and industrial AI both file about five times as many patents. Legal tech AI is the smallest category by far, just 910 patents. But it's also growing the fastest, more than tripling since 2022.
AI filers: patent count vs. citation-weighted quality
Intel holds less than a quarter of IBM's AI patent count, but the highest quality index of any AI filer checked.
Semiconductors
Top US semiconductor filers, all subclasses combined
BOE Technology, a Chinese company, files more US semiconductor patents than Intel, IBM, Tokyo Electron, Micron, or SK Hynix. TCL, another Chinese company, beats SK Hynix too.
Sub-focus areas, 2022 to present
HBM / 3D-stacked memory: top filers
Batteries
Batteries overall grew a modest 2.7% between 2024 and 2025. Solid-state battery patents, the chemistry every major automaker has committed development capital to, grew far faster.
Solid-state battery patent families by year
Solid-state filings went from 2,223 in 2022 to 4,601 in 2025, more than doubling. That growth also comes with a leadership change. LG Corp leads batteries overall, with 25,817 patents, more than one in ten of all battery patents we found. Narrow it down to solid-state chemistry specifically, and Toyota takes the lead instead, holding 1,735 patents against LG's 962, nearly double. The company leading the broad category and the company leading its fastest-growing corner are two different companies entirely.
Sub-focus areas, 2022 to present
Thermal management / safety: top filers
Battery safety and thermal management patents outnumber fast-charging and silicon anode patents combined. That's surprising, since range and charging speed are what actually sell EVs to consumers. LG leads here too, same as the broader battery category. General Motors and Hyundai are the only two non-Asian companies to make this list at all.
Battery filers: patent count vs. citation-weighted quality
BYD holds a fraction of LG Corp's patents, but the highest quality score of any battery filer we checked.
Robotics
Broad robotics filings actually fell 4.9% between 2024 and 2025. Narrow the same corpus to humanoid robots specifically, and the trend reverses completely.
Humanoid-adjacent patent families by year
Humanoid-adjacent filings went from 1,160 in 2022 to 2,790 in 2025, a 78% jump in the most recent year alone. The parent field looks flat. The humanoid sub-category inside it is anything but, which is exactly why a category-level ranking can hide the one technology actually worth watching.
Sub-focus areas, 2022 to present
Humanoid robots: China vs. the US name
Surgical robotics runs almost exactly even with humanoid robots, both well ahead of grippers and warehouse automation. That pairing never shows up in general "robotics boom" coverage, which is almost entirely about humanoids. The single highest-cited patent in this whole robotics dataset belongs to Intuitive Surgical, a name that rarely comes up in general robotics coverage either. Inside humanoid robotics, the four leading Chinese filers combined hold roughly eleven times Figure AI's filing volume. Figure AI's entire 59-patent portfolio is barely eighteen months old, a real head start gap against competitors who were building volume since 2022.
Robotics filers: patent count vs. citation-weighted quality
No dramatic outlier here. Boston Dynamics holds about a fifth of Fanuc's patents, but a similar quality score, not the big gap seen in semiconductors, batteries, or AI.
Robotics doesn't follow the same pattern. Quality scores here only range from 40.3 to 56.6, much tighter than semiconductors (41.4 to 70.7), batteries (39.8 to 72.1), or AI (39.3 to 65.5). Boston Dynamics comes closest to standing out. It holds 97 patents against Fanuc's 475, about a fifth of the volume, but almost the same quality score, 56.3 versus 55.4. Basically tied, not a reversal. Not every field hides a surprise like ASM Holding or BYD. It's still worth checking every time. You just won't always find one.
Autonomous vehicles
Sub-focus areas, 2022 to present
V2X, the technology that lets vehicles talk to each other and to roads, has almost three times as many patents as lidar and sensor systems, the technology that gets all the press coverage of self-driving cars.
Volume and quality answer different questions, checked across four fields, not just one
We first found this pattern in semiconductors, while researching gate-all-around and CFET transistors. Batteries confirmed it, with BYD. AI confirmed it again, with Intel. We checked robotics too, and it didn't hold there, which is worth knowing on its own. Here's the original semiconductor case one more time, since it's still the clearest example.
Semiconductor filers: patent count vs. citation-weighted quality
ASM Holding B.V. holds a fraction of TSMC's or Intel's patent count, but the highest citation-weighted quality score in this dataset.
Rank these nine filers by patent count, and ASM Holding comes in last. Rank them by quality instead, and it comes in first. BYD does the same thing in batteries. Intel does it in AI. Every "top filer" list on this page answers one question: who files the most. It doesn't answer a different question: whose patents actually matter. Those aren't the same thing. Robotics is the exception. Its quality scores only range from 40.3 to 56.6, much tighter than the other three fields, and nothing there shows the same kind of gap. Still worth checking every field, since you can't know in advance which ones will surprise you. One more note: the quality score is calculated separately for each dataset. A score of 72.1 in batteries and 70.7 in semiconductors aren't directly comparable. They only mean something within their own industry.
This analysis is about who's filing and how much. It doesn't tell you what any of that filing activity looks like once it's actually in prosecution, pending office actions, examiner assignments, response deadlines. If you want that for any company named above, insights.ipauthor.com tracks it separately, in real time. If you're the one filing, not just tracking who is, IP Author is built for the drafting and prosecution side of that same work.
What this means in practice
- A field's global ranking doesn't automatically apply to one country. Semiconductors and measurement both shift enough between the US and global rankings to change how you'd staff or specialize for each market.
- A flat parent field can hide a sub-category growing much faster, and the company leading the parent field isn't always the company leading that faster sub-category. Batteries and solid-state chemistry show this side by side.
- Leading the market and leading the patents are two different things. SK Hynix is the clearest example. Check both before assuming one implies the other.
- Patent count and patent quality are two different measurements. Check both before treating a volume ranking as a stand-in for importance.