We Analyzed 14M USPTO Filings: Here's What's Trending in 2026

AI-related trademark filings nearly doubled in a single year. Overall filing volume grew 5%. The fastest shift is hiding in the goods and services descriptions, where "artificial intelligence" is now the most common technology phrase in new applications.

Signa analyzed 14.1 million USPTO trademark records to separate actual trademark filing trends from the narratives that have stuck around too long.

• AI filings surged 86% year over year. Applications with "artificial intelligence" in their goods and services descriptions jumped from 11,449 in 2024 to 21,286 in 2025. No other technology category is growing this fast.
• China-origin applications recovered after a sharp correction, stabilizing around 148,000 per year after dropping 36% from the 2021 peak. The "China filing crisis" narrative is outdated.
• Use-in-commerce (Section 1a) leads intent-to-use (Section 1b) at 49% vs 44% of domestic filings. The conventional wisdom that intent-to-use dominates is wrong, at least for 2025.
• Filing volume is climbing again. CY2025 hit 625,254 applications, up 5% from 2024 and approaching pre-pandemic growth projections.

What are the biggest trademark filing trends in 2026? AI-related trademark filings rose 86% year over year, total USPTO application volume grew 5%, China-origin filings stabilized after the 2022 correction, use-in-commerce filings led intent-to-use filings, and Class 41 (education and entertainment) became the second-largest filing category by volume.

Every number comes from Signa's production database of 14.1 million USPTO records, cross-referenced with official USPTO reports. The queries are reproducible.

USPTO Filing Statistics: Volume Is Climbing Again

The USPTO received 625,254 trademark applications in calendar year 2025. That's roughly 1,700 new filings every business day.

Filing volumes jumped sharply in 2020 and 2021, peaking at 673,539 applications in CY2021. Pandemic-era e-commerce growth and a wave of speculative overseas filings drove that peak to roughly 35% above the 2019 baseline of 497,959.

Then came the correction. CY2022 and CY2023 saw volumes contract to 551,094 and 547,658 respectively. Some observers read this as structural decline. It wasn't. The speculative overseas filings washed out. Domestic filing volume held steady through the correction.

CY2024 marked the inflection point: 596,631 applications, up 9% from the trough. CY2025 continued the recovery at 625,254, putting the system back on the pre-pandemic growth curve. Draw a trendline from 2015 through 2019 and extend it forward. CY2025 lands close to where you'd expect.

Partial CY2026 data (272,631 applications through late May) suggests the growth is continuing, though it's too early to annualize reliably.

The trademark registration process is now essentially fully digital, with TEAS Plus e-filing exceeding 99%. This matters for data completeness: nearly every application is captured electronically at filing, giving analysts a near-complete picture of activity in real time.

Class 9 Still Leads. But Class 41 Is Closing the Gap.

Nice classes are the international system for categorizing the goods and services a trademark covers. There are 45 classes total (34 for goods, 11 for services), and every trademark filing maps to at least one. For a full breakdown, see the Nice classification guide.

Class 9 (software, electronics, mobile apps) remains the single largest Nice class at the USPTO, with 481,634 active registrations. Class 41 (education and entertainment) follows at 405,099. But the gap is narrowing.

The top five classes by filing volume in CY2025:

1. Class 9 (software, electronics, apps): 80,843 filings
2. Class 41 (education, entertainment, sporting activities): 77,209
3. Class 35 (advertising, business management, retail services): 69,866
4. Class 25 (clothing, footwear, headwear): 64,432
5. Class 42 (technology services, SaaS, scientific research): 59,086

The surprise is Class 41. The second-largest filing category is no longer clothing or business services, but education, entertainment, and digital media: a category shaped by creators, gaming, streaming, online courses, and content platforms.

Class 9 has the densest register at 481,634 active registrations. Class 41 (405,099) and Class 35 (403,616) are close behind. Together, the top five classes hold over 1.9 million registered marks. That density means more potential conflicts for every new application.

Class 25 (clothing) remains a high-volume filing category, driven by direct-to-consumer brands and streetwear labels. It's also one of the most contested classes for opposition filings.

Multi-class filings are increasing. A SaaS company typically needs protection across Class 9 (software products), Class 42 (platform services), and possibly Class 35 (business services). These multi-class applications inflate per-class counts but reflect genuine commercial breadth.

Class distribution explains what industries are filing. Geographic origin explains who.

China-Origin Filings: The Recovery Nobody Talks About

The biggest geographic story in recent USPTO trademark trends is not the decline in China-origin filings. It's the recovery.

China-origin applications peaked at 180,676 in CY2021 and then crashed to 114,967 in CY2022, a 36% drop. That correction was real and driven by deliberate USPTO enforcement changes:

• Stricter use-in-commerce requirements. The USPTO increased scrutiny of specimens (the evidence that a mark is actually being used in US commerce), catching fabricated evidence.
• US counsel requirement. Since August 2019, foreign-domiciled applicants must be represented by a US-licensed attorney. This added cost and friction to speculative filings.
• Specimen audit program. Random audits of registered marks required additional proof of use, cancelling registrations that couldn't demonstrate genuine commerce.

These changes cleaned out speculative filings, particularly China trademark squatting where filers registered marks preemptively to block or extort legitimate brand owners.

But the story doesn't end there. China-origin filings recovered to 154,540 in CY2024 and stabilized at 148,303 in CY2025. That's still 18% below the 2021 peak. But it's 29% above the 2022 trough. The speculative filings are gone. Legitimate Chinese brand owners are filing at higher rates than before the crackdown.

Domestic filings remain the majority: 394,389 US-origin applications in CY2025. The United Kingdom (8,482), Canada (7,716), and South Korea (7,350) round out the top foreign filers.

For companies building a global trademark filing strategy, the key insight is that CNIPA (China's domestic trademark office) still processes more applications annually than every other trademark office combined. The USPTO enforcement changes cleaned up the US register. They didn't change the global picture.

Filing Basis: Use-in-Commerce Leads, Not Intent-to-Use

In the United States, trademark applicants must specify a filing basis that describes their relationship to the mark:

• Section 1(a), use in commerce: The applicant is already using the mark in interstate commerce. Proof of use is submitted at filing.
• Section 1(b), intent to use: The applicant has a bona fide intention to use the mark but isn't using it yet. A Statement of Use must be filed before registration.
• Section 66(a), Madrid Protocol: The applicant filed through the Madrid system, an international framework that lets you file one application covering multiple countries, and designated the US.

The real CY2025 breakdown:

Use-in-commerce filings led intent-to-use by five percentage points in CY2025. That matters because it reverses the usual assumption that US filing activity is dominated by pre-launch intent-to-use applications. More applicants are filing after they've launched products, not before.

A higher share of use-in-commerce filings means more marks on the register are attached to actual products and services. The register more closely reflects the real marketplace than it would if speculative intent-to-use filings dominated.

Abandonment rates reinforce this. For mature filing cohorts (CY2020-CY2022), approximately 36% of applications were eventually abandoned. That includes both voluntary abandonments (applicant chose not to proceed) and involuntary ones (failure to respond to office actions, failure to file Statement of Use). The rate for CY2024-CY2025 cohorts appears lower (11-27%), but many of those applications are still in process.

AI Trademark Filings Nearly Doubled. The Land Rush Is Real.

Applications with "artificial intelligence" in their goods and services descriptions grew 86% year over year, from 11,449 in CY2024 to 21,286 in CY2025.

A note on methodology: "AI-related" here means filings where the goods or services description contains the phrase "artificial intelligence." This is a conservative text-matching approach. It captures filings that explicitly describe AI functionality but misses applications using only "machine learning," "neural network," or "generative AI." The true count of AI-adjacent filings is higher. The 86% figure is a lower bound on the growth rate.

The growth curve is accelerating, not plateauing. From 2020 to 2022, AI filings grew slowly (4,385 to 5,527, a 26% increase over two years). The inflection came in 2023, when generative AI went mainstream: 8,394 filings, a 52% jump. CY2024 added another 36%. CY2025 nearly doubled.

This pattern has precedent. Similar surges followed "cloud" in 2010-2013 and "blockchain" in 2017-2019. When a technology category reaches mainstream adoption, trademark filings follow 12 to 18 months later. The generative AI wave started in late 2022 when ChatGPT reached mass adoption. The trademark filing surge is arriving on schedule, but at a larger scale than previous technology waves.

The practical implication: clearance searches in AI-related classes (primarily Class 9, Class 42, and Class 35) are getting harder fast. Terms like "GPT," "neural," "gen," and "cognitive" appear in thousands of pending and registered marks. For anyone launching an AI product, the window for securing a clean, distinctive mark is narrowing with each quarter.

What These Trademark Filing Trends Mean for Professionals

The post-pandemic correction is complete. The US trademark system is growing again, but the composition of that growth has shifted.

For practitioners running clearance searches: the register is dense and getting denser. Class 9 holds 481,634 active registrations. Class 41 and Class 35 are each above 400,000. AI-related filings are surging into these already-crowded categories. Comprehensive searches need to account for both registered marks and pending applications.

For companies filing trademarks: filing basis typically depends on actual commercial status. The data shows that use-in-commerce filings (1a) slightly outnumber intent-to-use (1b). If you're already selling, file with specimens. If you're pre-launch in a fast-moving category like AI, the intent-to-use basis lets you secure a priority date before launch.

For the industry broadly: the China-origin filing correction is over. Applications from Chinese brand owners have recovered to sustainable levels. The enforcement changes worked as intended: speculative filings declined, legitimate commercial filings continued. Trademark monitoring of new publications remains important across all major filing origins.

What to watch in 2026: whether AI filing growth sustains or plateaus; whether Class 41 overtakes Class 9 in new filing volume; whether China-origin numbers hold steady or resume climbing; and whether the abandonment rate for the 2023-2024 cohorts settles at the historical ~36% or drops as filing quality improves.

This analysis describes trends in aggregate data. Consult a trademark attorney for guidance specific to your situation.

About the Data

Signa's production database contains 14,100,359 USPTO trademark records as of May 2026. All queries in this analysis ran against the production read replica using SQL against normalized, deduplicated data. Filing basis data comes from the trademark_filing_basis table; class distribution from trademark_classification; owner geography from trademark_owners joined with owners.

The AI filing analysis uses exact text matching for "artificial intelligence" in the goods_services_text field. A broader search including "machine learning," "neural network," and related terms would yield higher counts but was excluded due to query performance constraints on the full 14M-record dataset. The growth trend would be at least as dramatic.

To run your own analysis against this data, the Signa API supports filtering by Nice class, filing date range, jurisdiction, owner country, and status. Example:

curl https://api.signa.so/v1/trademarks \
  -H "Authorization: Bearer sig_live_..." \
  -H "Content-Type: application/json" \
  -d '{"filters": {"jurisdictions": ["US"], "nice_classes": [9], "filing_date": {"gte": "2025-01-01"}}, "options": {"aggregations": ["filing_year", "owner_country"]}}'

Signa's trademark API provides programmatic access to this full dataset, including filing basis, Nice class distribution, and owner geography filters. If this analysis raised questions about your specific market, explore the API at signa.so.