Copyrightreserved

1

Questel

GenerativeArtificial

Intelligence

2025PatentLandscape

Copyright?2025Questel–Allrightsreserved

Questel

TableofContents

AboutQuestel 3

ExecutiveSummary 4

I.Introduction 6

II.Methodology 8

1.Datasource&Searchstrategies 8

2.Taxonomy 9

III.Deeplearningglobalpatentlandscape 10

IV.FocusonMultimodalAI/DigitalHumans/IntelligentAgents 14

1.Introduction 14

2.MultimodalAI 15

3.IntelligentAgents 17

4.DigitalHumans 18

5.GenerativeAI 20

Appendices 22

2

Copyright?2025Questel–Allrightsreserved

Questel

AboutQuestel

Questelisatrueend-to-endintellectualpropertysolutionsprovidertomorethan

20,000clientsand1,5Musersacross30countries.Weofferacomprehensivesoftwaresuiteforsearching,analyzingandmanaginginventionsandIPassets.

QuestelalsoprovidesservicesthroughouttheIPlifecycle,includingpriorartsearches,patentdrafting,internationalfiling,translation,andrenewals.Thesesolutions,whencombinedwithourIPcostmanagementplatform,deliverclientsanaveragesavingsof30-60%acrosstheentireprosecutionbudget.

Questel’smissionistoallowinnovationtobedevelopedinanefficient,secure,andsustainableway.Behindthismission,QuestelconsidersthatCorporateSocialResponsibility(CSR)isabroad-basedmovementinbusinessthatencouragescompaniestotakeresponsibilityfortheimpactoftheiractivitiesoncustomers,employees,communities,andtheenvironment.TolearnmoreaboutQuestel,pleasevisit:

Morecontentisavailableonline,pleaseconsult:

/resources-hub/

3

Copyright?2025Questel–Allrightsreserved

Questel

ExecutiveSummary

Deeplearningcontinuestobeoneofthemostdynamicareasoftechnologicalinnovation,withpatentactivityshowingsustainedandacceleratinggrowth.BuildingonourpreviousstudyonDeepLearningandLargeLanguageModels,thisreportanalyzesthenextwaveofinnovationthroughpatentfilings,withaspecificfocusonMultimodalAI,IntelligentAgents,andDigitalHumans.Thesethreedomainsareemergingrapidlyandincreasinglyconvergingtowardmoreautonomous,interactive,andhuman-centricAIsystems.

Thepatentlandscaperevealsthatasmallnumberofglobaltechnologyleadersareshapingthistransitionbycombiningfoundationmodels,agenticcapabilities,andhuman-likeinterfacesintocoherentinnovationstrategies.

GOOGLEpositionsitselfasacoretechnologyleaderthroughitsGeminifamilyofmodels.Designedasnativelymultimodal,Geminiintegratestext,vision,audio,andvideowhileprogressivelyembeddingagenticreasoningcapabilities.GOOGLE’spatentstrategyreflectsastrongfocusoninternationalprotection,supportingitsambitiontodeploymultimodalandagent-basedAIonaglobalscaleacrossSearch,cloudservices,andproductivitytools.

BAIDUstandsoutasthemostverticallyintegratedplayer.WiththeERNIEmultimodalengine,GenFlowandAgentBuilderforintelligentagents,andarapidlyexpandingportfolioofdigitalhumantechnologies,BAIDUcoversallthreedomainsinaunifiedstack.Itspatentleadership,particularlyinvolume,highlightsastrategycenteredonlarge-scaledeployment.

NVIDIAdominatestheDigitalHumanslandscape.Whilenotpositioningitselfasageneral-purposeAIassistantprovider,NVIDIAsuppliestheessentialinfrastructure,platforms,andtoolchainsthatmakeembodiedAIpossible.ItsOmniverseandAvatarCloudEnginetechnologiessupporthighlyrealisticdigitalhumansandmultimodalinteraction,backedbyaverystronginternationalpatentportfolio.

MICROSOFTadoptsanenterprise-firstapproachthatconnectsmultimodalAIandintelligentagentsdirectlytobusinessworkflows.LeveragingbothitspartnershipwithOpenAIanditsownagentframeworks,MicrosoftintegratesCopilotacrossOffice,

4

Copyright?2025Questel–Allrightsreserved

Questel

cloud,andenterprisesoftware.Itspatentfilingsshowabalancedstrategy,combiningsolidportfoliosizewithahighproportionofinternationalfamilies,reflectingglobalambitionsinprofessionalandenterpriseAI.

IBMemergesasakeyleaderinIntelligentAgents,supportedbyitswatsonx.aiplatformandGranitefoundationmodelfamily.IBM’sstrategyfocusesonprofessionalandenterprise-gradeAIsystems.Whilelessvisibleinconsumer-facingapplications,IBM’sstrongpatentpositionconfirmsitsroleasacornerstoneplayerinagent-basedAIforcorporateenvironments.

5

Copyright?2025Questel–Allrightsreserved

Questel

I.Introduction

DeepLearningrepresentsacoretechnologicallayerwithinartificialintelligence,basedonmulti-layeredartificialneuralnetworksthatlearnhierarchicalrepresentationsfromlargedatasetsandexcelinperceptiontaskssuchasimagerecognition,speechrecognition,andnaturallanguageprocessing.Withinthistechnologicallayer,GenerativeAI(GenAI)emergedaround2016asasignificantbranchfocusedoncreatingnewcontenttext,images,video,audio,code,orsyntheticdatausingapproachessuchasVAEs,GANs,diffusionmodels,transformers,andparticularlylargelanguagemodels(LLMs).LLMsaretransformer-basedsystemstrainedonmassivecorporatounderstand,generate,andmanipulatenaturallanguage,enablingcapabilitieslikereasoning,summarization,translation,andconversationalinteraction.

Figure1-ThespectrumofAI:exposingthedifferentlayersofintelligentsystems

IncontrasttothesefoundationalAIlayers,severaldomainshavedevelopedasapplicationsbuiltontopofthem:

?MultimodalAIreferstosystemsthatcanprocess,understand,andgenerateinformationacrossmultiplemodalitiesordatatypestext,images,audio,video,sensordata—simultaneously.Byintegratingandreasoningacrossdifferentformsofinput,thesesystemsproduceaunifiedunderstandingandcangeneratecoherent

6

Copyright?2025Questel–Allrightsreserved

Questel

outputsthatspanmodalities(e.g.,acaptionedimage,anarratedvideo,orasynchronizedaudio-visualresponse).Thismultimodalcapabilityenablesricherperception,morenaturalhuman-machineinteraction,andmoreflexiblecontentgenerationthansingle-modalityAI.

?IntelligentAgentsareautonomousorsemi-autonomousAIsystemsthatperceivetheirenvironmentviasensors(whichcouldincludevision,audio,orothersensordata),makedecisionsbasedongivenobjectives,andtakeactionstoachievespecificgoals.Builtondeeplearning,multimodalunderstanding,andgenerativeorreasoningcapabilities,theseagentscanoperateindependently,learnfromexperience,interactwithhumansandotheragents,andadapttheirbehaviortochangingcircumstances.Thismakesthemsuitablefordynamictaskssuchastaskautomation,planning,human-agentcollaboration,responsiveassistance,oradaptivedecisionsupport.

?Finally,DigitalHumansareAI-poweredvirtualrepresentationsofhumanbeings,combiningcomputergraphics,animation,naturallanguageprocessing,speechsynthesis,andbehavioralAItocreaterealistichumanavatarscapableofautonomousorsemi-autonomousinteractionwithrealhumans.ThankstomultimodalAIandgenerativetechnologies,digitalhumanscanexhibitlifelikeappearance,voice,expressions,gestures,andbehavior,enablingapplicationssuchasvirtualassistants,virtualtrainers,digitalactors,orcompanionsystemsinhealthcare,education,entertainment,orcustomerservice.

7

Copyright?2025Questel–Allrightsreserved

Questel

II.Methodology

1.Datasource&Searchstrategies

ThedatasourceusedinthisstudyistheFamPatworldwidedatabasesearchtool,OrbitIntelligence.FamPatisaglobalcollectionofpatentapplicationsandgrantedpatents,organizedbysimplepatentfamilies,coveringmorethan100patentauthoritiesworldwide,includingsearchablefulltextfrom22patentoffices,providedbyQuestel.SinceeachFamPatrecordmaycontainmanyindividualpublicationevents,allwithdifferentdates,thereportusestheearliestknownofficeoffirstfilingdateforeachpatentfamily.Thisisconsideredtherepresentativepatentfamilymember,whichisusedtorefertothepatentfamily.TheOfficeofFirstFiling(OFF)orpriorityreferstothefirstapplicationforaparticularinvention,which,whenfiledatanypatentoffice,becomesthe“priorityapplication,”withthedateofthiseventdefiningtheprioritydate.Thecountryofthefirstfilingisdefinedastheprioritycountry.

Thetablesandchartsincludedinthereportusethisprioritydate,unlessotherwisenoted,becauseitprovidesthemostaccurateindicationofinventiveactivity.Thedefinitionofpatentsources,i.e.,thelocationfromwhichpatentfamiliesareemanating,isbasedontheOfficeofFirstFiling(OFF).Itshouldbenotedthatthisdefinitionisnot100%accurate;nevertheless,itprovidesausefulandfairmethodofidentifyingtheusualcountryoffirstfilingforentities,whichtypicallycoincideswiththeirhomepatentoffice.

Eachpatentfamilyislinkedtooneormoreentities,collectivelyreferredtoasthepatentowners.Incaseswherepatentswithinthesamefamilyareheldbymultipleowners,thefamilyisattributedtoallrelevantentities.Toenhancereadabilityandprovideaclearoverview,thevariousapplicantsmentionedinthisdocumenthavebeensystematicallycleanedupandconsolidatedundertheirrespectiveparentcompanies.Subsidiariesandaffiliatedentitiesaregroupedundertheirparentorganizationtoreflectaunifiedownershipstructure.Thisgroupingprocessisconductedusingacombinationofautomatedmethods(viaOrbitIntelligencedatabase)andmanualmethods.Thisgroupingwasmadetothebestofthepublicinformationavailableatthetimethisreportwascreated.

8

Copyright?2025Questel–Allrightsreserved

Questel

2.Taxonomy

ThispatentlandscapefocusesonDeepLearning(DL),withaparticularemphasisonthreerapidlyemergingapplicationdomains:MultimodalAI,DigitalHumans,andIntelligentAgents.Thesearchmethodologycombinesmultiplestrategies,leveragingdeep-learning-relatedkeywords,associatedtechniques,andrelevantpatentclassifications,suchasG06N3/02(computersystemsbasedonneuralnetworkmodels).Toensuretheresultsstrikeanoptimalbalancebetweenexhaustivenessandprecision,thesearchemployedBooleanlogicanditerativerefinement,enablingtheconstructionofarobustandreliabledatasetforanalysis.

Althoughpatentsdonotalwaysexplicitlydescribetheseapplications,examiningthosethatdoprovidesvaluableinsightsintohowthesetechnologiesarebeingimplementedandprotected.Thisapproachoffersauniqueandtechnicallygroundedperspectivethatcomplementscorporatecommunicationsormarketingnarratives,whichoftenemphasizeonlythemostvisibleortrendingusecases.

Toenrichtheanalysis,werevieweddomain-specificliteraturetoidentifyandclassifythemostrelevantapplicationsforMultimodalAI,DigitalHumans,andIntelligentAgents.Thisallowsustohighlighttheirpotentialacrossavarietyofsectorsandusecases,offeringactionableinsightsintotheserapidlyevolvingtechnologicaldomains.Adetailedlistoftheseapplications,alongwiththeirdescriptions,isprovidedintheappendixattheendofthisreport.

Figure2-AIapplications,focusonMultimodalAI,DigitalHumans,IntelligentAgents

9

Copyright?2025Questel–Allrightsreserved

Questel

III.Deeplearningglobalpatentlandscape

ExplosiveGrowth:TheDeepLearningBoomContinues!

ThedatarevealsthreemajorphasesinthedeeplearningandgenerativeAIrevolution:

2011–2016:TheDeepLearningRevolution:Breakthroughsinrepresentationlearning,convolutionalandrecurrentneuralnetworks,andGPU-acceleratedtrainingestablishdeeplearningasthedominantparadigminAI.

2016–2021:EmergenceofGenerativeAI:Theintroductionoftransformerarchitecturesenableslarge-scalegenerativemodelsandmarksthetransitionfromtask-specificAItowardfoundationmodels.

2021–Present:TheGenerativeAIExplosion:Theconvergenceofscalinglaws,architecturalinnovation,andmassivecomputationalresourcesdriveslargelanguagemodelsatunprecedentedscales(e.g.,ChatGPT),diffusion-basedimagegeneration(e.g.,StableDiffusion),andmultimodalsystemscombiningtext,vision,andaudio(e.g.,GPT-4V,Gemini).

Thechartbelowshowsthenumberofpatentfamiliespublishedperyear.Aspatentapplicationsaretypicallypublished18monthsafterfiling,publicationdatesshouldbeshiftedbackby18monthstomoreaccuratelyreflecttheactualinventionfilingdates.

Figure3-DeepLearning,andLLMfilingdynamic,2011-2024(numberofpatentfamilyVsprioritydate)

10

Copyright?2025Questel–Allrightsreserved

Questel

Ouranalysisofthe2011–2025periodprovidesanear-completeviewoftheremarkableevolutionofdeeplearning(DL)-relatedpatentactivity.Deeplearning,asdefinedinitsmodernformaround2011,hasgeneratedapproximately450,000patentfamiliesoverthistimeframe.Asnotedinourpreviousstudy

1

,theaccelerationofpatentfilingsinthedeeplearningdomaincontinueswithnoclearsignofstabilization,withannualgrowthratesapproaching50%intheearlyyears.Thissustainedtrajectory,fromnearlyzerofilingsin2011towellover100,000publishedpatentfamiliesin2024,underscoresthatinnovationinthefieldremainshighlydynamic,withlarge-scaleresearchandindustrialprogramsstillactivelyexpanding.

Oneofthemoststrikingdevelopmentsistherapidriseoflargelanguagemodels(LLMs).Appearingonlyafter2020,LLMsstartedfromalmostzeroandexperiencedrapidgrowth,particularlyfrom2022onwards,illustratingtheexplosionofresearch,engineering,andcommercializationthatfollowedthearrivaloffrontier-scalemodels.Theirexpansioniscloselylinkedtofundamentaladvancesindeeplearning(DL):largerarchitecture,improvedtrainingpipelines,andadvancesinself-supervisedlearninghavecreatedidealconditionsfortheriseofLLMs.

Figure4-DeepLearningTopIndustrialApplicants,2011-2024

1

/resourcehub/consolidation-and-opportunities-in-llm-domain/

11

Copyright?2025Questel–Allrightsreserved

Questel

Figure5-LLMsTopIndustrialApplicants,2011-2024

Figures4and5illustratethemainassigneesshapingthepatentlandscapeinthefieldsofDeepLearningandLargeLanguageModels,rankedbythetotalnumberofpatentfamilies.Theyprovideinsightsintothescaleofinnovation,representedindarkcolors,andtheextentofinternationalpatentprotectionamongthemainplayers,representedinlightcolors.

Deeplearningpatents:leadershipisconsolidating

RegardingtheDeepLearningtechnologyfield,themajorplayerscontinuetoreinforcetheirpositionswithimpressivegrowthintheirpatentportfolios.

Baiduranksfirstwithaconsiderablelead,holding7,670patentfamilies.Thisdominantpositionisfurthersupportedbyasignificantnumberofinternationalpatentfamilies(1,347).BaiducontinuestoassertitsleadershipinDeepLearning,rankingnumberoneinpatentapplicationsforthethirdconsecutivetimesinceourfirststudy

2

,therebyconsolidatingitspositionasthemostactiveglobalinnovatorinthisfield.Thissustaineddominancereflectsalong-termandlarge-scaleIPstrategyanchoredincoredeeplearningtechnologies.

2

/wp-content/uploads/2021/11/2021-Deep-Learning-Patent-Landscape-

short-report-.pdf

12

Copyright?2025Questel–Allrightsreserved

Questel

Insecondplace,SamsungElectronicsholds2,665patentfamilies,withahighnumberofinternationalfilings(1,578),whichmadeittheleaderintermsofinternationalprotection.TencentTechnologyranksthirdwith2,283patentfamilies,backedby482internationalprotections.CloselyfollowedbyPingAnTechnologywith2,234patentfamilies,791ofwhichareinternationalfamilies.Infifthposition,IBMmaintainsarobustpresencewith2,129patentfamilies.With435internationalfilings.TheStateGridCorporationofChinashows2,102patentfamiliesbutonly2internationalpatentfamilies.Alibaba,inseventhplace,holds1,868patentfamilies,251ofthemareinternationallyprotected.Finally,Googlerankseighthwith

1,767patentfamilies,supportedbyanotablyhighlevelofinternationalprotection(1,097).

BaidudominatesLLMpatents,withTencent,MicrosoftandGoogleformingastrongsecondtier

BaiduleadstheLLMspatentlandscapewith1,471patentfamilies,clearlyaheadofallotherassignees,confirmingitsdominantpositioninlargelanguagemodel–relatedinnovation.Baiduhasfurtherstrengtheneditsposition,decisivelytakingthelead,movingfromsecondplaceinthepreviousyear

3

tothetopglobalrankintermsofpatentapplicationvolume.ThisprogressionhighlightsBaidu’sacceleratinginvestmentinLLMtechnologiesandunderscoresitsgrowinginfluenceinnext-generationfoundationmodels.

TencentTechnologyfollowsinsecondplacewith1,139patentfamilies,whileMicrosoftTechnologyLicensingandGoogleoccupythenextpositions,combiningsubstantialportfoliosizeswithcomparativelystronginternationalpatentcoverage.ThesecondtierisformedbySamsungElectronicsandAlibaba,eachholdingmorethan600patentfamilies,thoughwithdifferinglevelsofinternationalprotection.Mid-rankedplayerssuchasIBM,PingAnTechnology,NEC,Intel,Huawei,Qualcomm,andNVIDIAhavemoremoderateportfoliosizes,generallyaccompaniedbylowerinternationalfilingvolumes.

3

/resourcehub/consolidation-and-opportunities-in-llm-domain/

13

Copyright?2025Questel–Allrightsreserved

Queste

IV.FocusonMultimodalAI/DigitalHumans/IntelligentAgents

1.Introduction

Artificialintelligenceiscurrentlyenteringanewphasedrivenbythreecloselyrelatedinnovationareas:multimodalAI,digitalhumans,andintelligentagents.Ourpatentanalysisidentifiesaround15,000patentfamiliesinMultimodalAI,8,000inDigitalHumans,and

6,000inIntelligentAgents.Thesetechnologiesareclearlyemergingtogetherandreinforcingeachother,creatinganewgenerationofAIsystemsthatcanperceive,communicate,andactinwaysthatwerenotpossibleafewyearsago.

MultimodalAIreferstoAIsystemsthatcanunderstandandgeneratemorethanonetypeofdataatthesametime,suchastext,images,audio,andvideo.

IntelligentagentsareAIsystemsdesignedtopursuegoalsautonomouslybyreasoning,planning,usingtools,andtakingactionsovertime.

DigitalhumansareAI-poweredvirtualcharactersthatcanspeak,showfacialexpressions,andinteractwithusersinahuman-likeway.

MultimodalAI,DigitalHumansandIntelligentAgentsfilingdynamics

7000

6000

MultimodalAI

5000

4000

3000

DigitalHumans

2000

1000

IntelligentAgents

0

201720182019202020212022202320242025

Figure6-Multimodal,DigitalHumansandIntelligentAgentsfilingdynamics,2017-2024

14

Copyright?2025Questel–Allrightsreserved

Questel

2.MultimodalAI

MultimodalAI:wheremodelleadershipandpatentpowerdefinethenextAIrace

Theyear2025markedaturningpointformultimodalAI,withmajorlaunchesofnewmultimodalenginesbytheleadingplayersinthefield.TheselaunchesclearlyshowthatthetechnologicalraceisacceleratingandthatbuildingstrongpatentportfoliosinmultimodalAIwillbeakeystrategicassetforthefuture.Whilesomecompaniesdominatepublicvisibilityandmarketadoption,othersarepositioningthemselvesthroughintensivepatentfilingstrategies.

Figure7-TimelineofmajorAIModel,2023-2025

MultimodalAIisoneofthemostimportantfoundationsofcurrentAIinnovation.Itsobjectiveistoenablemachinestounderstandandcombinemultipletypesofinformation.suchastext,images,audio,andvideowithinasinglemodel,inawaythatisclosertohumanperception.Ratherthanprocessingeachmodalityseparately,multimodalmodelslearnsharedrepresentationsthatconnectdifferentinputsandallowmoreadvancedreasoningandinteraction.

OPENAIisoneofthemostvisibleinnovatorsinthiscategory.WithmodelssuchasGPT-4,GPT-4V,andGPT-4o,OPENAIfocusesonunifiedmultimodalreasoningacrosstext,images,andaudio,withstrongemphasisonreal-timeinteractionandagentic

15

Copyright?2025Questel–Allrightsreserved

Questel

capabilities.GPThasbecomealmostsynonymouswithAIinpublicspaces.However,despitethisleadershipinperceptionandadoption,OPENAIdoesnotappearamongthetoppatentowners.

GOOGLEisanothermajorcoreenginedeveloper.TheGeminifamilyofmodelswasdesignedasnativelymultimodalfromtheoutset,integratingtext,images,audio,andvideowithinasinglearchitecture.GOOGLE’sstrategyemphasizesdeepreasoning,long-contextunderstanding,andtightintegrationwithitsecosystem,includingSearch,Android,andcloudservices.Thistechnologicalleadershipisalsoreflectedinastrongandinternationallyorientedpatentportfolio.

BAIDUplaysasimilarrolewithitsERNIE4.xmodels.ERNIEisamultimodalfoundationmodelcapableofunderstandingtext,images,audio,andvideo.Thisstrategyisstronglysupportedbypatentfilings,positioningBaiduasbothatechnologyandIPleader.

OtherimportantenginedevelopersincludeANTHROPIC(Claude3),iFLYTEK(SparkLLM),ALIBABA(Qwen,withastrongenterprisefocus),andIBM(Watsonx.aiandGranite,alsotargetingenterpriseusecases).Open-sourcecontributorssuchasMETA(notablythroughLLaVA-stylemodels)andMISTRALalsoplayarole,althoughgenerallyatamorelimitedscalecomparedtothelargestfoundationmodelproviders.

Figure8-MultimodalAItopindustrialplayers,2017-2024

16

Copyright?2025Questel–Allrightsreserved

Questel

GOOGLEleadsthepatentlandscapewiththehighestnumberofmultimodalAIpatentfamiliesandastronglevelofinternationalprotection.BAIDUfollowsclosely,withalargeportfoliothatreflectsitslong-termcommitmenttomultimodaltechnologies.

TENCENT,ALIBABA,ANDPINGANTECHNOLOGYalsoholdsubstantialnumbersofpatentfamilies,thoughtheirfilingsaremoreheavilyconcentratedondomesticprotection.

MICROSOFT,HUAWEI,ANDSAMSUNGELECTRONICSstandoutfortheirbalancedstrategies,combiningsignificantpatentvolumeswithahighproportionofinternationalfilings.TraditionaltechnologyleaderssuchasAPPLE,NVIDIA,andMETAshowmoremoderatepatentvolumesbutstillmaintainselectiveinternationalcoverage.

3.IntelligentAgents

Fromconversationtoaction:intelligentagentsarebecomingtheoperatinglayerofAI

IntelligentagentsareAIsystemsdesignedtooperateautonomouslyoverextendedperiods.Ratherthansimplyrespondingtoprompts,theycanpursueobjectives,breaktasksintosteps,useexternaltools,andadapttheirbehaviorbasedonfeedbackandmemory.Thisfieldhasexpandedrapidlywiththeriseoflargelanguagemodels.

Figure9highlightskeyassigneesinthefieldofIntelligentAgents,showingboththetotalnumberofpatentfamiliesandtheirinternationalcoverage.

17

Copyright?2025Questel–Allrightsreserved

Questel

Figure9-IntelligentAgentstopindustrialplayers,2017-2024

IBM(withwatsonx.aiplatformandtheGraniteFoundationModelFamily)leadsintermsoftotalvolumepatentvolumeinthisfield.Strategically,IBMpositionsintelligentagentsfocusonsolutionsforprofessionalsandcompanylevel.

GOOGLE(withGeminiAgents),BAIDU(withGenFlow/AgentBuilder)andMICROSCROFT(withCopilot&agentframeworks)followintermofpatentfilingwithverylargepatentportfolioandpositionsGeminiasanagenticmodelcapableofreasoning,planning,andinteractingwithtools.MicrosoftbuildsonthesefoundationswithCopilotproductsintegratedintoenterprisesoftware,enablingtaskautomationacrossdocuments,emails,andworkflows.

OpenAI(9patentfamilies,including5IPFs),NVIDIA(11patentfamilies,including6IPFs),andHuawei(also11families,including9IPFs)holdfewerpatentsoverallbutretainagoodshareofinternationalfamiliesrelativetothesizeoftheirportfolios.

Ontheotherhand,AMAZON,ALIBABA,PINGANTECHNOLOGY,TENCENTT