WhereWillTomorrow’s

AI

Geniuses

Go?July2025By

NikolausS.

Lang,

LeonidZhukov,

EtienneCavin,and

Johann

HarnossEvena

modestdecline,

by

USstandards,intheinflowof

AI

talentcouldsignificantly

boostothercountries’effortstobecometechinnovation

hubs.Thiswould

be

particularlybeneficialtotheGenAI“middle

powers”—includingthe

EU,

the

UK,Canada,

Japan,andsomeGulf

countries—thatare

hopingtocompeteasglobalsuppliersofthe

technology.Theopportunityis

ripeforthosethat

beton

researchbreakthroughsassourcesof

competitiveadvantage.

Infact,thestakesforattractingtopAI

researchtalent—the

AIgeniusesof

tomorrow—willonlyincreaseas

GenAIfoundation

modelscontinueto

becomecommoditized.

In

thisendstate,innovationwill

bethe

maindifferentiator,

andcountriesabletocapitalizeon

the

shift

stand

towin

the

most.TheglobalcompetitionforAItalentis

not

new,

butitsintensity

hasincreasedasa

resultof

changing

US

policies.

While

manyCEOsare

understandablyfocusedontariffsandthe

risingcostof

datacenterexpansioninthe

US,

we

believethatchangesinimmigration

policiesand

publicfundingfor

R&D

may

havegreater

long-termeffectson

globalcompetitionforAI

leadership.Theworld’s

bestand

brightestAI

minds

have

long

beendrawn

tothe

US.

Butas

other

countries

increasinglyseektoattract

these

researcherswhile

retainingtheir

own

home-growntalent,the

map

of

AIinfluencecouldshift.BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?2Despitethis

uncertainty)

recent

moves

by

UStechgiantssuggestthattariffswill

not

underminethe

US’sglobal

lead

incomputing

power.AsofFebruary

2025,

Microsoft,Alphabet)Amazon)and

Metaweresettocollectivelyspend

morethan$320

billionincapexthisyear)

about

30%

more

thantheyspentin

2024.AndfollowingtheTrumpAdministration’s

reciprocaltariffsannouncementonApril

2,earningscallsforthefirstquarter

of2025showed

nosignsof

the

UStechgiantsscaling

backtheirAIinfrastructurespending.

Infact)

Metaincreasedits

2025capexguidancefrom

between$60

billionand

$65

billionto

between$64

billionand$72

billion—a

revisionthatimplicitly

pricestheall-intariff

effectat6%

to

10%

of

constructioncost.AmidTariffUncertainty,

USTech

Giants

ForgeAheadOnApril

1)

2025)the

US

Departmentof

Commerceopenedaninvestigationintothe

nationalsecurityimplicationsof

importingsemiconductorsandsemiconductor

manufacturingequipment.The

results—

whichasof

thiswritingarestill

pending—will

determine

thetariffrateappliedtosemiconductorsandservers,directlyimpactingthecostof

datacenterexpansionin

the

USandtheavailabilityof

computing

power.

(Serversaccountfor

morethan

half

ofa

USAIdatacenter’stotal

costof

ownership.)BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?3EXHIBIT1USTech

Giants

Havethe

Fundsto

Expand

Computing

Power,

as

Do

US

GenAI

DevelopersCASH

ON

HAND

FOR

US

TECH

GIANTS($BILLIONS)1

FUNDS

RAISED

BY

LEADING

GENAI

LABS(TTM,$BILLIONS)220.6Thisis

partlyasensible“wait

and

see”

response

onthepartofUStechgiants.

Butit’salso

unsurprisingin

light

of

other,tariff-independentfactors.

First,there’sthesheercapital

powerof

thesecompanies,whichallowsthemtoabsorbthe

highercostsof

datacenterexpansionfor

thesakeof

whattheyopenly

regardas

an

existentialimperativeof

AIsupremacy.Second,withtheTrumpAdministration’s

repealof

the

Biden-eraAI

Diffusionframework,

US

hyperscalersarevirtually

unconstrainedin

termsofpursuingdatacenterexpansioninother

countries.Furthermore,TSMC’sexpanding

productionfootprintinthe

UScould

reduceexposuretotariffsfor

USdata

centers

relyingon

leading-edgesemiconductorsandservers,includingthosedesigned

by

Nvidiathatarecurrently

manufacturedabroad.US

hyperscalers—and

leadingGenAIdeveloperssuchas

OpenAI,Anthropic,andxAI—arealsostartingwithasizable

leadin

bothcapitalandcomputing

powercomparedtocompaniesinothercountries.(The

sameis

truewhenitcomestoinstalledcomputing

powercapacity

inthe

US,which

reaches40gigawatts,comparedto

8gigawattsinthe

EUasawhole,

for

example.)

UShyperscalerscollectivelysiton$145

billionin

cash,

and

leadingGenAI

playersinthe

US

have

raisedover$59billioninthe

lasttwelve

monthsalone—morethan

23

timesas

muchasthethree

largest

non-US

GenAIdeveloperscombined.(SeeExhibit1.)Againstthat

background,weexpectthe

USto

holdits

lead

incapitalandcomputing

power—two

hugesources

ofadvantageintheglobalGenAI

race. USOthercountriesSources:Tracxn;company

reports;Crunchbase;

Reuters;

DatacenterDynamics;

BCG

Henderson

Instituteanalysis.1As

of

May

2025.2TTM=trailing

twelve

months

for

period

ending

May

20,

2025.

For

OpenAI,

?gure

includes

$10

billion

in

upfront

investment

from

the

total

of$40billionraisedinApril

2025;the

remaining$30

billionisslated

for

later

in

2025,

partly

conditional

on

OpenAI’s

converting

into

a

for-pro?t

company.28.8

28.823.3BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?4Amazon

Microsoft

Meta

Google

OpenAI

DatabricksCohere

MistralAI0.70.7AnthropicZhipuAI66.215.311.512.0xAI1.1The

LargerTalent

Pool:TheOverallAIWorkforceOverthe

pastthreeyears,the

USsawa

netinflow

of32,000foreignAIworkers,or

roughly7%of

the

country’s

total

pool(488,000asof2025).1

Theseworkersare

even

moreimportantto

leading

UStechcorporations—where,

inaggregate,theyoccupy

nearly40%of

US-based,AI-related

roles.(SeeExhibit2.)Giventheircritical

roleintheAIworkforce,anexodus

offoreignworkersalreadyinthe

US—ora

declineinattractivenessofUS-based

rolesforqualifiedimmigrants—

could

poseaserious

riskfor

UStechfirms.

Indeed,somereports

have

pointedtostatisticsshowingdecliningtechtalentflowsintothe

USasa

signthat

this

riskis

alreadymaterializing.2

Butthistrend

predates

recentchangesinUS

policy

by

morethantwoyears.

Moreimportantly,it’slargelyattributabletodeclining

hiring

ratesinthe

UStech

sectorand

lower

levelsof

overallglobaltalent

mobilityover

thatsame

period.(SeeExhibit3.)Whenweconsiderthe

potentialeffectsofrecent

US

policy

shiftsonanother

keyenablerof

GenAIstrength—talent—

the

picturegets

morecomplex.

Here,animportantdistinction

needsto

be

made

betweentheworkerswho

havethetechnicaltrainingtodevelopandimplementAI

solutionsandthe

researchers(usuallywithadvanceddegreesincomputerscienceand

relatedfields)who

collectively

pushthetechnologicalfrontier.While

new

US

policiesonimmigrationwill

likely

resultina

decreaseof

foreigntalentcomingtothe

US,inourviewthesechangesare

not

likelyto

materially

underminethe

US’sabilityto

retainandattractAIworkers

in

the

nearfuture.

Butthosesame

policyshifts,coupledwithreductionsin

public

R&Dexpenditures,could

haveasignificanteffectonthe

nextgenerationof

AIgeniuses—

thesmalleryetdisproportionatelyimportantgroupof

top

AI

researcherswho

havefordecadesflockedto

USacademicinstitutions.The

ImpactonAITalent1.

AIspecialistsaredefinedasworkerswith

knowledgeindeep

learning,computervision,

reinforcement

learning,and

neutral

networks,amongotherskills.2.

“Reports:

US

Losing

EdgeinAITalent

Pool,”Semafor,

May2,2025.BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?5

Non-US

undergraduatedegree

recipientsEstimated

share

ofUS

undergraduate

degree

recipientswho

are

foreign

born2Sources:

LinkedIn;

universitycommondatasets;

BCG

Henderson

Instituteanalysis.1Basedoncountrywhere

undergraduatedegreewas

obtained.2Re?ectstheweightedaverageshareof

international

undergraduatestudents

at

the

US

universities

represented

in

theAI

employee

base

of

the

companies

showninthe

chart.EXHIBIT3DecliningAITalent

In?owstothe

US

Have

Been

Driven

by

a

Hiring

Slowdown

and

Overall

Reduced

MobilityINDEXED,SEPTEMBER2022=1EXHIBIT2US

GenAI

Firms

Have

a

Sizeable

Share

ofForeign-BornAIWorkersESTIMATEDSHAREOF

US-BASED,AI-RELATED

JOBS

HELD

BY

FOREIGN

NATIONALSAT

LEADING

USGENAI

LABS11.00.80.70.60.50.3AIspecialistnet

in?owstothe

USGlobalAItalent?owsUSsoftware

developer

jobpostings1Sources:

Indeed

(via

Federal

Reserve

Economic

Data);

BCGGlobalTalentTracker;

BCG

Henderson

Instituteanalysis.

1Dailysoftwaredevelopment

job

postingson

Indeedinthe

US,

averaged

out

over

previous

12

months.10%–12%26%BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?61.00.80.60.40.80.60.3FTEweightedaverage0.2Sept.

2022Sept.

2023Sept.

2024AnthropicDec.2024Databricks36%–38%MicrosoftOpenAIGoogleAmazonMeta28%25%27%27%17%24%24%21%xAIEXHIBIT

4ForeignAIWorkers

Have

ConcernsAbout

US

Policy

ShiftsInthe

lastthree

months,

have

non-US-basedAIworkerscited

US

policies

or

perceptions

as

negativelyimpacting

their

opennessto

relocatingtothe

US?(%OF

SURVEYED

RECRUITERS)Admittedly,thereis

muchgreatervarianceamongrecruiterswhenaskedaboutthe

nextsixto

twelve

months;

aconsiderableshareexpectaslight

downturn

in

the

US’s

abilitytoattractand

retainforeignAIworkers

overthatperiod.

However,veryfewexpectthedownturnto

besignificant.Andforgood

reason.(SeeExhibits5.1一5.4.)Despitethechanging

policyclimate,the

US

remainsahighlycompetitivedestinationforqualifiedforeignworkers.

Accordingto

H-1Bvisaapplicationfilingsfrom

2024,salariesfortechnicalAI

rolesattop

USGenAI

labs

likeOpenAI,Anthropic,andxAIareonaverage

twice

as

high

asButwhen

pushcomestoshove,immigration

policyshifts

haven’t

hada

materialeffectonthe

attractivenessofUS-basedtech

jobs,

recruiterssay.Some

have

noted

more

challengesattractingor

retainingforeign

nationalsforsuch

jobs—buta

large

majority

haveobserved

nochange,or

an

actualimprovement,intheattractivenessofUScompanies.The

neteffect,accountingforallthesereportedshifts,

has

been

negligibleinthe

lastthree

months:a

1%declinein

reportedattraction

and

a

3%

declinein

retentionforforeign

nationalsinAI-related,

US-based

roles.comparable

rolesatforeign

peerssuchas

Cohere(Canada),AI21

Labs(Israel),

MistralAI(France),Technology

Innovation

Institute(UAE),andG42(UAE).Thisassessment

may

understatethegapintotalcompensation,whichtypicallyincludesstockoptionsandbonusesinadditionto

salaries.Thecompensationgapcouldwideneven

more,

as60%

of

the

recruiterswesurveyedexpect

UStechcompaniestoenhancefinancialcompensation

packagesinthe

nextyear

toattractthe

besttalent.Togetaclearersense

of

what

has

happened

so

far

thisyear,wesurveyedagroupof

40

senior

recruiters

at

tech

companiesandtalentagencies.3

Theytold

usthat

USpolicyshifts

have

beenshowing

upindiscussionswithprospective

recruits,with

mostconcerns

being

relatedto

stricterimmigration

rulesand

processes.(SeeExhibit4.)Source:

BCG

Henderson

Institutesurvey.Note:

N=40.

Surveyconductedwithseniorin-house

recruitersfromtopAI

companies

and2025.

Because

of

rounding,

percentages

shown

do

not

add

up

to

100%.BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?7RPOs/agenciesworkingwith

highlyquali?edAIworkersin

May3.Survey

conducted

between

May14and

May30,

2025.Did

not

citepolicyconcernsYes,sometimesYes,

but

rarelyYes,oftenNot

sure18%33%33%13%5%Out?ows3%Signi?cantdecrease8%Slightdecrease35%Nochange45%Slightincrease3%Signi?cantincrease8%UnsureIn?ows5%Signi?cantincrease25%Slightincrease25%Nochange30%Slightdecrease8%Signi?cantdecrease8%UnsureSource:

BCG

Henderson

Institutesurvey.Note:

N=40.

Surveyconductedwithseniorin-house

recruitersfromtopAI

companies

and

RPOs/agencies

working

with

highly

quali?ed

AI

workers

in

May2025.

Because

of

rounding,

percentages

shown

for

each

chart

do

not

add

up

to

100%.EXHIBIT5.2SalariesforAI

Roles

at

Leading

US

LabsAre

onAverage

Double

Thoseof

Non-US

Labs1SALARIES

FOR

TOP

AI-RELATED

POSITIONS($THOUSANDS)EXHIBIT5.1Recruiters

Expectthe

USto

Slightly

Decline

as

anAITalent

Magnet

inthe

Next

YearExpected

change

in

the

movement

of

AI

workers

to/from

the

US

in

the

next6–12

months(%OF

SURVEYED

RECRUITERS)Sources:

US

DepartmentofLabor;

levels.fyi;

BCG

Henderson

Instituteanalysis.Note:

Rolesanalyzedinclude

“Memberof

thetechnicalsta?”for

OpenAI,Anthropic,xAI;

“AI

Research

Scientist”

for

Meta;

“Applied

Scientist

III”

forAmazon;

“ML/AISoftware

Engineer”forCohere,A121

Labs,

MistralAI;

“Machine

Learning

Engineer”forG42;

“Software

Engineer”

for

TII.Analysis

does

not

includestockoptions

and

bonuses.1ExcludingChina.BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?8Non-USAverage:~$134K1USAverage:~$267KAnthropicAI21

LabsMistralAIOpenAIAmazonCohereMetaG42158248353320133203124127127214xAITIISalariesandotherformsof?nancialcompensation53%Opportunitiesforcareergrowth

(incl.entrepreneurship,access

tofrontier

research)30%Immigration/visaopportunities

and

path

to

10%permanent

residencyQualityoflife/culture(incl.personal/familyconsiderations

and

political/socialstability)8%Source:

BCG

Henderson

Institutesurvey.Note:

N=40.Survey

conducted

with

senior

in-house

recruiters

from

top

AI

companies

and

RPOs/agencies

working

with

highly

quali?ed

AI

workers

in

May

2025.EXHIBIT5.4Financial

Compensation

Remainsthe

LeadingAttractionfor

AIWorkersTOP

FACTOR

LEADING

NON-US

AI

TALENT

TO

SEEK

RELOCATION

TO

THE

US(%OF

SURVEYED

RECRUITERS)EXHIBIT5.3Recruiters

Expect

USTech

Firmsto

Improve

Financial

Packagesto

Compensatefor

Policy

ShiftsExpectedchangeinsalaryand?nancial

compensation

o?ered

by

UScompanies

to

attract

foreign

AIworkers

in

the

next6–12

months(%OF

SURVEYED

RECRUITERS)15%Source:

BCG

Henderson

Institutesurvey.Note:

N=40.

Surveyconductedwithseniorin-house

recruitersfromtopAI

companies

and

RPOs/agencies

working

with

highly

quali?ed

AI

workers

in

May2025.

Because

of

rounding,

percentages

shown

do

not

add

up

to

100%.BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?90%Slightly

lowerSigni?cantly

higherSigni?cantly

lowerSlightly

higherNochange45%35%5%Shareof

foreignnationals67%55%TopAI

researchers

at

US-basedinstitutions1TheSmallerTalent

Pool:Top

AI

ResearchersThe

picture

looks

markedlydifferentwhenweconsiderthe

muchsmaller

poolofhighlyskilledacademic

researcherspushingthefrontierof

AI.There,the

USisfar

moredependentontalentinflowsfromabroadthanin

theprivatesector.

In

2023,foreign

nationalsaccountedfor55%

ofall

USdoctoratesincomputerscienceand

math,

nearly

three-quartersof

whomwereChineseor

Indian

nationals.

These

highlyspecializedstudentstendtostayinthe

USafter

receivingtheirdegrees,therebycontributingtothe

USinnovationecosystem.

It’s

nosurprisethat,of

the

100

most-cited

US-basedAI

researchers,67%areforeignnationals.(SeeExhibit6.)This

relativelysmall

poolofacademic

researchers

playsa

vital

roleinthe

longarcof

theGenAI

race.

While

UStechnologycorporations

have

become

researchpowerhousesintheirown

right,academia

remainscritical

totheinnovationecosystem.Academicinstitutionsare

by

designthe

best

placestofosterexploratory

researchof

the

sortthatoftenyieldsfundamentalintellectual

property(IP)

breakthroughs—andevery

nowandthen,flashesof

genius

that

leadtogreatadvancementsintheirfields—precisely

because

universitiesare

unfettered

byconcernsaboutnear-termfinancial

returns.Thesymbiotic

relationship

betweenacademicandcorporate

researchis

reflectedintheir

respectivesharesof

spendingalongthe

R&D

pipeline.

Inthe

US,academiaisthe

largestcontributorto

basic

research,while

businesses

playthe

largest

roleinapplied

research(63%)

andexperimentaldevelopment(92%).

Inshort,academiais

key

toenablingtheexplorationofideasthat

areeventuallytaken

up

by

businesses.(SeeExhibit7.)Such

breakthroughsintime

becomecatalystsforcorporate

R&D.

Recent

reportsofMeta

reducingtheresourcesallocatedtoits

FundamentalAI

Researchgroup

infavorofmorecommerciallyorientedGenAI

applications,

if

true,illustrate

how

businesses

maychoosetofavorinvestmentswithclear

returnsoverfrontier

researchwith

an

uncertain

payoff.EXHIBIT6ForeignTalent

Makes

Up

a

Large

Share

ofUSAI

ResearchersSources:SurveyofEarned

Doctorates

2023

(NationalScience

Foundation);AMiner;

BCG

Henderson

Institute

analysis.1N=139,corresponding

to

the

US

share

of(i)the

world's

top100

researchers

across

AI

sub?elds

and

(ii)

the

world's

top

100

machine

learning

researchers,

bothbasedon

numberof

citations.Shareof

foreign

nationalsestimated

based

on

non-US

bachelor

degrees

and

additional

background

research

when

relevant.

IranBangladesh

India··-

··

————Turkey

.

.· South

KoreaMainlandChina0%10%20%30%40%50%SHAREOFTOTAL

FOREIGN-BORN

USCOMPUTERSCIENCEAND

MATH

PHD

RECIPIENTS,

BY

PLACEOFORIGIN%OF

FOREIGN

DOCTORATE

RECIPIENTS

WITH

AN

INTENT

TOSTAY

INTHE

USBOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?10

Taiwan

100806040200PhD

recipientsincomputerscienceand

math本報(bào)告來(lái)源于三個(gè)皮匠報(bào)告站（）,由用戶Id:349461下載,文檔Id:731915,下載日期:2025-07-24●SaudiArabia UKCanadaThistypeof

divisionoflabor

betweenacademiaand

private

corporations

has

beencriticaltothedevelopmentof

GenAI

itself.Google’s2017Transformerarchitecture,thecornerstoneof

today’s

large

language

models,wasonlypossible

becauseof

the(then-unfashionable)workondeep

neural

networkscenteredaroundthe

Universityof

Toronto

inthetwodecades

prior.Whenideasincubatedinacademia

beginto

mature,

privateinvestmentinAIstartsto

rise—boostingapplied

researchinthefield.(SeeExhibit8.)The

recent

historyof

AI

mayseemtosuggestthatthelocationofacademic

researchis

not

particularlyimportant,

as

UScompanies

have

beenthe

beneficiariesofbreakthroughsincubated

primarilyinCanadianinstitutions

betweenthe

1970sand

1990s.

Butacademic

researchinstitutionscanconfersignificantadvantagesto

acountry’s

innovationecosystem,as

bothtalentdevelopment

hubs—

attractingandtrainingthetalent

necessaryto

buildadvancedtechfirms—andasincubatorsofbusinessideas

intheirown

right.(Forevidence,

look

nofurther

thanSiliconValley.)

Empirical

researchshowsthatthespillover

ofacademic

knowledgeisstrongly

localized:

more,andbetter,corporate

patentsare

produced

near

universities.4Andwhileacademic

researchisoftenopen

andconferslittle

IPadvantage,thisisachoice,

not

aninevitability.Universitiescan

restrictaccesstoor

monetizetheirresearch,andtheyappear

more

likelyto

do

so

nowthan

theydidinthe

recent

past:inthe

US,academia’sshare

of

allAI

patentfilingswiththeWorld

Intellectual

PropertyOrganization

rose

by40%

between

2021and

2023comparedtothe

periodfrom

2013to

2020.Insum,where

researchers

liveandwork

matters.Andbecauseso

manyof

themareimmigrants,theirgeographic

footprintis

highlysensitivetoimmigration

policy,accessto

researchfunding,andfinancialcompensation.Ontheimmigrationfront,thesuspensionofnew

F-1and

J-1visas,

aswellas

plansto

limittheOptional

PracticeTrainingprogramfor

postgraduationemployment,will

makeitmorechallengingfor

USacademicinstitutionstoattract

foreignstudents.EXHIBIT

7Academia

and

Business

Play

Complementary

Rolesinthe

Innovation

EcosystemUS

R&D

EXPENDITURES,

BY

SECTOR,SOURCE

OF

FUNDS,AND

TYPE

OF

R&D(2022,$BILLIONS) AcademiaBusinessFederalgovernmentOthersSource:

Researchand

Development:

USTrendsand

InternationalComparisons

(NSF);

BCG

Henderson

Instituteanalysis.1Experimentaldevelopmentissystematicwork,drawingon

existing

knowledge

gained

from

research

and

practical

experience,

that

is

directed

to

producing

new

materials,

products,anddevices;toinstalling

new

processes,systems

and

services;

orto

improving

substantially

those

already

produced

or

installed.4.Naomi

Hausman,“University

Innovation

and

Local

Economic

Growth,”Review

of

Economics

and

Statistics

104,

no.

4

(2022);

Sharon

Belenzon

and

MarkSchankerman,“Spreading

the

Word:Geography,

Policy,and

Knowledge

Spillovers,”Review

of

Economics

and

Statistics

95,

no.

3

(2013);Anna

Valero

andJohnVan

Reenen,“The

Economic

ImpactofUniversities:

EvidencefromAcrosstheGlobe,”

NBERWorking

Paper

No.w22501(August2016);

Leonie

Koch

and

MartinSimmler,“How

ImportantAre

Local

KnowledgeSpilloversofPublic

R&DandWhat

DrivesThem?”

EconPolWorking

Paper42,ifo

Institute–Leibniz

Institutefor

Economic

Researchatthe

UniversityofMunich(February2020).Basic

research

Applied

research

Experimentaldevelopment1BOSTON

CONSULTING

GROUP

WHERE

WILL

TOMORROW’S

AI

GENIUSES

GO?11$129$160$59610080604020045%36%16%63%92%Cutsto