Full transcript
オープニング
0:00You have compiled and published a
0:01report providing a detailed analysis of
0:03Japan's digital deficit . In the report ,
0:05we created two categories : " hidden
0:07digital deficit " and " SDX deficit . "
0:09Combining the automotive ,
0:11telecommunications , and industrial
0:13machinery sectors , we concluded that in
0:16a pessimistic scenario , there would be
0:18a deficit of approximately 45.3
0:20trillion yen . It is crucial to report
0:22on technologies that can contribute to
0:24the nation by addressing issues unique
0:26to Japan . In San Francisco , they
0:28probably don't even say " AGI " or " AI "
0:30anymore . Recently , people at frontier
0:33labs have started calling it " PAS
0:35frontier , " and what happens after that
0:37is they just naturally roll out the new
0:40Opus 5.5 . To justify this valuation ,
0:42you have to do something that
0:43essentially destroys every existing
0:45market ; otherwise , it’s unjustifiable
0:48. They might not know if they are
0:51heading toward ruin or not , but they
0:52run while acknowledging that they are
0:54destined to keep running at full speed .
0:58For learning about real estate
0:59investment , go to Rakumachi . Hello to
1:04everyone watching the Rakumachi Channel
1:06. Our theme today is : Can Japan break
1:08its digital deficit structure ? We will
1:10be speaking with this guest today . He
1:12is Mr. Sando Tsuda , who just resigned
1:13from the Ministry of Economy , Trade and
1:15Industry at the end of August . Mr.
1:16Tsuda , thank you for joining us today .
1:18>> Thank you for having me .
1:19>> During your time at the ministry , you
1:21led a team of young bureaucrats under
1:23the age of 30 to compile and publish a
1:25detailed report analyzing Japan's
1:27digital deficit . Combined with your
1:30look , you have an image as a unique
1:32bureaucrat to those who follow economic
1:34media , but your true profession is that
1:36of an entrepreneur and engineer . This
1:39time , you didn't just stop at the
1:41analysis report ; you resigned after one
1:44year to start a business based in Japan
1:46to earn foreign currency . What is
1:48missing in order to solve the
1:50structural problem of Japan's digital
1:52deficit ? And how will the situation
1:54change due to the current AI
1:55development race ? I hope to discuss
1:57this in detail today . First of all , why
1:59did you decide to resign and start a
2:01business at this particular time ?
データに飲み込まれる世界
2:04>> I wrote the Digital Economy Report last
2:06year ,
2:09>> Well , I released it before Golden Week
2:11in 2025 , and the subtitle of that
2:13report really captured what I wanted to
2:16say most .
2:18>> Uh-huh . It was titled " Survival in a
2:21Borderless Digital Market in a World
2:23Consumed by Data . " Back then , the idea
2:27of a " world consumed by data " didn't
2:29really click with many people , though
2:31the " borderless digital market " part
2:34sort of did .
2:36>> You had Tesla and BYD releasing these
2:38software-defined products ,
2:43>> And Elon Musk was also rolling out
2:45Starlink , while Jensen Huang had
2:47started using the term " Physical AI . " I
2:54think there was already a sense that
2:56things were becoming borderless .
3:03>> But over this past year , as frontier
3:05labs have been churning out models , I
3:07think everyone has started to
3:08intuitively grasp that data and
3:10business-specific context are the true
3:12roots of corporate value .
3:18>> In that sense , I titled it " Data Eats
3:20the World . " Marc Andreessen wrote "
3:28Software Eats the World " back in 2011 ,
3:31and about 15 years later , as this
3:33paradigm is truly taking hold in the
3:35market , I’ve realized that while
3:37I’ve been working on policy as a
3:39chief architect for national data
3:41management infrastructure and
3:43open-source projects , we finally need
3:45real players to step up .
3:52>> At this juncture ,
3:55>> The reality of industrial policy is
3:57that it only works when you have
3:59players in the game ; in blue ocean
4:01markets , there are very few effective
4:03policy measures you can take . That's
4:07why I intentionally included some tough
4:09truths in the report . I’m not just
4:14speaking from a policy perspective ;
4:16I’ve spent more time as a business
4:18owner , and I felt it was time for a
4:20next-generation player to lead the way ,
4:22which is why I’m here now .
4:26>> In terms of my own career , I felt this
4:28was where I could find the greatest
4:29leverage , so that's why I chose to
4:31start my own business . Conventionally ,
4:33when talking about the digital trade
4:35deficit , people often point out how
4:37everyone buys iPhones . Then you use
4:38Amazon to order goods , post on
4:40Instagram , or watch videos on YouTube ,
4:43and every time you do , money — Japanese
4:45yen — flows to the US through
4:47subscriptions , advertising , and other
4:49channels . That was the typical
4:52explanation , but now , the digital
4:54deficit is increasingly spilling over
4:56into the manufacturing sector ,
4:58including cars like Teslas and even
5:00rockets . So that is what you mean by a
5:03digital market without boundaries ?
5:05Exactly . I coined the term " SDX , " or "
5:11Software-Defined Everything , " to
5:13describe this hidden deficit , where the
5:15hierarchy between hardware and software
5:17is reversed , and software increasingly
5:20defines the hardware and the market
5:22itself ; and that's exactly what's
5:24happening with automobiles — at the end
5:26of the day , Tesla doesn't just want to
5:28sell cars , but is focused on something
5:30like Mobility as a Service . The car is
5:37merely a device they sell to provide
5:39that service . In that sense , they build
5:43their own in-vehicle OS and develop
5:45their own autonomous driving and ADAS
5:47technology .
5:50>> To do that , their software components
5:53must be loosely coupled — what's called
5:55an E / E architecture — so they build
5:57cars to be more like software than
5:59traditional vehicles , and they even go
6:01as far as building and installing their
6:03own Superchargers , a business model
6:06that is largely unthinkable for
6:07Japanese OEMs . So , including those
6:12aspects , across various areas — even
6:14extending into the physical realm —
6:16there's a shift happening . In San
6:20Francisco , they’ve recently been
6:22calling it " bits to atoms . " We're
6:25gradually shifting from bits to
6:27physical atoms , but even then , those
6:29atoms are ultimately being defined by
6:31software . The idea that we can make
6:35businesses more scalable by defining
6:37physical things through software is
6:39definitely spreading . And it’s clear
6:43that what’s driving this is AI
6:44frontier models . Since writing this
6:48digital economy report , I feel like I'm
6:51in a situation where I’m witnessing
6:53the very things I wrote about actually
6:55playing out in reality .
日本の弱点はミドルウェア
6:57>> Yes .
6:58>> You mentioned that an important point
7:00is that even if the government invests ,
7:02there must be players to receive that
7:04investment . The current Kishida
7:05administration has announced 370
7:07trillion yen in industrial investment
7:09through 2040 , but is there a current
7:11reality where Japan is facing a
7:13shortage of players in this specific
7:15digital field ?
7:16>> Hmm .
7:16>> Well , that's true . You see , software
7:19isn't just one thing ; there are various
7:21types of businesses . Generally ,
7:23software is structured in three layers :
7:26the application layer , the middleware
7:28layer , and the OS layer . Historically
7:31speaking , the OS came first . There was
7:33Windows , then came other systems , and
7:36afterward , " applications " emerged , in
7:38the sense of applying those operating
7:40systems . At first , there were only
7:42operating systems and applications . But
7:47as these applications grew to include
7:49SaaS and various peripheral tools , it
7:51became clear that they shared common
7:53components . This led to the emergence
7:58of data infrastructure , authentication
8:01systems , and recently , ontologies and
8:03AI foundation models — all of which
8:05serve as middleware . In Japan , the SaaS
8:10business began to expand right around
8:13the time I was living in Estonia , from
8:152019 to 2020 , when many SaaS companies
8:18started going public .
8:20>> SaaS in the US began in the early 2000s
8:23, so it took about 19 years for similar
8:27B2B , not B2C , SaaS models to finally
8:30emerge in Japan . In that sense ,
8:33regarding applications , the SaaS
8:35business is gradually starting to
8:37appear in Japan as well .
8:40>> However , it brings us back to the
8:41question I raised earlier : what about
8:43the OS and the middleware ?
8:46>> Ultimately , with AI — specifically
8:48foundation models — now being
8:50integrated into workflows as " AI agents
8:53, " or what they call " Agentic AI " in
8:55the US , allowing them to operate
8:57autonomously , applications as
8:59functional tools are becoming less
9:01necessary . Historically , it was said
9:06that SaaS consisted of three elements :
9:09the GUI for the user , the SOR — or
9:11System of Record , which is the data
9:13infrastructure — and finally , the
9:15business logic . Basically , before the
9:20arrival of AI , software was just a
9:23collection of functions . It was a game
9:29of taking structured input and
9:30processing it into structured output .
9:36So , humans had to observe this business
9:38environment , structure that diverse
9:40information , and input it into
9:42databases because the logic couldn't
9:44process it otherwise . That's why we’d
9:48go out of our way to build GUI
9:50interfaces for every single SaaS , using
9:52dropdowns to input text , slowly
9:54structuring data piece by piece so the
9:57logic could read it . But what’s
10:00changed in this AI world is that , for
10:02better or worse , AI can now handle
10:05various data as-is . It can handle text
10:08directly , and images — well , text as
10:10images — are something it’s become
10:12fully capable of handling lately . In
10:15that sense , we’ve reached a point
10:17where all you really need as an
10:19interface is a chat window .
10:21>> And so , ultimately , the so-called ...
10:23>> Well , back in February when Claude
10:25released " Computer Use , " people talked
10:27about it being the end of SaaS , even
10:29calling it the " SaaS Apocalypse . " But
10:32what we realized afterward is that the
10:34companies surviving are the ones that
10:36hold the data , the ones that hold the
10:38actual business logic . The conclusion
10:40now is that companies holding only the
10:42UI are the ones being weeded out .
10:44That’s true .
10:46>> In that sense , the application layer is
10:48being dismantled by AI , and the place
10:50where it’s re-crystallizing is in
10:52middleware — foundational areas like
10:54data infrastructure and authentication ,
10:57along with computational resources .
11:01Capital is increasingly being funneled
11:03into that computing space , so we really
11:05should be supporting it , but when you
11:07think about who those players are in
11:09Japan , nothing really comes to mind .
11:13There are Japanese companies , of course
11:15, but the market is truly small .
11:18>> But in San Francisco , for example , the
11:20companies selected for things like Y
11:22Combinator , the star players are
11:24actually that middleware . Yeah .
11:27>> Including business tools and developer
11:29tools .
11:30>> America is betting heavily on those
11:32areas , but it's a capital-intensive
11:34business .
11:36>> But ultimately , Japan's focus leans
11:38more toward application layers or what
11:40we call system integration , where human
11:43resources are embedded into businesses
11:45as a form of labor — it’s really
11:47skewed toward operations .
11:54>> In that sense , Japan clearly lacks
11:56middle-layer services or what we call
11:58well-managed product-based services , so
12:00there are very few of those in Japan .
12:07>> I feel that personally stepping in to
12:08fill that gap is where I can provide
12:09added value , and that’s what I’m
12:11currently thinking about as I start a
12:12new venture .
12:13>> Yeah .
12:13>> That’s how I decided on it .
12:15>> First , I’d like to ask for a detailed
デジタル赤字は日本特有なのか
12:17explanation of the current digital
12:19deficit , as I’m sure many people
12:21haven’t read your previous report . To
12:24start , it’s been a long time since
12:26people began pointing out the digital
12:28deficit caused by the internet giants
12:30— the GAFAM companies — and it’s not
12:32just Japan that lacks companies like
12:34Apple or Google . In fact , no country
12:36other than the U.S. has them . Yeah . So ,
12:39I’m curious , are there many countries
12:41other than Japan suffering from a
12:43digital deficit , or is this a problem
12:45unique to Japan ? To put it simply , this
12:48is a problem unique to Japan . Now ,
12:51apart from Japan , there are other
12:53countries facing a digital deficit ,
12:55such as Germany , for example .
12:58>> But Japan and Germany are fundamentally
13:00the two countries closest to each other
13:02in terms of the scale of their digital
13:04deficits . The biggest difference in
13:06their industrial structures is that
13:08Germany has companies like Siemens and ,
13:10of course , SAP . Well , it’s not an
13:12exaggeration to say that SAP is
13:14effectively a U.S. company , but the
13:16fact remains that those companies earn
13:18foreign currency , so in terms of
13:20industrial structure , the digital
13:22deficit isn't seen as a major issue
13:24there . Now , looking at the actual macro
13:28figures , in Japan's case , there's what
13:30we call a " digital deficit " — which
13:32I'll discuss later — that refers to a
13:34payment excess in the digital-related
13:36balance of payments , a category within
13:39the services balance of our
13:40international balance of payments
13:42statistics . This basically feeds
13:46directly back into the economic
13:48environment , acting as a fundamental
13:50factor .
13:50>> Yes . You have described this as a "
13:52structural yen depreciation , " right ?
13:54>> That's correct . It does feed directly
13:56back into those areas . However , for
13:58Germany , for better or for worse ,
14:00>> it is a member of the Eurozone , so if
14:02Germany is struggling on its own , it
14:04can be offset by other countries . In
14:06that sense , as written in the report ,
14:08the digital deficit is fundamentally a
14:11>> Japan-specific problem .
14:12>> Right . Now , one thing we must be
14:15careful about here is that , although
14:17some people say this digital deficit
14:19itself is the problem , as I just
14:20mentioned , this is a matter of the
14:22balance of payments . Therefore , having
14:26a large deficit is also a reflection of
14:29a situation where domestic demand
14:31exceeds external demand , so in that
14:33sense , these figures alone cannot be
14:35labeled as one-sidedly bad .
14:38>> As I have noted here , when looking at
14:41economic policy , industrial policy , and
14:43fiscal and monetary policy , there are
14:45trade-offs involved , so the deficit
14:48itself is not necessarily the problem .
14:51However , this digital deficit is indeed
14:54a mirror reflecting the structural
14:56problems Japan faces , particularly in
14:58software .
15:01>> So , the position of this digital
15:02economy report is that we must identify
15:05the issues within that reflection .
15:07>> Right .
15:08>> If we had very high purchasing power
15:10and were buying lots of foreign
15:11products , leading to a deficit , that
15:13might be fine because it would mean the
15:15public is prosperous . But in the case
15:16of Japan's digital deficit , it seems
15:18that is not the case . So , you mentioned
15:20earlier that the nature of the digital
15:23deficit in countries like Germany is
15:24slightly different from Japan's . What
15:27exactly makes Japan different from
15:29other countries ?
15:31>> To understand that , let me first
15:32briefly explain what the digital
15:34deficit numbers actually look like .
15:37>> These numbers are from 2023 , so they
15:39are a bit old , but basically , as I
15:41mentioned earlier , the digital deficit
15:43refers to the situation where payments
15:46in the " digital-related " category
15:47within the service balance are
15:49exceeding earnings .
15:50>> On the left side , we have our country's
15:52macro financial situation used to
15:54calculate the current account balance .
15:56With investment income as our top
15:58revenue source , after various
15:59deductions , the current account balance
16:02comes out to 21.4 trillion yen . This "
16:06digital-related balance " within the
16:08service balance is what we refer to as
16:10the digital deficit , and it is
16:12currently composed of five specific
16:14items . These are management consulting ,
16:18computer services , copyright , and
16:20information and communications ;
16:22basically , these first three items
16:24account for about 97 % of the total . So ,
16:26we can ignore the last two for now .
16:29Regardless , it is crucial to understand
16:31what these three actually represent .
16:36Looking at the name " management
16:38consulting , " one might assume we are
16:40just overusing foreign consulting
16:42services , but in reality ,
16:43digital-related advertising makes up
16:45almost everything under this management
16:47consulting category . In other words ,
16:51this includes things like advertising
16:53payments made to companies such as
16:55Google or Meta .
16:56>> Yeah . Does it also include the ads
16:57playing for the people watching this
16:59video right now ?
16:59>> Oh , absolutely , those types of ads are
17:01included as well . As for computer
17:03services , as the name implies , this
17:05refers to things like SaaS and PaaS .
17:08For B2B , that includes services like
17:10Salesforce or SAP , which I mentioned
17:13earlier , and for B2C , it includes fees
17:15for services like Netflix or Spotify .
17:19>> Now , regarding copyright investment
17:22usage fees , the content is a bit hard
17:24to grasp , but it essentially includes
17:26licenses for things I mentioned earlier
17:29, like Linux , Windows , and iOS , so that
17:31makes up a significant portion . In
17:38short , if we look at these three
17:40categories collectively , both payments
17:42and receipts are growing , but it is
17:44clear that the amount Japan is paying
17:46to foreign companies is quite high ,
17:48which is the direct root cause of this
17:50digital deficit . Yeah .
17:56>> Yes .
17:57>> Number one in investment property
17:58listings . To start real estate
18:00investing , search for the Rakumachi app
18:02or " Rakumachi . "
拡大する「隠れデジタル赤字」
18:04>> Regarding this deficit , do you expect
18:06it to remain at this level , or will it
18:08continue to grow in the future ?
18:10>> Well , for starters , under a pessimistic
18:13model , we know that it will swell to 28
18:15trillion yen by 2035 .
18:18>> Wait , that is only in terms of the
18:20service account balance . As I mentioned
18:23at the beginning , as this
18:25software-defined world continues to
18:27arrive , where do sales for things like
18:29Tesla or BYD show up ? They are
18:31basically included in the trade balance
18:33. Yeah . So , they are included as
18:36products , with the software accounted
18:38for as an accessory . The question is ,
18:42how much does that actually amount to ?
18:46Also , as these software-defined
18:48products sell , to what extent will
18:50existing hardware-defined markets be
18:52destroyed ? We need to consider this
18:57decline in competitiveness , so in this
18:59report , we created two new deficit
19:01categories : " hidden digital deficit "
19:04and " SDX deficit . "
19:06>> When you combine those across the
19:08automotive , industrial machinery , and
19:11telecommunications industries , a
19:13pessimistic scenario projects a total
19:15deficit of approximately 45.3 trillion
19:18yen . As I mentioned earlier , this is
19:21strictly a matter of balance of
19:23payments statistics , so the figures
19:25themselves carry various implications .
19:29One thing I can say is that the 28
19:31trillion yen figure from the
19:32pessimistic model I just mentioned also
19:35accounts for disruption caused by AI .
19:37>> If we look at this , for example , the
19:40digital deficit as of last year was
19:426.85 trillion yen . And it is expected
19:45to grow from there to 28 trillion yen .
19:48However , even with this digital deficit
19:50already factored in , we are currently
19:52seeing a service account deficit of 2.9
19:55trillion yen . So , where is the income
19:58coming from to offset these payments ?
20:00It’s coming from inbound tourism and
20:02industrial property rights .
20:04>> Industrial property rights , for
20:06instance , consist of licensing fees for
20:08inventions and patents in the
20:10manufacturing industry , but we can't
20:12expect these inventions to triple
20:14overnight , can we ? This clearly depends
20:17on seasonality and the industry itself ,
20:19so it will likely remain relatively
20:21flat with some fluctuations . As for
20:23inbound tourism , if we consider whether
20:26the number of foreign visitors to Japan
20:28can triple from here , or if revenue can
20:30grow threefold , there are clearly
20:32limits . Therefore , when the
20:34digital-related balance grows from 6.85
20:37trillion to 28 trillion , that remaining
20:3922 trillion or so will be added
20:41directly to the service account deficit
20:44, meaning the service account deficit
20:46could reach around 22.9 trillion yen .
20:51Regarding the trade balance , as
20:52mentioned earlier , if the deficit
20:54continues to grow , this confirms the
20:56structural issue behind the yen's
20:58depreciation that Kama-san is talking
21:00about — the fundamental problem lies
21:03primarily within the digital industry .
21:09While there is certainly the impact of
21:11interest rate gaps between Japan and
21:12the U.S. in the financial sphere , when
21:14looking at the fundamentals , it's
21:16possible that this digital-related
21:18balance is the most significant factor .
21:25In that sense , when analyzing Japan's
21:27yen-depreciation structure , it becomes
21:29clear from a national perspective that
21:31we must address the fact that our
21:33industries aren't growing — whether
21:35that's due to a lack of alternatives or
21:37losing out to foreign competitiveness
21:39— and we need to intervene properly .
21:43>> Yeah . In addition to the existing
21:46digital deficit , Japan’s core
21:48business model of exporting automobiles
21:50is now being threatened by SDX .
21:53Furthermore , since many of the current
21:55ways Japan earns foreign currency don't
21:58scale well , the balance of payments
22:00will become increasingly unstable ,
22:02which is quite concerning . Yeah .
22:04However , if left unchecked , this
赤字が拡大する業界構造
22:06digital deficit is expected to increase
22:08by about four times in a pessimistic
22:10scenario . I'd like to ask in more
22:12detail what exactly causes such a surge
22:14. You mentioned disruption by AI as an
22:17example ; could you explain what that
22:19entails specifically ?
22:20>> First off , this figure of 28 trillion
22:22yen is based on a base scenario of 18
22:25trillion yen , so we estimate that the
22:2710 trillion yen delta is coming from AI
22:30.
22:31>> And the area that is bleeding the most
22:33is Japan's SI sector , which is being
22:35significantly eroded , according to the
22:37results . First , when considering the
22:42root cause of Japan's structural
22:43problems with digital deficits , there
22:46are obviously many factors , but they
22:48can be broadly divided into two : the
22:50market issue and the business model
22:52issue . Starting with the business model
22:56issue , let's look at the current state
22:58of Japanese business . This map of
23:02Japan's current business structure
23:04might be hard to read , but the way to
23:06interpret it is that the gray areas
23:08represent earnings for Japanese
23:10companies from overseas , and the red
23:12areas represent payments to overseas .
23:16The larger the area , the larger the
23:18market . At a glance , you can see that
23:21the section labeled " SI " is the largest
23:24. This is a so-called labor-intensive
23:26industry , and it uses a calculation
23:28method known as " man-month pricing , "
23:30which is based on how many people were
23:32assigned for how many hours . Therefore ,
23:34it goes without saying that both its
23:36growth rate and profit margin are low .
23:38However , in a sense , it's just like
23:40contract AI development — you can
23:42steadily generate cash in the short
23:44term . While it is a reliable way to
23:47make money in that sense , you
23:49ultimately have to keep catering to the
23:51client , which means you lose the time
23:53needed to build actual products . This
23:55holds true for both small AI startups
23:57and large SI vendors ; if you keep
23:59earning money this way , you end up
24:01doing nothing more than serving as a
24:03client's errand runner . Yeah .
24:05>> Conversely , looking at where overseas
24:07companies are making money , as shown in
24:09the middle with applications ,
24:10middleware , computing infrastructure ,
24:12and advertising , it is clearly these
24:14capital-intensive businesses . Once
24:17software is built , they generate
24:19revenue through managed services ,
24:21operational maintenance , transaction
24:23fees , or marketplace commissions . These
24:27are the ways overseas companies are
24:29clearly making money , and ultimately ,
24:31the root of Japan's structural digital
24:33deficit lies in the difference in
24:35capital efficiency . So , you have
24:37businesses that won't scale unless you
24:39keep adding more people — essentially
24:41linear growth — versus businesses that
24:43scale rapidly just by increasing
24:45electricity usage . These are what we
24:47call exponential businesses . The gap
24:49between these two is what primarily
24:51appears in the digital trade deficit
24:53structure , and as I mentioned , since AI
24:55is clearly going to keep cutting into
24:57SI service fees , that value is shifting
25:00toward the middleware providers I spoke
25:02of earlier . And it’s also shifting
25:04toward the computing infrastructure
25:05side . When you think about it , it
25:07becomes clear that our digital deficit
25:09is bound to grow ; that’s what we call
25:11a problem on the business structure
25:13side . It means that even in the digital
25:15sector , Japanese companies have a
25:17tendency to choose labor-intensive
25:19business models .
FDEはマーケティング!?
25:20>> By the way , " FDE " has recently become
25:22something of a buzzword ,
25:24>> and while I think it’s a global trend
25:26, I feel like there are more and more
25:28companies in Japan claiming that they
25:30are doing FDE themselves . Has this led
25:33to any actual changes in how people
25:35work ? Well , to put it simply , for most
25:38Japanese companies , FDE is just a
25:40marketing term .
25:41>> It is . Now , as for why Palantir
25:45originally started doing FDE , they
25:47describe it as moving from a gravel
25:49road to a paved highway , and they do
25:51this because they have an ontology
25:53middleware product .
25:56>> What is an ontology ?
25:57>> To put it simply , it’s a mechanism to
26:00represent the business context that AI
26:02uses , technically through a graph —
26:04that's the simplest way to explain it .
26:07>> So , they possess this software product ,
26:09this middleware product . They work
26:15closely with their clients , acting
26:17almost like an on-site CTO to observe
26:20what's happening within the company and
26:22build a digital twin that allows AI to
26:24be utilized — that is the Palantir
26:26business model , and as they go from
26:28client to client , they start to
26:30identify common components . Things like
26:35, " Hey , this and that are basically the
26:37same . " And feeding this back into the
26:40product — that R & D function is the true
26:43purpose of an FDE . Yeah .
26:46>> So , when companies like Anthropic or
26:48OpenAI talk about FD , they have their
26:51own foundational AI models . They embed
26:55themselves with customers to understand
26:57how they should tune these models
26:59during the pre-training phase or how to
27:01customize them for the client . When
27:05they identify common components , they
27:07feed that back into their product . That
27:11loop is what FD is supposed to be .
27:13However , in many cases , what systems
27:16integrators call FD is just extracting
27:18the superficial aspects of customer
27:21requirements . Having a product is
27:24exactly what gives FD operations their
27:26true depth .
27:27>> I really think that looking at this
27:29business structure , it's essential not
27:31to misunderstand this point and to see
27:33it for what it really is .
27:34>> Yeah .
27:35>> So , simply being on the front lines
27:37isn't enough ; the part about using the
27:39data received to add value to future
27:41products is missing , right ? Even if
シリコンバレーと勝負せよ
27:43it's hard to become a platformer like
27:45them , there are still opportunities in
27:47the middle layer , as you mentioned ,
27:49Tsuda-san , aren't there ? Why hasn't
27:51Japan entered this space until now ?
27:53What are your thoughts on this ?
27:55>> Well , I think there are two reasons .
27:57The first is exactly what I alluded to
27:59earlier when I mentioned there was
28:01another reason — it's a market issue .
28:03Yeah .
28:04>> Yes .
28:05>> Basically , this software business —
28:07including the middle layer — is a
28:09winner-take-all market . It’s a market
28:13where whoever has the largest market
28:15and the most capital takes everything
28:16through economies of scale . Software is
28:20a business that fundamentally has no
28:22borders , and recently , thanks in part
28:25to AI , we’ve reached a point where we
28:27can overcome language barriers , customs
28:29, and localization issues at a
28:31relatively low cost . This means that
28:34tools developed in the U.S. are coming
28:36straight to Japan in their original
28:38form , creating a situation where
28:39borders effectively don't exist . We
28:42have to do business in this environment
28:44, but traditionally , Japan has had a
28:46market that is just large enough to be
28:48distracting .
28:48>> Because of that , there are many cases
28:50where companies just focus on the
28:52domestic market . For better or worse ,
28:55in the Japanese market , if you go
28:57public , you can generate a decent
28:59income as an entrepreneur , and even for
29:01VCs , you can provide returns to your
29:03LPs by consistently hitting singles
29:05within Japan without needing to go
29:07abroad . This isn't just an issue for
29:11entrepreneurs ; historically , it’s
29:13been a situation where you only needed
29:15to operate within Japan . Furthermore ,
29:18in the past , companies were protected
29:20by business customs and regulations ,
29:23which often led Japanese software to
29:25become Galapagos-like , specialized
29:27solely for Japan . However , whether it's
29:31applications or middleware , as I
29:33mentioned earlier , this is a business
29:35that must scale and targets the largest
29:37possible markets . When you think about
29:41where that is , it's clearly the United
29:43States — San Francisco , or rather ,
29:45Silicon Valley . So , if you really want
29:51to win in software , the premise for
29:53scaling is clearly to go straight to
29:55San Francisco , find product-market fit
29:57there , leverage the massive amount of
29:59capital available , and then expand
30:01across the country and the world . In
30:06fact , looking at the numbers , there is
30:08an 8.5 - fold difference in ICT spending
30:11between Japan and the U.S. alone .
30:13Therefore , even if you’re using the
30:16same product with the same features ,
30:18companies in San Francisco are starting
30:20with an 8.5 - fold head start . Both in
30:22terms of credibility and capital .
30:26>> In terms of the number of product
30:28customers as well , the fact that the
30:30environment was confined to the
30:32Japanese market is one of the
30:34structural problems in Japan ;
30:35particularly for middleware , the
30:37customers are applications . As I
30:43mentioned , it’s a three-tier
30:45structure , and middleware sits below
30:47the applications . Their direct
30:51customers are the applications , to
30:53which they provide , for example , APIs
30:55or integration features . Even with
31:01current AI — for instance , Salesforce
31:04recently appeared at their event with
31:06Dario from Anthropic — the reason they
31:08say it’s not an AI replacement is
31:10because they believe if deterministic
31:13functions for a CRM are already defined
31:15, AI should just use them via tool
31:17calls .
31:21>> Whether it's through APIs or MCPs ,
31:23companies that hold the business logic
31:25and the CRM won't be replaced . In that
31:29sense , Salesforce is gradually moving
31:31down from the application layer into
31:33the middleware layer . If you think
31:36about where these applications are most
31:38concentrated , it’s San Francisco . So ,
31:43in that sense , if you don't start in
31:45the U.S. from the beginning , middleware
31:47won't have customers or room to grow .
31:53Plus , because middleware deals with
31:55highly technical areas , VCs often can't
31:57easily perform due diligence . I believe
32:03that is why these companies don't
32:05emerge from Japan — or really anywhere
32:07outside of the ecosystem in San
32:09Francisco .
32:11>> Having heard all this , I understand
日本に勝ち筋はある？
32:13very well why you , Tsuda-san , are
32:14starting a business and what your
32:16concerns are , but listening to you , it
32:18seems that starting a company in Japan
32:19is overwhelmingly at a disadvantage .
32:21>> Yeah .
32:22>> That’s exactly why there haven't been
32:23many before , and ultimately , even if
32:25you start a business in Japan , the
32:26competitors you're fighting are in
32:28Silicon Valley .
32:29>> And there are so many elite companies
32:30there , aren't there ? With that in mind ,
32:32I'd like to ask why you specifically
32:34chose to start your company in Japan .
32:36>> Yes . Well , as I mentioned at the start ,
32:39there is a symbolic aspect to it — the
32:41idea that I am going to lead the charge
32:43on the digital economy report — but
32:45there are various other reasons , too .
32:48At the end of the day , interest rates
32:50in Japan are relatively low , and
32:52compared to San Francisco , you can
32:54probably hire top-tier engineers at a
32:56salary that's more than ten times
32:58cheaper . In that sense , Japan has a
33:02certain advantage when it comes to
33:04Japan-US arbitrage and capitalizing on
33:07those gaps . And when it comes to simply
33:10basing your headquarters in Japan ,
33:11there are actually a lot of different
33:13ways to go about it . Places like South
33:18Korea , the Nordic countries , Israel ,
33:20and the UK have small domestic markets ,
33:22so the " born global " mindset — starting
33:25without borders and going to the global
33:27market from day one — is deeply
33:29ingrained there . For both entrepreneurs
33:34and investors , I think there is so much
33:37we can learn from them , especially
33:38South Korea . For example , there is a
33:43company called Coupang , often referred
33:45to as the Amazon of South Korea , which
33:47is headquartered in Korea . But their
33:52registration is actually in Delaware ,
33:54in the US , where they hold a parent
33:55company with the Korean headquarters
33:57underneath it . In other words , they
34:02keep the holding company — or rather ,
34:04the investment interface — in San
34:06Francisco , while the actual capital
34:09flows into Korea to fuel global
34:10expansion . China does something similar
34:16with companies like Alibaba or
34:17ByteDance , though that is partly due to
34:19the unique reason that offshore
34:21investment isn't allowed within China .
34:25>> There are various ways to structure it ,
34:27like raising capital in Singapore or
34:29the US and then channeling it back into
34:31the domestic company , so " basing your
34:33headquarters in Japan " can actually
34:35mean many different things .
34:39>> While we talk about the digital trade
34:41deficit , as the report notes , the
34:43domestic remittance ratio is basically
34:45the deciding factor . So , in terms of
34:51schemes , one way is for a Japanese
34:53company to earn foreign currency in the
34:56US while keeping the IP in Japan , and
34:58perhaps handle fundraising in Delaware .
35:04Another way is to keep the legal
35:06registration and headquarters entirely
35:08in Japan . The fact that companies
35:14aiming to be " born global " are actually
35:16using these various schemes is not
35:18widely known , and I really feel that
35:20Japan needs to learn from these
35:22approaches . Yeah .
35:26>> Have there not been many Japanese
35:28companies doing things this way until
35:29now ?
35:30>> That's right . Not really . The hurdle
35:34for raising funds in the US is
35:36extremely high , to begin with .
35:38>> Yeah . Right , exactly . You have to
35:41really get into the inner community and
35:43be able to negotiate aggressively in
35:44English . From their perspective , they
35:48also wonder why you’re bothering to
35:49come all the way from Japan . For
35:51example , the UK is strong in finance ,
35:53so they can combine that with tech for
35:55FinTech . In the Nordic countries where
35:57I've been , they have strong gaming and
35:59education sectors , so they do EdTech or
36:01GameTech . Israel is strong in military
36:03and cybersecurity , so they have
36:05DefenseTech . They can build a narrative
36:07around cybersecurity and such . In
36:10Japan’s case , while companies like
36:12Caddi have emerged recently , there have
36:15been very few people doing things like
36:17manufacturing multiplied by SaaS .
36:21Because of that , even in San Francisco ,
36:23they’d ask , " Why come here when you
36:25have a large market like Japan ? " and it
36:29was hard to answer the " why " question
36:31— that is , why you are worthy of
36:33investment — which I think was a major
36:35factor .
36:37>> I see . Well , even at Y Combinator ,
36:39middleware is a very popular and highly
36:42competitive field . Could you tell us ,
データの価値が高まる時代に
36:44within the scope of what you can share ,
36:46specifically what kind of services or
36:48products you intend to build ?
36:51>> Yes . What we intend to do at my company
36:55is , as I mentioned earlier , current AI
36:57— especially in LLM , or Large Language
36:59Model architecture — is pre-trained by
37:02extensively crawling internet text data
37:04. Recently , that hasn't been enough ,
37:09and there’s been news about companies
37:11buying up massive amounts of books from
37:13used bookstores to scan them . Either
37:16way , they are learning based on what
37:18exists publicly in the world . So , what
37:23happens next is the question of whether
37:25it’s okay to train AI on a company's
37:28domain-specific business context ,
37:30business logic , rules , and data . Yes .
37:35>> As I said earlier , for a company today ,
37:37the only true competitive advantage is
37:40their data and their context . If you
37:43give that away , then eventually ...
37:45>> Software itself has become a commodity .
37:47Hardware , too , while important in its
37:49own right , is effectively a commodity
37:51as well . Considering this , as
37:53competitive advantages continue to
37:55diminish , the question becomes how to
37:56manage your unique domain context while
37:58maintaining governance over the AI .
38:00>> Mm-hmm . The market is gradually
38:02emerging for how to manage it in a
38:04decentralized way . Even data
38:06infrastructure companies like Snowflake
38:08or Databricks , which I mentioned
38:10earlier , have long histories , but as
38:12you delve deeper ...
38:13>> Well , Databricks started with Apache
38:15Spark , so they've been at it for a very
38:17long time , going all the way back to
38:19the AMPLab era at UC Berkeley . Before
38:22that , there was a company called Immuta
38:24for data management , and now they've
38:26reached " hectocorn " status , with a
38:28market cap of around 15 trillion yen .
38:31The market for data management ,
38:32implementing guardrails , and managing
38:34context is growing rapidly , and
38:36that’s where I want to focus my
38:38efforts . Having served as a chief
38:40architect for distributed data
38:42management for the government , and
38:44knowing there are many more scalable
38:46technical approaches to areas like
38:48Palantir’s ontology , I see a
38:50significant opportunity . So , I am
38:53planning to build a middleware platform
38:55for this field , which is what we call
38:57context management . Speaking of
39:00competitive advantages in data ,
39:02Palantir is quite well-known in Japan .
39:05Alex Karp has been sounding the alarm
39:07— though it may be a bit of
39:09position-taking — that by using closed
39:11models like those from Anthropic or
39:13OpenAI , your intellectual property will
39:16be stolen by them , and ... ...
39:18>> your business will end up being copied .
39:20It’s unclear if that’s actually
39:22happening , but he is certainly raising
39:24those concerns . What are your thoughts
39:26on that ? Could you please share them ?
39:28>> Yes . As a premise ,
39:29>> Yes . They’re what you call " tech bros
39:33" in San Francisco , but it’s actually
39:35really worth listening to what they
39:37have to say . For example , recently ,
39:42people at Frontier Labs started talking
39:44about " P-star , " and Dario began
39:46claiming that frontier development has
39:48gone too far and that we all need to
39:50stop it . Elon and Sam from OpenAI
39:55seemed to agree with that , but what
39:57happened afterward ? They just went
40:00ahead and released Opus 3.5 . First and
40:04foremost , we have to recognize that on
40:06the surface , a lot of what they say is
40:08just positioning talk . I think that
40:13position-driven rhetoric shows up
40:14really clearly in this conflict between
40:17open-weight and proprietary models .
40:21After the whole Open / Hugging Face
40:23incident , the " Coalition for Secure AI "
40:25was established , and it’s really
40:27split things down the middle . In other
40:31words , companies like Anthropic , OpenAI
40:33, and Google DeepMind aren't joining it
40:36.
40:36>> But on the flip side , companies like
40:38Palantir , Amazon , Nvidia , and
40:40Databricks are saying that we need to
40:43be careful . At the end of the day ,
40:45that’s just a reflection of their own
40:47positioning . For instance , the more
40:50models are running inference , the more
40:53AWS or Microsoft’s data centers make ,
40:55which is huge profit . That’s their
40:58business , and the more that workload
41:00increases , the more Jensen Huang makes
41:03from GPUs . Conversely , the more data
41:05management there is , the more
41:07Databricks profits . In that sense ,
41:09Palantir is also doing context
41:11management in a business sense , so they
41:13can switch models however they like .
41:17Everyone sounds like they’re talking
41:19about grand causes , but it’s
41:20important to remember that about half
41:22of it is just positioning talk .
41:24>> Yeah . Looking at how the camps are
41:26divided , it’s pretty clear which side
41:28they stand on based on where their
41:30business interests align . Mm .
41:32>> Yeah . That’s true .
41:33>> Did you know that condo prices have
41:35increased 1.4 times over the last five
41:37years ? Your condo might have gone up in
41:40value too . Sales appraisal requests are
41:43increasing right now . Search for "
41:44Rakumachi Satei . " The talk about
人類滅亡リスクはシリアスか
41:47humanity potentially going extinct ,
41:49mentioned by a former Anthropic
41:51researcher , is a bit separate from the
41:53business context , but I think it’s
41:55actually a concern to some extent . How
41:57seriously do researchers on the ground
41:59actually
42:01>> take these concerns ?
42:03>> Well , I think the people on the core
42:05teams actually do feel that way . I
42:08often talk to people at DeepMind and
42:10other frontier labs , and in San
42:12Francisco , people don't even say the
42:14word " AI " anymore — if you say it ,
42:16you’ll basically get laughed at . But
42:21people are starting to believe that ,
42:23yes , you actually can run that
42:25recursive self-improvement loop where
42:27it learns by improving itself . That’s
42:34half true ; while it’s technically
42:35possible , there are issues like data
42:37center power shortages due to a lack of
42:40turbines , and the fundamental problem
42:42of not having enough data . So , it’s a
42:49situation where there’s no conclusion
42:51yet , but everyone recognizes that these
42:53frontier labs are running at a pace
42:55that even they find alarming . But this
43:00is really what you would call a
43:02prisoner’s dilemma in game theory .
43:05>> And who is involved in this
43:06prisoner’s dilemma ? It’s with
43:09investors , with competitors , and
43:11finally , there's the security dilemma
43:13between the US and China . They are in a
43:17dilemma in each of these relationships .
43:21At the very least , investors — not just
43:23the VCs , but also the LPs behind them
43:26— are thinking that to justify these
43:28valuations , these companies have to do
43:30something that essentially disrupts
43:32every market .
43:34>> Right .
43:35>> They really think they can't justify
43:37them otherwise . Investors primarily
43:40expect that they can't protect their IP
43:42, and as I mentioned , the competition
43:44is so fierce that even if someone says
43:46they want to quit , nobody believes it ,
43:49so they can't stop . Even if we could
43:53hypothetically stop within the US , the
43:55question of whether China would stop is
43:58a different story , and they clearly
43:59wouldn't . Everyone is running at full
44:04speed , knowing they are destined to do
44:06so , regardless of whether they’re
44:08heading toward ruin . Because of this ,
44:13the core teams are exhausted , and we're
44:15seeing more and more people leaving
44:17lately , as well as a growing number of
44:19people who are genuinely concerned .
44:22>> I see . Well , it might be that they are
44:23leaning toward catastrophic thinking
44:25precisely because they are exhausted .
44:26>> That could be true . This is truly a
44:28mental health issue , and frankly , even
44:30San Francisco itself is , well , I don't
44:32know . It’s a barren land ; there
44:34really is nothing there . It’s like
44:37they're keeping the brightest minds
44:39isolated in a boot camp-style
44:40environment in that desolate place , and
44:42if they keep doing that for one or two
44:44years , you can really feel that it's
44:46taking a toll on people when you're on
44:48the ground there .
44:50>> Yeah . They were pushing themselves hard
44:53, thinking their mission would be
44:55complete once AGI arrived , but nobody
44:57really knows how things will turn out .
44:59>> Yeah . Well , nobody really knows . Even
45:02from an investment perspective , they
45:04are diversifying their portfolios .
45:05That’s how uncertain it is , and I
45:07think even they themselves don't know
45:09if that's truly the case . If
45:12electricity becomes the ultimate
45:14deciding factor , some say China might
45:16have the upper hand due to their
45:17geography ; what is the discussion like
45:20regarding that ?
45:23>> I’m not an expert on power , so there
45:25are parts I don’t fully understand ,
45:27but as YC and other VCs often say about
45:29" bits and atoms , " in the end , those who
45:31control the power source and the
45:33communication infrastructure — meaning
45:36not just electricity , but the fiber
45:38optics and undersea cables for data
45:40centers — will be the ones who hold the
45:42power . Everyone recognizes this as a
45:46problem , and in that sense , we’re
45:48seeing things like gas turbines being
45:50ramped up , or even backorders for
45:52turbines because there’s simply not
45:54enough electricity .
45:57>> In a way , that’s actually good news
45:59for Japanese heavy industry , as they
46:01stand to profit from the situation .
46:05However , that alone won't be enough ,
46:07and if we look at the recent Hugging
46:09Face incident from a broader , more
46:11detached perspective , we have to
46:13consider cybersecurity . If all
46:18firewalls could be breached , it’s not
46:20just data center power that’s at risk
46:22; theoretically , the power we consume
46:24in our daily lives could be hacked and
46:26virtualized . It’s two sides of the
46:31same coin with the debate on guardrails
46:33, so I don’t think there’s a simple
46:35conclusion to be drawn . I see . Well ,
46:39while these doomsday scenarios might
46:40not be entirely impossible , it’s
46:42practically impossible to base our
46:44daily actions on them , so the reality
46:46is that we just have to keep moving
46:48forward with our business as usual
46:49under these conditions .
46:51>> Yeah .
46:51>> By the way , Tsuda-san , for yourself
46:53personally — whether you call it ASI or
46:55something else — if a superintelligence
46:58were to arrive , would you be excited or
47:00a bit afraid ?
47:01>> Hmm . Well , first of all , there is no
47:03set definition for superintelligence ,
47:05so it’s quite a moving target , isn't
47:07it ? The goalposts keep shifting , so I
47:10can’t say for sure , but I do think
47:12the singularity will eventually arrive
47:14at some point .
47:16>> I believe so . And so , regarding that ,
47:19how we as humanity can accept it is
47:21important . Of course , governance
47:26discussions are truly vital , including
47:28whether humans can properly pull the
47:30kill switch .
47:32>> We really need to treat this as a
47:34genuine , practical issue . Not as
47:36something that might come in the
47:38distant future , like ten years from now
47:40, but as something that could arrive
47:42tomorrow . With that mental preparation ,
47:44I think it’s just about considering
47:47what I can do now — nothing more ,
47:49nothing less .
47:49>> Yeah .
47:50>> I see . You mentioned that the power
データ主権と国際協調
47:53issue is one thing , and the
47:54availability of high-quality data is
47:56another key to AI progress . Could you
47:59elaborate further on this matter of
48:01data sovereignty ?
48:02>> Yes . Well , I’d like to make a slight
48:06correction ; I don't actually use the
48:08term " data sovereignty " all that much .
48:10First of all , " sovereignty " is a term
48:13I’m familiar with since my background
48:15is in the interdisciplinary area
48:17between philosophy , political science ,
48:19and sociology . In fact , this word is a
48:21political science term . It originated
48:24in the 16th century with a man named
48:26Jean Bodin . From its classical
48:29definition , one of the most essential
48:31elements included in the concept of
48:33sovereignty is the notion of external
48:35recognition . To explain that , when
48:41people talk about sovereignty , many
48:43think only internally — like
48:44territorial integrity , protecting one's
48:47own borders , and exercising rights
48:49within them . However , sovereignty is
48:54actually a concept that is established
48:56relatively . In an anarchic
49:00international political society , the
49:02term " sovereignty " also includes a very
49:05large portion regarding being
49:06recognized as independent by other
49:09countries . Sovereignty in the sense of
49:13protecting a nation from external
49:15threats is called Westphalian
49:17sovereignty , which stems from the state
49:19system established after the Peace of
49:21Westphalia . While this concept is
49:25certainly important , what's missing
49:27from current discussions on sovereignty
49:30— particularly regarding digital
49:32sovereignty — is the idea of "
49:33international legal sovereignty , " which
49:35involves external recognition , and "
49:37interdependence sovereignty , " which is
49:40about exercising control within
49:41interdependent relationships . I bring
49:45this up because technology does not
49:47emerge in a vacuum . Just like academic
49:50papers , technology is always built upon
49:52prior research . It emerges as a
49:55progression from what came before ; it
49:57doesn't just suddenly appear out of
49:59nowhere , completely unrelated to
50:01everything else . Given this , as I wrote
50:04in my own notes on SIS-Notations ,
50:06technological sovereignty is not
50:08isolationism . It exists relatively
50:12within the broader context of technical
50:14relationships , building upon the
50:16foundational work of respectable
50:17frontier labs and researchers like
50:19those at UC Berkeley I mentioned
50:21earlier . Therefore , I don't think we
50:26should frame the discussion in an
50:28isolationist way — saying , for example ,
50:30" this is our technology , so we won't
50:32use anything from abroad , " or simply
50:34reinventing the wheel on our own . Of
50:39course , as someone involved in the
50:40Digital Garden project , I understand
50:43the need to create alternatives to
50:45overseas services . However , if we try
50:48to protect data sovereignty solely from
50:51the perspective of Westphalian
50:52sovereignty , we will only end up
50:54isolated and decoupled . In that sense ,
50:58I believe we must continue to face the
51:01frontier — which is to say , keep
51:02engaging with San Francisco — even if
51:05it is difficult to do so . Whether as a
51:08nation or as entrepreneurs , we must
51:10ensure that in the name of
51:12technological sovereignty , we don't
51:14fall into isolationism . I believe we
51:17need to firmly rethink how we approach
51:20this digital deficit and our software
51:22businesses , and I say this as a
51:23reminder to myself as well .
51:26>> I see . Whether it's from the
51:28perspective of protecting valuable
51:30corporate and government data , or
51:32improving the structural digital
51:33deficit , we must not close ourselves
51:35off in either respect . So you mean we
51:38must not become isolationist . That's
51:40right . These days , there are so many
奈良時代スピリッツで世界へ
51:42people who compare the current need for
51:44reform in Japan to the Bakumatsu or
51:46Meiji Restoration eras .
51:48>> I think that's true to an extent , and
51:50many business leaders are talking about
51:52the " Sampo-yoshi " philosophy of Omi
51:54merchants lately . But considering that
51:57Japan is no longer the frontier , I feel
51:59we need to look even further back — to
52:02the era of the Kentoshi envoys to Tang
52:04China .
52:06>> In the Nara and Heian periods , China
52:08was overwhelmingly the frontier ,
52:10bringing advanced culture , laws , and
52:13systems . We sent people there , and
52:17people came from there . Back then ,
52:20Emperor Shomu used Buddhism — a new
52:23religion — to consolidate national
52:25power . In that sense , since the
52:30frontier is clearly in San Francisco
52:32today , we must send our own " Kentoshi "
52:35to San Francisco , go there properly ,
52:37and ensure that what we learn is fed
52:39back into Japan . We cannot let that
52:44relationship wither away .
52:46>> I see . So , it's about the importance of
52:48being humble , communicating , and
52:49maintaining a learning mindset .
52:51>> Yes . I would like to ask one thing : in
52:55this situation where Japan has fallen
52:57behind — and the term " digital defeat "
52:59is often heard — is there any area
53:01where Japan can maintain a competitive
53:04advantage , what you might call " Japan
53:06Leverage " ?
53:07>> Well , I think there are various areas
53:09for that . In terms of soft power , there
53:11really aren't many countries like Japan
53:14where social capital has been built up
53:16to such a high degree in every corner .
53:18In that sense , we should use this soft
53:20power to get frontier-level people to
53:23come to Japan . Many companies ,
53:25including those in the hardware sector ,
53:27are actually looking to get a foothold
53:29in Japan right now . I know many people
53:32say things like Japan will become the
53:34Switzerland or the Singapore of Asia ,
53:35but I think ...
53:36>> When I go to San Francisco , I feel that
53:38when I say I'm Japanese , people are
53:39relatively willing to listen to what I
53:41have to say . Yes , that's exactly it . I
53:43think that's one factor . They always
53:45ask things like , " How is Japan's
53:47hardware industry doing ? " or " What's
53:49the energy demand like ? " or " Is there
53:51any good software coming out ? " The fact
53:54that they're interested is a good sign .
53:56So , in that sense , we should leverage
53:58that to win in Japan . People often talk
54:00about the content or manufacturing
54:02industries , but there are new markets
54:04like the middle-layer software I work
54:06in , so I think Japanese people should
54:08properly ...
54:09>> Go to San Francisco and , as I wrote on
54:11Note , report on the technology that's
54:14actually here in the country . I think
54:16that's very important . As a means to
54:18that end , as I mentioned earlier , the
54:20relative cost of engineers in the US is
54:22so high that you can't even hire top
54:24talent from frontier labs for 50
54:26million yen per person anymore . So , in
54:29that sense , if you hire Japanese people
54:31— especially those engineers who are
54:33currently undervalued — for that same
54:3550 million yen , you can attract top
54:37talent . I think the solid approach is
54:40to identify where the arbitrage points
54:42between Japan and the US lie and make
54:44use of them .
54:45>> That’s true , isn't it ? Since the
54:46market has been large enough to sustain
54:48itself domestically until now , and
54:50there aren't many people doing this ,
54:52there are businesses with huge growth
54:54potential depending on how you look at
54:56them . That was the story of how you are
54:58now taking on the challenge . Well , I
55:00think it’s about time to wrap up .
55:03It’s quite rare for us to have
想像を超えた時代へ
55:05someone in their 20s on our show , even
55:06though we are a business media outlet .
55:10After watching this video , I’d love
55:12for you to share a message — whether
55:14it’s for your own generation or even
55:16older — about what you want to do .
55:18>> I was born in 1997 and I’m 29 this
55:21year . Thinking that I’ll live for
55:26about another 50 years , I look back 50
55:29years from 1997 , and that’s right
55:31when the first commercial computers
55:33were being implemented . Back around
55:381955 , nobody could have imagined that
55:40we’d reach a point where AI operates
55:43autonomously like this .
55:45>> No one could have pictured a world like
55:47" Ghost in the Shell , " so surely , no one
55:49can imagine what will happen in the
55:51next 50 years . We’ll likely reach the
55:55limit of what humans can imagine , and
55:57then a new era will follow . As the next
56:04generation at this turning point , I
56:06believe we must take responsibility for
56:08shaping and designing this country of
56:10Japan and the world itself . Regarding
56:17the technical report I mentioned
56:19earlier , this is a term included in
56:21what Kiichiro Toyoda compiled after his
56:24father , Sakichi , founded Toyota , based
56:26on the Toyota precepts . In the past ,
56:31this was called " industrial patriotism ,
56:33" which later had a dark history of
56:35being tied to the state . However , its
56:40true meaning lies in the story of how
56:43Sakichi created the Type G automatic
56:45loom — a technology recognized even in
56:47the UK , the heart of the Industrial
56:49Revolution — and then handed over the
56:51licenses he gained to his son and the
56:54next generation . This allowed them to
56:59build a new automotive industry by
57:01developing the Model T-style car he had
57:03seen in New York . This story shows that
57:08we must think not just about industry ,
57:10but also about what we can leave behind
57:13for the next generation in terms of
57:14technology . It’s not just about
57:20earning and making a profit ; it’s
57:22about what we can leave for AI natives
57:24who might find their entry-level jobs
57:26replaced by AI . If we don't think
57:30seriously about this , even at our age ,
57:33I believe there won't be a future for
57:35Japanese or global society . In that
57:40sense , we are currently in a funding
57:42phase and looking for funding members ,
57:44but as members of the same generation ,
57:46I hope we can work hard together , so
57:48let's enjoy this era . Thank you very
57:53much . Once again , today we have been
57:55speaking with Mr. Tsuda , a former METI
57:57bureaucrat and current software
57:59engineer . I look forward to your future
58:01success as an entrepreneur , and I would
58:03be happy to have you on the show again
58:05if the opportunity arises . Thank you
58:07again for joining us today . Thank you
58:09very much . Rakumachi , Rakumachi ,
58:12Rakumachi .
58:21>> Fudosan Toshi wa Rakumachi .