【PRI】Behind the U.S.-China “Freeze”: Harvard Rethinks the AI Race — Is There Really a Country That Will Win?
English text appears at the bottom of this post.
【PRI】米中「現状凍結」の陰で――ハーバードが問い直すAI競争、「勝つ国」は本当にあるのか
在米ジャーナリスト 髙濱 賛 (Tato Takahama)
Pacific Research Institute
2026年9月25日
「米国と中国、AI競争で勝つのはどちらか。それとも、誰も勝たないのか」
こんな問いを投げかけたのが、ハーバード大学ケネディ・スクールの経済学者ジェイソン・ファーマン氏である。
ハーバード大学、と聞くと、日本人はどうも少し姿勢を正してしまう（笑）。だが、今回のファーマン氏の話は、難しい経済理論ではない。むしろ、いま世界中が当然のように使っている「米中AI競争」という言葉そのものを、いったん疑ってみようという話である。
9月24日、ワシントンでトランプ米大統領と中国の習近平国家主席が会談した。貿易、関税、レアアース、台湾など、米中間にはいくつもの懸案がある。その中にAIもあった。
今回の首脳会談で、米中は大きな対立をいったん抑え込む方向を確認した。貿易休戦については延長が合意されたが、AIをめぐる核心的な技術問題まで決着したわけではない。むしろ、AIについては、競争を続けながら事故や悪用などの共通リスクについて対話する、という非常に難しい領域が残った。Reutersも、首脳会談には大きな突破口がなかったと報じている。
私はここに、今回の米中首脳会談を理解するうえでの重要なポイントがあると思う。
米中関係は「決着」したのではない。「現状凍結」したのである。
そして、その凍結の陰で、むしろAI問題の複雑さが浮かび上がってきた。
9月24日付のハーバード・ガゼットで、ファーマン氏は「AI競争にはゴールがない」と語っている。
米国と中国は現在、最先端AIモデルの二大勢力だという。しかし、最先端モデルを作った国が、そのまま世界のAI経済を支配するとは限らない。
ここが重要だ。
これまでのインターネット産業では、Googleの検索やMetaのSNSのように、一つのサービスが巨大な利用者基盤を獲得すると、ネットワーク効果によって「勝者総取り」に近い現象が起きた。
ところがAIは少し違う。
ChatGPTを使っている人がいても、別の人がClaudeを使うことはできる。ユーザーは比較的簡単にサービスを乗り換えられる。
だから、最も優れたAIモデルを作った企業が、必ずしも経済全体を独占するとは限らない。
ファーマン氏は、AIの「最先端」が特に重要になる分野として、サイバーセキュリティーや軍事、諜報などを挙げている。ここでは、2番手では意味がない場合がある。
しかし、経済全体を考えると、もっと重要なのは別の問題だという。
それは、AIを「誰が最初に作ったか」ではなく、「誰が仕事や産業の中でうまく使うか」である。
つまり、「AIの最先端を制した国」と「AIを最も上手に社会へ組み込んだ国」は、同じ国とは限らない。
この一文は、いまの米中AI競争を見るうえで、かなり重要ではないか。米国では「中国に負けるわけにはいかない」という議論が強い。もちろん、国家安全保障の観点から見れば、それには理由がある。中国より優れたAIを軍事やサイバー分野で使えることは、米国にとって大きな意味を持つ。
しかしファーマン氏は、そこに別の事情も入り込んでいるのではないかと指摘する。
AI企業は開発競争を止めたくない。政策担当者の中には規制を緩めたい人もいる。そのため、「中国との競争」を理由にAI開発を急ぐことが、国家安全保障上の必要性と企業側の都合の両方を含んでいる可能性があるというのである。これは米国のAI政策を考える上で、なかなか鋭い指摘である。
もっとも、米中がAIをめぐって競争していること自体は疑いようがない。問題は、競争しているからといって、すべてが対立になるわけではないことだ。
ここでもファーマン氏の話が興味深い。
同氏は今年5月、中国を訪問し、中国の大学教授らと話した。その際、AIによる雇用への影響、教育への影響、AIの安全性などについて、中国側にも米国とよく似た懸念があることを知ったという。
米国の大学教授が心配していることを、中国の大学教授も心配している。米国でAIによって仕事が奪われることを心配している人がいれば、中国にも同じ心配をする人がいる。
AIが人間の制御を離れることへの懸念も共通している。つまり、米中には「競争する理由」がある一方で、「協力する理由」もある。
9月24日のトランプ・習近平会談で習氏がAIについて、米中は競争しているが、それ以上に協力する理由があるとの趣旨を述べたのは、この文脈で見ると興味深い。両国はAIのリスクや利益について対話し、AIの悪用を防ぐ必要があるとも述べた。
もちろん、これを米中AI協調の始まりと見るのは早すぎる。
米国は中国の技術力が軍事転用されることを警戒している。中国側も米国の技術封鎖を警戒している。
半導体、計算能力、AIモデル、データ、レアアース――AIをめぐる競争は、すでに経済安全保障そのものになっている。
実際、トランプ政権は中国とのAI協議を進めながら、米国の技術的優位を維持することを明確にしている。9月18日、スコット・ベッセント財務長官は中国側との協議でAIについて、オープン型、クローズド型モデル、安全策などを議題にすると説明した。米中が共通リスクを避けながら、両国のAIシステムが完全に分断されることも避けたい、という考え方である。
ここに、今回の「現状凍結」の意味がある。凍結とは、問題を解決したということではない。むしろ、問題があまりにも複雑なので、全面衝突を避けながら時間を稼ぐことである。そしてAIの場合、その時間が特に重要になる。なぜなら、技術の進歩が外交交渉より速いからだ。
さらに、ファーマン氏はもう一つ、見逃せない問題を指摘している。それは「AIバブル」である。現在、AI関連の設備投資と需要は猛烈な勢いで増えている。ところが、生産性が同じような速度で上昇しているという証拠は、まだ十分に見えていない。
もしAIによって期待されたほど生産性が上がらなければ、巨額の投資をどう回収するのか。
AIによって生まれるはずの利益を企業が十分に確保できなければ、投資を支えてきた金融市場にも影響が及ぶ。
ファーマン氏は、世界金融危機のような事態になるとは予想していない。しかし、AIに巨額の資金が流れ込んでいるだけに、期待が外れれば「かなり大きな問題」になる可能性があると指摘する。
ここまで来ると、「米国と中国のどちらがAIに勝つのか」という問いは、少し小さく見えてくる。
本当の問題は、AIによって何が変わるのか。そして、その変化を誰がうまく社会に取り込めるのか、である。
米国と中国がAIの最先端を競う。その一方で、両国ともAIによる雇用問題、教育問題、安全問題、金融問題を抱える。
競争相手が同じ問題を抱えているなら、そこには協力の余地がある。
しかし、だからといって技術を無制限に共有できるわけではない。
ファーマン氏自身も、米中間で知識の完全な封じ込めは難しいと認めている。
米国が開発したものを中国が学ぶ。中国が開発したものを米国が学ぶ。AIの知識は、核兵器のように完全に一国の中に閉じ込めておくことが難しい。だから必要なのは、完全な分離でも、無条件の協力でもない。
「どこまで競争し、どこから協力するのか」という境界線を探すことになる。
これは、今回の米中首脳会談が「現状凍結」にとどまったからこそ、これから始まる問題でもある。
日本はどうするのか。9月22日、高市早苗首相とトランプ大統領との日米首脳会談でも、AI、半導体、重要鉱物を含む経済安全保障分野での協力強化が確認された。日本政府は、AIなどの技術競争で日米協力を進め、経済安全保障上の強靱性を高めるとしている。
これは当然の方向である。しかし、日本にとって問題は、米国と一緒に中国に対抗することだけではない。
AIの最先端を持つ米国と中国の間で、日本がどのような役割を果たすのか。
AIを日本の製造業にどう組み込むのか。教育ではどう使うのか。医療ではどう使うのか。そして、防衛・サイバー安全保障ではどこまでAIに依存するのか。さらに、米中がAIの安全性について協議するとき、日本はその議論にどう参加するのか。
ここまで考えると、日本が「米国か中国か」という二者択一だけでAI問題を考えるのは狭すぎる。日本に必要なのは、米国との技術協力を深めながら、自国のAI利用能力を高めること。そして、米中が競争しながらも共通のリスクについて話し合う場に、日本自身も入っていくことである。
今回のトランプ・習近平会談で、AI問題は「解決」されなかった。だからこそ重要なのである。貿易では「現状凍結」ができる。関税を上げるのを待つこともできる。レアアースの輸出をめぐっても、一定期間の取り決めはできる。
しかし、AIの進歩そのものを凍結することはできない。技術は進む。企業は投資する。研究者は発見する。
米国は中国を警戒し、中国は米国を警戒する。その一方で、AIがもたらす雇用、教育、安全性の問題は、国境を越えて同じように人間社会に降りかかってくる。
だから、米中関係の「現状凍結」の先にあるAI問題は、従来の米中覇権競争だけでは説明できない。
ハーバードのファーマン氏が問いかけた「AI競争に勝つ国は本当にあるのか」という疑問は、米国にも中国にも、そして日本にも向けられている。
最先端AIを持つことが勝利なのか。
AIを軍事に最も早く取り入れることが勝利なのか。
AI企業の株価を最も高くすることが勝利なのか。
それとも、AIを人間の仕事、教育、医療、産業の中に最も上手に組み込み、人間がAIを制御できる社会を作った国こそが、長い目で見ればAI時代を生き残るのか。
米中首脳会談が「現状凍結」で時間を稼いだいま、日本にとって本当に重要なのは、米中のどちらが勝つかを待つことではない。
その時間を使って、日本自身が何をするかである。
ハーバードから聞こえてきたのは、「米国か中国か」という答えではない。
むしろ、
AI時代に、本当に勝つとは何なのか。
という、もっと根本的な問いだった。
参考文献
- Harvard Gazette, “Who’ll win AI race? U.S.? China? Maybe no one.” September 24, 2026. Harvard Gazette
- Reuters, “Trump welcomes China’s Xi with fanfare, but no breakthroughs emerge.” September 24, 2026. Reuters
- Reuters, “Five takeaways from Trump’s summit with Xi in Washington.” September 24, 2026. Reuters
- Reuters, “US Treasury’s Bessent to discuss AI, rare earths with China’s He, source says.” September 18, 2026. Reuters
- Ministry of Foreign Affairs of Japan, “日米首脳会談,” September 22, 2026. 外務省
- Ministry of Foreign Affairs of Japan, “Address by Prime Minister TAKAICHI Sanae at the Eighty-first Session of the United Nations General Assembly,” September 22, 2026. 外務省
- The State Council of the People’s Republic of China, “Xi says China, U.S. have more reasons to cooperate than compete in AI,” September 25, 2026. 中国政府網
- Associated Press, “China and the US are competing for AI dominance but have shared concerns over safety.” September 21, 2026. Associated Press
【Cultural Newsへのご支援のお願い】
本稿は、髙濱賛による連載「A Japanese View from America」の一篇です。
Cultural Newsは、日系社会と米国社会を結ぶ架け橋として、報道、論評、文化活動を続けています。こうした活動を継続するためには、読者の皆様のご支援が欠かせません。
本記事に共感いただけましたら、Cultural Newsへのご寄付をご検討いただければ幸いです。皆様からのご支援は、取材活動、記事制作、地域社会への情報発信を支える重要な基盤となります。
今後ともご愛読とご支援を賜りますよう、心よりお願い申し上げます。
Cultural Newsへのドネーションは、下記リンクからお願いします。
https://www.culturalnews.com/donation
【PRI】Behind the U.S.-China “Freeze”: Harvard Rethinks the AI Race — Is There Really a Country That Will Win?
By Tato Takahama
Pacific Research Institute
September 25, 2026
The summit between President Donald Trump and Chinese President Xi Jinping in Washington on September 24 produced plenty of ceremony, but no major breakthrough on the most difficult issues separating the United States and China, including artificial intelligence, trade, Taiwan, and critical minerals. Reuters described the meeting as heavy on symbolism but light on substance.
In that sense, the summit may be better understood not as a historic settlement, but as a temporary “freeze” in a relationship defined by competition.
Yet there is another question that may prove more important than whether Washington and Beijing have temporarily stabilized their differences.
Who will actually win the AI race?
That is the question raised by Harvard economist Jason Furman in a September 24 article in the Harvard Gazette. His answer is striking, perhaps no one.
Furman, a professor of economic policy at Harvard Kennedy School and in Harvard’s Department of Economics, argues that the United States and China are the two principal players in frontier AI. Both countries are investing enormous sums of money in technology, and political leaders increasingly describe their competition as a race.
But, Furman points out, an AI race has no clear finish line.
That observation matters because the phrase “winning the AI race” can mean very different things.
Being first in frontier AI—the most advanced models and systems—may matter enormously in areas such as cybersecurity, military applications, intelligence, and the value of leading technology companies.
But technological leadership is not necessarily the same thing as economic leadership.
The country that develops the most advanced AI may not be the country that makes the most effective use of AI across its economy and society.
That distinction is especially important for Japan.
Japan does not need to decide whether it is on the American side or the Chinese side of an abstract race. It needs to determine how effectively it can use AI in manufacturing, education, medicine, finance, government, defense, and cybersecurity—and how it can contribute to international discussions about the safety and governance of technology.
This is where Furman’s argument becomes particularly interesting.
The United States has increasingly framed its competition with China in AI as a matter of national security. There are legitimate reasons for doing so. The most advanced AI systems could have direct implications for military capabilities, intelligence gathering, cyber operations, and other areas of national power.
At the same time, Furman raises another possibility: the language of a national “race” can also encourage governments and companies to move as quickly as possible, sometimes before the broader economic benefits and risks are fully understood.
That does not mean the competition is imaginary. It means that the definition of victory is still unsettled.
There is another reason the idea of a simple U.S.-China race is misleading.
AI is not a technology that can easily be divided into two national camps.
Knowledge spreads. Researchers move. Companies compete globally. Models learn from enormous bodies of information. Techniques developed in one country can be studied, adapted, improved, or incorporated elsewhere.
Furman therefore argues that completely containing AI knowledge may be nearly impossible.
This creates an unusual situation.
The United States and China may compete intensely over the most advanced AI systems while simultaneously facing many of the same problems.
What will happen to jobs?
How should schools and universities respond?
How should societies deal with misinformation and other forms of misuse?
Who should be responsible when an AI system causes harm?
And how much control should human beings retain over increasingly capable machines?
These are not uniquely American or Chinese questions.
Indeed, Xi Jinping emphasized precisely this point during his September 24 meeting with Trump. According to China’s official account of the talks, Xi said that the United States and China, while competing in AI, have even more reasons to cooperate. He called for continued dialogue on the risks and benefits of AI and for joint efforts to prevent its misuse and abuse. He also said AI should remain under human control.
That does not mean the strategic competition has disappeared.
It means that competition and cooperation are likely to coexist.
The same pattern can already be seen in the broader U.S.-China relationship.
The September 24 summit produced no dramatic settlement of the fundamental disputes between Washington and Beijing. But both governments have an interest in preventing competition from turning into uncontrolled confrontation.
The result is something closer to managed competition than to either full cooperation or complete decoupling.
AI may follow the same pattern.
There is another problem with the race metaphor: money.
The enormous investment now flowing into AI has created a powerful economic boom. Companies are spending heavily on computing infrastructure, data centers, chips, and new AI systems.
But the economic payoff has not necessarily appeared at the same scale.
If AI investment continues to rise while productivity gains remain smaller than expected, the gap could eventually become a financial problem.
In other words, there may be an AI investment boom without an equally large AI productivity revolution—at least not yet.
That is another reason why simply asking which country is “ahead” can obscure the more important question.
What happens after technology is developed?
A country may have the most powerful AI model in the world and still fail to transform its broader economy if businesses, schools, hospitals, government agencies, and workers cannot use the technology effectively.
Conversely, a country that does not produce every leading AI model could gain enormous economic value by becoming exceptionally good at adopting and integrating AI.
The country that leads in frontier AI and the country that makes the best use of AI across society may not be the same country.
For Japan, that distinction should not be overlooked.
Japan already has important reasons to work with the United States in AI and related technologies. At their September 22 summit in New York, Prime Minister Sanae Takaichi and President Trump agreed to further strengthen cooperation in economic security, including AI, semiconductors, and critical minerals.
That cooperation gives Japan an important foundation.
But cooperation with the United States cannot substitute for Japan’s own decisions about how to use AI.
Japan must decide where AI can strengthen its industrial base, how it should be incorporated into education and medicine, how government should use it, and how national security and cybersecurity should adapt.
It also needs to participate actively in international discussions over AI safety and governance.
The question is therefore not simply whether America or China will win.
The more difficult question is what “winning” will mean in an age when AI itself is changing the meaning of technological power.
The Trump-Xi summit may have produced a temporary freeze in some areas of the U.S.-China confrontation. But a freeze is not a settlement. It simply creates time.
The real question for Japan—and for every other country watching the U.S.-China competition—is what it will do with that time.
Harvard’s Furman offers a useful warning against treating AI as a race with a finish line.
There may be no single winner.
There may instead be countries that develop the most advanced technology, countries that use it most effectively, and countries that learn how to manage its risks.
Those categories may overlap. They may not.
The age of AI is only beginning.
The question is no longer simply, “Who will win the AI race?”
It is this:
What will it actually mean to win in the age of AI?
Note: English translation was prepared by the author. Translation assistance tools may have been used.
In that sense, the summit may be better understood not as a historic settlement, but as a temporary “freeze” in a relationship defined by competition.
Yet there is another question that may prove more important than whether Washington and Beijing have temporarily stabilized their differences.
Who will actually win the AI race?
That is the question raised by Harvard economist Jason Furman in a September 24 article in the Harvard Gazette. His answer is striking, perhaps no one.
Furman, a professor of economic policy at Harvard Kennedy School and in Harvard’s Department of Economics, argues that the United States and China are the two principal players in frontier AI. Both countries are investing enormous sums of money in technology, and political leaders increasingly describe their competition as a race.
But, Furman points out, an AI race has no clear finish line.
That observation matters because the phrase “winning the AI race” can mean very different things.
Being first in frontier AI—the most advanced models and systems—may matter enormously in areas such as cybersecurity, military applications, intelligence, and the value of leading technology companies.
But technological leadership is not necessarily the same thing as economic leadership.
The country that develops the most advanced AI may not be the country that makes the most effective use of AI across its economy and society.
That distinction is especially important for Japan.
Japan does not need to decide whether it is on the American side or the Chinese side of an abstract race. It needs to determine how effectively it can use AI in manufacturing, education, medicine, finance, government, defense, and cybersecurity—and how it can contribute to international discussions about the safety and governance of the technology.
This is where Furman’s argument becomes particularly interesting.
The United States has increasingly framed its competition with China in AI as a matter of national security. There are legitimate reasons for doing so. The most advanced AI systems could have direct implications for military capabilities, intelligence gathering, cyber operations, and other areas of national power.
At the same time, Furman raises another possibility: the language of a national “race” can also encourage governments and companies to move as quickly as possible, sometimes before the broader economic benefits and risks are fully understood.
That does not mean the competition is imaginary. It means that the definition of victory is still unsettled.
There is another reason the idea of a simple U.S.-China race is misleading.
AI is not a technology that can easily be divided into two national camps.
Knowledge spreads. Researchers move. Companies compete globally. Models learn from enormous bodies of information. Techniques developed in one country can be studied, adapted, improved, or incorporated elsewhere.
Furman therefore argues that completely containing AI knowledge may be nearly impossible.
This creates an unusual situation.
The United States and China may compete intensely over the most advanced AI systems while simultaneously facing many of the same problems.
What will happen to jobs?
How should schools and universities respond?
How should societies deal with misinformation and other forms of misuse?
Who should be responsible when an AI system causes harm?
And how much control should human beings retain over increasingly capable machines?
These are not uniquely American or Chinese questions.
Indeed, Xi Jinping emphasized precisely this point during his September 24 meeting with Trump. According to China’s official account of the talks, Xi said that the United States and China, while competing in AI, have even more reasons to cooperate. He called for continued dialogue on the risks and benefits of AI and for joint efforts to prevent its misuse and abuse. He also said AI should remain under human control.
That does not mean the strategic competition has disappeared.
It means that competition and cooperation are likely to coexist.
The same pattern can already be seen in the broader U.S.-China relationship.
The September 24 summit produced no dramatic settlement of the fundamental disputes between Washington and Beijing. But both governments have an interest in preventing competition from turning into uncontrolled confrontation.
The result is something closer to managed competition than to either full cooperation or complete decoupling.
AI may follow the same pattern.
There is another problem with the race metaphor: money.
The enormous investment now flowing into AI has created a powerful economic boom. Companies are spending heavily on computing infrastructure, data centers, chips, and new AI systems.
But the economic payoff has not necessarily appeared at the same scale.
If AI investment continues to rise while productivity gains remain smaller than expected, the gap could eventually become a financial problem.
In other words, there may be an AI investment boom without an equally large AI productivity revolution—at least not yet.
That is another reason why simply asking which country is “ahead” can obscure the more important question.
What happens after technology is developed?
A country may have the most powerful AI model in the world and still fail to transform its broader economy if businesses, schools, hospitals, government agencies, and workers cannot use the technology effectively.
Conversely, a country that does not produce every leading AI model could gain enormous economic value by becoming exceptionally good at adopting and integrating AI.
The country that leads in frontier AI and the country that makes the best use of AI across society may not be the same country.
For Japan, that distinction should not be overlooked.
Japan already has important reasons to work with the United States in AI and related technologies. At their September 22 summit in New York, Prime Minister Sanae Takaichi and President Trump agreed to further strengthen cooperation in economic security, including AI, semiconductors, and critical minerals.
That cooperation gives Japan an important foundation.
But cooperation with the United States cannot substitute for Japan’s own decisions about how to use AI.
Japan must decide where AI can strengthen its industrial base, how it should be incorporated into education and medicine, how government should use it, and how national security and cybersecurity should adapt.
It also needs to participate actively in international discussions about AI safety and governance.
The question is therefore not simply whether America or China will win.
The more difficult question is what “winning” will mean in an age when AI itself is changing the meaning of technological power.
The Trump-Xi summit may have produced a temporary freeze in some areas of the U.S.-China confrontation. But a freeze is not a settlement. It simply creates time.
The real question for Japan—and for every other country watching the U.S.-China competition—is what it will do with that time.
Harvard’s Furman offers a useful warning against treating AI as a race with a finish line.
There may be no single winner.
There may instead be countries that develop the most advanced technology, countries that use it most effectively, and countries that learn how to manage its risks.
Those categories may overlap. They may not.
The age of AI is only beginning.
The question is no longer simply, “Who will win the AI race?”
It is this:
What will it actually mean to win in the age of AI?
Note: English translation was prepared by the author. Translation assistance tools may have been used.
