Microsoft absorbs inflection.

微軟吸收拐點。

The big organizational change happening at Microsoft around AI.

微軟圍繞人工智能進行的重大組織變革。

Amazon consumes the debt.

亞馬遜消耗了債務。

Its startup was valued at more than a billion dollars.

其初創公司估值超過 10 億美元。

Now, most of the company is now going to Amazon.

現在，該公司的大部分業務都交給了亞馬遜。

Google and character AI, three of the biggest tech companies swallowing three of the biggest startups in the AI world without actually acquiring them.

谷歌和人工智能公司（character AI），這三家最大的科技公司吞併了人工智能領域最大的三家初創公司，卻沒有真正收購它們。

The reality is everywhere you look, we see rapid concentrations.

現實情況是，我們到處都能看到快速集中的現象。

They're quelling competition, poaching top talent and skirting scrutiny, while regulators are still playing checkers.

它們在遏制競爭，挖走頂尖人才，規避審查，而監管機構仍在下跳棋。

I'm Deirdre Bosa and this week on Tech Check, the AI M&A fake out.

我是迪爾德麗-博薩（Deirdre Bosa），本週的 Tech Check 欄目是人工智能併購假象。

Character AI was one of the most promising startups in the age of generative AI.

Character AI 是生成式人工智能時代最有前途的初創公司之一。

Personalized AI.

個性化人工智能。

How are thee?

你好嗎？

Let's get to the bottom of this.

讓我們一探究竟。

Highly engaging.

極具吸引力。

Loyal user base.

忠實的用戶群。

Somebody who sends one message today is active for on average two hours in that day. $150 million in funding at a billion dollar valuation.

今天發送一條資訊的人在當天平均活躍兩個小時。融資 1.5 億美元，估值 10 億美元。

Character.ai launched its software last September and has had more than 173 million visits last month, with a 61% increase from March.

Character.ai 於去年 9 月推出了自己的軟件，上個月的訪問量已超過 1.73 億次，比 3 月份增長了 61%。

And a founder who pioneered generative AI.

還有一位開創了生成式人工智能的創始人。

When I got started working on these large language models and inventing this field, like in 2016 to 2018, like very few people were excited.

當我開始研究這些大型語言模型併發明這一領域時，比如在 2016 年到 2018 年，好像很少有人感到興奮。

There was just one problem.

只有一個問題。

The startup couldn't make money.

創業公司賺不到錢。

And according to reports, it struggled to generate enough revenue from paid users.

據報道，該公司很難從付費用戶那裡獲得足夠的收入。

Character.ai, billion dollar valuation, pre-revenue, very much feels like the 2010 App Store where you gain a lot of users and kind of figure out the monetization later.

Character.ai，估值十億美元，尚未實現盈利，感覺很像 2010 年的 App Store，先獲得大量用戶，然後再想辦法盈利。

So it turned to a well-established money printing mega cap, Google, which was happy to cut a deal, especially since it was arguably more interested in the talent than the technology.

於是，它求助於一家老牌印鈔巨頭--谷歌，後者很樂意與它達成協議，尤其是它對人才的興趣可以說超過了對技術的興趣。

Character.ai co-founder, Noam Shazir, was an author of a groundbreaking paper that laid the foundation for today's age of AI.

Character.ai 的聯合創始人諾姆-沙茲爾（Noam Shazir）是一篇開創性論文的作者，這篇論文奠定了當今人工智能時代的基礎。

It would bring Shazir back to Google, along with a fifth of his employees and license Character's technology without having to outright buy the startup in an acquisition.

這將使沙茲爾及其五分之一的員工重返谷歌，並獲得 Character 的技術許可，而不必通過收購直接買下這家初創公司。

Every person on earth is going to be inventing their own applications.

地球上的每個人都將發明自己的應用程序。

It's a playbook that skirts the regulatory crackdown on big tech dominance, provides an exit for AI startups struggling to make money and allows mega caps to pick up the talent needed in the AI arms race.

這套遊戲規則可以規避監管部門對大型科技公司主導地位的打擊，為掙扎在盈利邊緣的人工智能初創公司提供出路，並讓巨型公司在人工智能軍備競賽中獲得所需的人才。

There's a very small number of researchers, AI practitioners that operate at the frontier of the technology.

只有極少數研究人員和人工智能從業人員在技術前沿開展工作。

There's probably only thousands right now and the hyperscalers, the mega caps want that talent and they're looking to get it any way they can.

現在可能只有幾千人，而那些超大規模企業和巨型企業都想得到這些人才，他們正在想方設法獲得這些人才。

A playbook that Microsoft pioneered.

微軟首創的玩法。

In March, it inked an unusual deal with Inflection, a prominent AI startup with a chat bot to rival ChatGPT's.

今年 3 月，該公司與 Inflection 簽訂了一份不同尋常的協議，後者是一家著名的人工智能初創公司，其哈拉機器人可與 ChatGPT 相媲美。

Again, not an outright acquisition, just a $650 million agreement that would allow Microsoft to use Inflection's AI models and hire away most of its staff, including founder Mustafa Suleyman, whom they put in charge of its entire AI unit.

同樣，這也不是一次徹底的收購，只是一份價值 6.5 億美元的協議，允許微軟使用 Inflection 的人工智能模型，並僱傭其大部分員工，包括創始人 Mustafa Suleyman，讓他負責整個人工智能部門。

He co-founded DeepMind back in 2010.

早在 2010 年，他就與他人共同創辦了 DeepMind 公司。

That was later acquired by Google.

該公司後來被谷歌收購。

In 2022, he co-founded a company called Inflection.

2022 年，他共同創辦了一家名為 Inflection 的公司。

He did that with Reid Hoffman.

他和裡德-霍夫曼（Reid Hoffman）就是這麼做的。

In June, Amazon and Adept AI struck a deal.

今年 6 月，亞馬遜與 Adept AI 達成協議。

Amazon reportedly paid at least $330 million to license Adept's technology with $100 million retention bonus to employees who jumped ship.

據報道，亞馬遜至少支付了 3.3 億美元來許可 Adept 的技術，並向跳槽的員工發放了 1 億美元的留用獎金。

A big cost center for these startups is GPUs, those chips, cloud computing.

這些初創公司的一大成本中心就是 GPU、這些芯片和雲計算。

And what does Amazon have?

亞馬遜有什麼？

They've got AWS, they've got the cloud computing side in spades.

他們擁有 AWS，在雲計算方面也是如魚得水。

So if you're a startup and you're looking to partner with one of these companies or do some sort of licensing deal, you're gonna look at one of the big tech companies, likely Amazon, Google, Microsoft.

是以，如果你是一家初創公司，並希望與這些公司中的一家合作或達成某種許可協議，你就會把目光投向大型科技公司，很可能是亞馬遜、谷歌和微軟。

For the mega cap and the startup employees who were hired away, it's win-win.

對於巨型公司和被聘用的初創公司員工來說，這是雙贏。

The founders no longer feel pressured to monetize their business models.

創始人不再感到商業模式貨幣化的壓力。

Being swallowed up by a highly profitable mega cap, that buys them time to build for the longer term.

被高利潤的巨型企業吞併，為他們贏得了更長遠的建設時間。

As these founders are looking to scale their companies, they might see headwinds, any number of headwinds.

當這些創始人希望擴大公司規模時，他們可能會遇到各種不利因素。

And it could just be that at this point in time, the best home for them is, again, a mega cap.

可能只是在這個時候，他們最好的歸宿還是巨型帽。

And their cloud computing costs needed to develop gen AI models running into the hundreds of millions of dollars, they get absorbed by the tech giants that own the infrastructure.

而他們開發基因人工智能模型所需的雲計算成本高達數億美元，這些成本會被擁有基礎設施的科技巨頭吸收。

We definitely talk to founders that wanna develop the best technologies and just have access to GPU resource and not fight for it.

我們肯定會與那些想開發最佳技術、只想獲得 GPU 資源而不想為之奮鬥的創始人交談。

And I think that's why you've seen some funds spin up clusters of their own to give founders access.

我想這就是為什麼你會看到一些基金成立了自己的集群，為創始人提供訪問權限。

For the mega caps, it's a quick and easy way to get the best talent in the field.

對於大公司來說，這是在該領域獲得最佳人才的快捷方式。

And that was likely a motivation when Google purchased DeepMind back in 2014, an example of an all-out acquisition that is highly unlikely to get approved today.

這很可能也是谷歌在 2014 年收購 DeepMind 時的動機之一。

DeepMind was one of the most prestigious, independent AI research labs a decade ago.

十年前，DeepMind 還是最負盛名的獨立人工智能研究實驗室之一。

Google was early and managed to acquire it before gen AI was even on the radar of regulators.

谷歌先行一步，在監管機構還未注意到人工智能基因時就成功收購了它。

Today though, AI is squarely in their crosshairs and gen AI the biggest arms race in tech.

如今，人工智能已成為他們的目標，而人工智能基因也成為科技界最大的軍備競賽。

Gen AI aside, Apple, Google, Amazon, and Meta, they all have pending cases before the antitrust chiefs of the FTC and the DOJ.

除 Gen AI 外，蘋果、谷歌、亞馬遜和 Meta 都有未決案件要提交給聯邦貿易委員會和司法部的反壟斷主管。

Google has already found itself on the losing end with a federal court ruling it has a monopoly in the search market and now faces a potential breakup.

聯邦法院裁定谷歌壟斷了搜索市場，現在谷歌可能面臨解體。

Big tech might think they're outsmarting antitrust enforcers and sidestepping scrutiny, but they could be playing with fire.

大型科技公司可能會認為他們在智取反壟斷執法者，逃避審查，但他們可能是在玩火。

The regulators of today, they do have a few moves up their sleeves.

今天的監管者，他們確實有幾招。

They're more agile and aggressive than ever.

他們比以往任何時候都更加敏捷和積極。

The FTC has already opened an antitrust probe into Microsoft's deal with inflection, investigating whether Microsoft structured that deal to gain control of the startup while also dodging review.

美國聯邦貿易委員會已經開始對微軟與 inflection 的交易進行反壟斷調查，調查微軟是否通過安排該交易獲得了對這家初創公司的控制權，同時也逃避了審查。

And the agency has also launched an informal inquiry into the Amazon ADEPT deal.

該機構還對亞馬遜 ADEPT 交易展開了非正式調查。

Even Congress has taken notice with three US senators calling for an investigation.

就連美國國會也注意到了這一點，三位參議員呼籲進行調查。

Lena Khan and others have looked at it.

莉娜-汗和其他人都研究過這個問題。

It's not that there's been wholesale mergers of these things, but even the partnerships unto themselves.

這並不是說這些東西已經進行了大規模合併，而是說它們本身就是合作關係。

Their main concern isn't just whether or not the acquisition happened on a technical level.

他們最關心的不僅僅是收購是否在技術層面上實現。

At this moment, AI is a nascent technology that could catalyze enormous growth and innovation.

目前，人工智能是一項新興技術，可以催化巨大的增長和創新。

There is enormous opportunity here for a vibrant market, but we also see risk.

這裡蘊藏著充滿活力的巨大市場機遇，但我們也看到了風險。

We see risk of consolidation.

我們看到了整合的風險。

We see risk of monopolization.

我們看到了壟斷的風險。

It's whether or not the act is ultimately anti-competitive, especially in a nascent, growing space where having competition is key to making sure that there's innovation instead of the same incumbent behemoths getting ever larger and more powerful.

關鍵在於該法案最終是否有反競爭之嫌，尤其是在一個新生的、不斷髮展的領域，競爭是確保創新的關鍵，而不是讓同樣的在位巨頭變得越來越大、越來越強。

Because yes, megacops, they aren't swallowing startups whole.

因為是的，大警察們，他們並沒有把創業公司整個吞下。

They're leaving behind a shell of a company and some of its employees, still functioning but not competitive, which also brings into sharp relief just who gets left holding the bag.

他們留下的只是一個空殼公司和部分員工，雖然仍在運轉，但已經失去了競爭力，這也讓誰會被拋棄的問題變得更加突出。

On the surface, these deals seem like win-win, but there are employees and investors who are left behind.

從表面上看，這些交易似乎是雙贏的，但卻留下了員工和投資者。

In the Adept deal, for example, Amazon only hired away the AI researchers, according to the New York Times.

例如，據《紐約時報》報道，在 Adept 的交易中，亞馬遜只聘用走了人工智能研究人員。

The teams working on products, sales, and other areas weren't so lucky.

而負責產品、銷售和其他領域的團隊就沒那麼幸運了。

They don't reap any of the rewards that their former colleagues do, a new cushier gig at a proven big corporation without the desperate search for viable revenue.

他們不會像以前的同事那樣獲得任何回報，在一家成熟的大公司裡找到一份更舒適的新工作，而不用拼命尋求可行的收入。

Instead, they're left trying to find a way to piece together a new business without its top talent.

相反，他們只能在沒有頂尖人才的情況下，想方設法組建新的企業。

In a number of deals of this type, the remaining company still has access to the IP that's been created and certainly enough resources as part of the transaction to continue to hire talent and build upon that IP.

在許多此類交易中，剩下的公司仍可使用已創建的知識產權，而且作為交易的一部分，肯定有足夠的資源繼續聘用人才，並在知識產權的基礎上繼續發展。

Then there's the VC money that initially founded those startups.

還有最初創立這些初創公司的風險投資資金。

Those that backed Character, Adept, and Inflection, they're being made whole or almost whole through licensing deals, but it is a far cry from the returns they were looking for.

那些為 Character、Adept 和 Inflection 提供支持的人，通過授權交易獲得了全部或幾乎全部的回報，但這與他們所期望的回報相去甚遠。

Typically, VCs bet on bigger payoffs.

通常情況下，風險投資公司賭的是更大的回報。

A minimum five to 10 times rate of return is considered a benchmark for success.

至少 5 至 10 倍的回報率被認為是成功的基準。

One insider who wanted to remain anonymous tells me that there could be more of these types of deals to come, noting that there are many more overvalued AI startups that have yet to find viable business models.

一位不願透露姓名的業內人士告訴我，未來可能會有更多此類交易，他指出，還有很多被高估的人工智能初創企業尚未找到可行的商業模式。

Cohere, for example, is another still-independent Gen AI darling that was recently valued at more than $5 billion.

例如，Cohere 是另一個仍然獨立的 Gen AI 寵兒，最近估值超過 50 億美元。

It reportedly has less than 25 million in annualized revenue.

據報道，該公司的年收入不到 2500 萬美元。

A lifeline from a mega cap, that could be a welcome development as the pressure to monetize builds.

在貨幣化的壓力不斷增加的情況下，來自巨型上限的生命線可能是一個值得歡迎的發展。

But as regulators get wise to the pseudo acquisition, will big tech be willing to tempt them with more deals or will the AI unicorns be left to fend for themselves as the big get bigger?

但是，隨著監管機構對這種偽收購行為的警覺，大型科技公司是否願意用更多的交易來誘惑他們，或者說，隨著大公司變得越來越大，人工智能獨角獸們是否只能自生自滅？

However it plays out, the AI M&A fakeout has changed the landscape for generative AI.

無論結果如何，人工智能併購假象已經改變了生成式人工智能的格局。