Jensen Huang walked onstage and called $MRVL the “next trillion-dollar company.”
The stock moved 32% in a single session. $56 billion in market cap. One afternoon.
Everyone asked: is $MRVL really better than $AVGO? But missed the whole point.
Jensen wasn’t making a product argument.
He was pointing at something much larger.
The AI infrastructure buildout has reached a scale where the conversation about who has the *best* product is being temporarily overridden by a more urgent one:
Who has product at all?
The numbers are difficult to process with normal intuition.
The top five hyperscalers are projected to spend $602 billion in capital expenditures in 2026 alone. $MSFT, $META, $GOOG, and $AMZN combined are on track for over $300 billion annually, just on data center infrastructure.
Global data center power consumption is expected to increase by nearly 126 GW every year through 2028.
Goldman Sachs projects a 165% increase in power demand by 2030 versus 2023 levels.
A typical hyperscale data center consumes as much electricity as roughly 100,000 households. The largest campuses now under construction demand approximately twenty times that.
This is not a growth trend. This is a structural reordering of where energy, capital, and compute flow on a planetary scale.
When demand moves at this velocity, it breaks the normal logic of competitive markets.
In a normal market, the best product captures share and second place loses ground. That math requires supply to be the constant and quality to be the variable.
Flip those, make quality the constant and supply the variable and the dynamic inverts completely.
Everything that can handle the workload gets absorbed. The constraint isn’t who built the better chip.
The constraint is who can deliver at all, at the scale being demanded, within the timelines hyperscalers are operating on.
That’s the environment we’re in right now.
And it’s the environment that makes the entire AI infrastructure stack; compute, networking, power, land, cooling one of the most consequential investment territories of the next decade.
The Marvell moment wasn’t a product endorsement. It was a signal about the nature of the demand.
When even second place becomes a trillion-dollar company, you start to understand the size of what’s coming.
—BP
This is not financial advice. Personal analysis. Do your own research.
Nvidia CEO'su Jensen Huang'a, 'hayatında tanıdığınız en zeki kişi kim' sorusuna cevabı:
- Tanıdığım en zeki insan üniversite giriş sınavından berbat bir puan bile almış olabilir.
- Herkes yazılım programlamanın nihai akıllı meslek olduğunu düşünüyordu.
- Yapay zekanın çözdüğü ilk şey ne oldu? Yazılım programlama.
- Zeki tanımı çoğu insanın düşündüğünden çok farklı.
- Gerçek zeka: Teknik yetenek + İnsan empatisi + Söylenmeyeni anlama becerisi
- Köşelerin ötesini görebilen insanlar gerçekten, gerçekten akıllıdır.
- Sorunları ortaya çıkmadan önce önleyebilmek - sadece havayı hissettiğin için.
- O hava: Veri + Analiz + İlk prensipler + Yaşam deneyimi + Bilgelik + Diğer insanları hissetmek
- İşte bu zekadır.
- Geleceğin zeki tanımı bu olacak.
Ve o kişi SAT'den berbat bir puan alabilir.
全球十大富豪裡面,最有媒體曝光率的就是馬斯克和黃仁勳,這兩個人,既沒像暴發戶一樣到處炫富,也不像怕人家覬覦財富一樣隱姓埋名,他們好像負有任務一樣,要和世人,要和青史有一個交代而到處行走,這是什麼「為富之道」?
黃仁勳和川普去了北京之後,留在北京到處吃喝,彼時網上有人開玩笑,「好像一群同學到亞洲參訪,然後其它美國同學隨團回國,但這個亞裔同學,好像第一次發現不當少數民族的好處,變成注目的焦點,而非常享受這個獨自留下的時刻。」也許,但不只這樣。歷史從來沒有一個亞洲人經營過世界最大的公司,從來沒有一個東亞裔在世界十大首富,郭台銘、李嘉誠、馬雲、孫正義都沒達到這個境界,更不用說達到黃仁勳在產業的重要性,而且黃仁勳還可以在世界舞台,絲毫不勉強的和各地顯要名人,平起平坐,侃侃而談,這是多少人從小學英文,從小看著歐美通俗文化而嚮往而永遠達不到的地步,因此他在亞洲各地受到瘋狂的歡迎,是完全可以理解的。
但黃仁勳除了享受當超級巨星外,有什麼理由這樣到處行走,跑一個近似虐待自己的行程?我不知道這是不是他的理由,但我猜測他如此折磨自己的一個原因是,這是人類歷史的一個關鍵時刻,AI即將永遠改變人類社會,從來沒有一個科技,有這麼深遠而迅速的影響,人類社會處於一個不確定的時刻,因此有很多對AI有懷疑的聲音,這個聲音會因為工作型態發生改變,因為財富和社會地位迅速重組,而越來越大,反對的力量會集結,可能在政治上形成阻擋AI進步的勢力。面對這樣的阻力,身為最大AI公司的負責人,也可以說是最了解AI科技的人之一,黃仁勳有責任去帶頭對抗這個力量。對抗這個反對的力量有很多的方法,可以在政黨政治裡用錢影響,也可以像黃仁勳一樣,給AI一個人性的面孔。「原來是這樣一個可親可愛的人在發展AI科技,所以AI應該不可怕吧!」所以黃仁勳一直要上媒體,一直要接近群眾,不斷地避免爭議,但像個人一樣講話、吃飯、跳舞,只是跟在他身邊的那些財閥大佬,看起來就勉強而辛苦了。
而如果說黃仁勳是AI的代言人,那馬斯克就是AI的工程師。當你要變成歷史上第一個trillionaire,財富、名聲、事業等等,通通不重要了,你還要追求什麼?我想馬斯克體現的人生哲學,就是,It's all about the journey, not the destination (終點不重要,過程本身就是一切),如果你享受這個困難的過程,勝過人生可以有的所有奢華,那當然是繼續工作,不斷工作。這一點,是所有工作狂都可以理解的,但不同的是,馬斯克選了一個幾乎永遠到不了的終點,至少在他有限人生達不到的終點,所以他可以一直「享受」這個工作的快樂與痛苦。曾子曰,「士不可以不弘毅,任重而道遠。仁以為己任,不亦重乎?死而後已,不亦遠乎?」馬斯克正在展現儒家的極致精神。
炫富、享受人生、慈善事業,這些傳統發達致富之後的選擇,都不是黃仁勳和馬斯克有興趣的事。我們何其有幸,生於這個時代,親身體驗從來沒有的科技,更親眼目睹一個更高的人生境界。
Larry Ellison: "Tesla 的 AI 跟普通 AI 不一样,因为车和机器人面对的是物理世界。
开车时,危险可能在微秒级发生。系统必须立刻看见、理解、避让,不能把请求发到远端服务器,再等模型返回。
所以 Tesla 车和机器人都必须有本地算力。自动驾驶不是聊天机器人,延迟不是体验问题,而是安全问题。"
BREAKING:
Elon Musk will make a major announcement at ASML's private technology conference Thursday.
One day before the expected SpaceX IPO.
ASML makes the machines that print the world's most advanced chips.
Without ASML.
No Nvidia. No TSMC. No advanced AI hardware.
Elon speaking at the most important chip machine company on earth.
The day before the biggest IPO in history.
Is not random.
Watch Thursday closely.
INSTEAD OF WATCHING NETFLIX TONIGHT.
Spend 1 hour with this.
Claude AI FULL COURSE that teaches you how to BUILD and AUTOMATE anything.
The people who watch this tonight will wake up tomorrow with a new skill.
Watch it and Bookmark it now.
Jensen Huang, CEO de Nvidia, te está diciendo dónde invertir en 2026.
Ha dirigido personalmente el capital de Nvidia hacia 8 empresas específicas por un total combinado de más de 45.000 MILLONES de dólares.
Aquí es donde está poniendo su dinero la empresa más importante de la economía de la IA.
La lista completa:
→ OpenAI: 30.000 M$
El mayor compromiso de los 8. Nvidia financia la infraestructura de computación de OpenAI desde dentro. OpenAI es también su mayor cliente individual.
→ GLW Corning: 3.200 M$
Vidrio óptico y fibra para conectar físicamente los clústeres de IA. Sin ello, no puedes mover datos entre millones de GPUs.
→ IREN: 2.100 M$
Proveedor de IA en la nube con una de las posiciones energéticas más sólidas de Norteamérica.
→ MRVL Marvell: 2.000 M$
Chips de red personalizados que mueven datos entre GPUs a gran escala.
→ LITE Lumentum: 2.000 M$
Láseres y componentes ópticos para la columna vertebral de fibra de cada data center de IA.
→ COHR Coherent: 2.000 M$
Transceptores de fibra óptica que conectan clústeres de GPUs dentro de los centros de datos.
→ CRWV CoreWeave: 2.000 M$
GPU-as-a-service. El mayor cliente cloud de Nvidia fuera de los hyperscalers.
→ NBIS Nebius: 2.000 M$
Infraestructura de IA en la nube. Construyendo silenciosamente capacidad GPU a hiperescala para los laboratorios de IA.
Lo que Nvidia compra es donde irá el dinero a continuación.