- Nvidia is the world's most popular市紅 maker of artificial intelligence chi請女ps.
- It has released its la林話test chip, which is cl放廠aimed to be nine times fast高黑er than its predecessor.
- Only AMD has a chance to challeng輛你e Nvidia's AI chip dominance

If there's one company currently (or ev低雪en in the foreseeable future) dominatin好高g the tech industry, it's Nvi筆讀dia. The tech giant's produ土人cts, especially AI chips, are 子如the industry's most valuable r他城esource. So much so t報林hat NVIDIA's market capital家草ization has surpassed $著輛1 trillion, making it the店歌 most valuable chipm河站aker in the world, signaling a cle分南ar boom in demand for AI.

The company is the biggest beneficiary家街 of the rise of ChatGPT物化 and other generative arti電人ficial intelligence (AI) app書商lications, almost all o生東f which are powered by its ve男謝ry powerful graphics processors.校鄉 Prior to this, Nvidia's chips were火務 also widely used to pow友這er traditional AI systems,嗎術 and during the cryptocurrency boom鄉錯, demand for the chips increased as sys如為tems in the industry also re哥說lied on their processing power.

Nvidia monopolized the AI 員關market with its chips, and it di現的dn't start in the post-ChatGPT er熱謝a. Nvidia founder and CEO冷嗎 Jen-Hsun Huang has bee跳是n so certain about AI 吧去in the past that most of his certaintie音街s have felt like hig水站h-profile rhetoric or marketing 醫呢gimmicks.

How NVIDIA Got Ahead of the Curve to Re來鐘ach the AI Chip Inflection Point

NVIDIA, led by Jen-Hsun Huang, has al愛店ways been on the cusp 資章of development - fir是器st gaming, then machine learning醫玩, then cryptocurrency mining, data 為紙centers, and now artificial inte舊體lligence. Over the past d自兒ecade, the chip giant h吧他as developed a unique portfolio of 會吧hardware and software如你 products designed to democr了舞atize AI and enable the company to be間雪nefit from adopting AI workl務雜oads.

But the real turning point 科機came in 2017, when N道這vidia began adapting GPUs 什遠to handle specific AI calculations. T理間hat same year, Nvidia, whi子低ch typically sells chips or board自林s for other companies' sys水雜tems, also began selling complete comp唱兵uters to perform AI tas路鐘ks more efficiently.

Some of its systems are now the的小 size of supercomputers, ass線視embled and run using proprietary鄉明 networking technology and thousands of去畫 GPUs. Such hardware may ru可暗n for weeks to train the latest刀購 AI models. For some comp下坐etitors, competing w厭弟ith companies that sell computers, so海短ftware, cloud services, train間湖ed AI models and processors is very 術鐘difficult.

By 2017, tech giants like Google, M鄉姐icrosoft, Facebook and 年北Amazon had bought more Nvidi件訊a chips for their da一靜ta centers. Organizations like Massach你車usetts General Hospital use章去d Nvidia chips to spot an分弟omalies in medical images like 很水CT scans.

By then, Tesla had already an秒厭nounced that it would install 個要Nvidia GPUs in all of家唱 its cars to enable autonomous drivi歌刀ng. nvidia chips provide a strong 算通base of capabilities for virtual rea術讀lity headsets, like th黑我e ones that Facebook and HTC are bri生們nging to market. Nvidia h姐如as a reputation for 習麗consistently delivering 呢睡faster chips every few years.

Looking back, it's s農問afe to say that for mo什到re than a decade, Nvi不做dia has established a nearly imp煙那enetrable lead in producin友自g chips that perform complex AI tasks l美和ike image, facial, and speech re老玩cognition, as well as generating 也就text for chatbots like Op醫喝enAI's ChatGPT.

The biggest benefit? NVIDIA a身開chieved its dominance by r離月ecognizing AI trends early放亮, customizing its chips 你線for those tasks, and then develop聽分ing the critical software that hel時個ps AI development. As the 體鐘New York Times (NYT) not制年es, NVIDIA has gradually transformed 很身itself into a one-stop stor司看e for AI development.

While Google, Amazon, Me要資ta, IBM and others also produc開子e AI chips, Nvidia now accounts for 村我more than 70 percent of global AI ch笑錢ip sales, according to res外人earch firm Omdia. It's even more promi白讀nent in training generative 外內AI models. Microsoft alone has spent拍做 hundreds of millions of術訊 dollars on tens of thousands都話 of Nvidia A100 chips to help build Cha明車tGPT.

In an article by Don Cla照金rk of the New York Times, he said, 新化"By May, the company's positio店章n as the clearest winner of t短服he artificial intelligence revolut些朋ion had become clear, with quarterly山民 revenues projected to grow by 64 perc是山ent, far outpacing Wall Stre黃子et's expectations."

Generative AI's 'iPhone m懂森oment': Nvidia's H100媽動 AI chip

"We're in the iPhone moment o報知f AI," Jen-Hsun Huang said外鄉 at the GTC conference in March. 黃國He was also quick to point out N喝不vidia's role in the beg去開inning of this AI wave: he brought a DG坐志X AI supercomputer to Open票來AI in 2016, and that hardware w車高as eventually used to build劇南 ChatGPT.

At GTC earlier this year, Nvidia unve用哥iled the H100, the successor t舊厭o Nvidia's A100 GPU, which河一 underpins modern lar慢些ge-scale language modeling efforts, 相樂and Nvidia claims that its H100, as計和 a GPU, delivers AI training that's u電匠p to nine times faster than the A100,山鐘 and inference that's up人月 to 30 times faster than the A100.

While Nvidia did not文吧 discuss pricing or chip allocat湖光ion policies, industry executives an近放d analysts have said tha睡自t each H100 costs betw得對een $15,000 and more than $40,000, dep民銀ending on packaging and other f森做actors, and is roughly 友開two to three times the cost of the 看舞previous generation A100 chip.

However, H100 continues to face se光跳vere supply shortages am長熱id soaring demand and easing short地友ages in most other chip categories. R你月euters reports that analyst唱不s believe Nvidia will only be熱短 able to meet half of the demand, w公志ith its H100 chips selling fo門用r twice the original $20,000 price t計到ag. According to Reuters, this trend 也時could continue for several quarters.

There's no denying th妹東at the surge in demand is coming from 這窗China, where companie唱火s are stockpiling chips廠高 due to U.S. chip export restric姐民tions. The Financial這小 Times reports that leading Chin鄉船ese Internet companies have ordered $你城5 billion worth of chips我信 from NVIDIA. The supply-dem少們and divide will inevitably lead some河好 buyers to turn to N師亮VIDIA's rival AMD, which is lo子但oking to challenge the company說道's most powerful AI w樂遠orkload offering with its M1300X 但農chip.

What's next?

As NVIDIA shares have tripled this y家明ear and the company's m城舞arket valuation has i從計ncreased by more than $700 billion, 好機making it the first chip compan信市y to be worth trillions 知來of dollars, investors are場慢 expecting the chip-體又designer to forecast higher-than-expe制高cted quarterly revenue when it report動綠s results on Aug. 23rd.

NVIDIA's second-quarter earnings w雨河ill be the biggest test of the ar要物tificial intelligence hype cycl在議e. "What NVIDIA reports in its冷術 upcoming earnings re哥就port will be a barometer f道物or the entire AI hype," Forrester ana舊討lyst Glenn O'Donnell to時友ld Yahoo. "I expect the results to b身如e excellent because deman但票d is so high, which mea章一ns NVIDIA can make higher pro鐘樂fits than other companies."

On the supply side, NVIDIA wo鄉吧n't be the only supplier, but門煙 what puts the company in a 廠會better position than its他問 rivals is its leading position an大靜d patent-protected technology, which ha答制s been proposed at an early st河開age. Some market experts believe that關兒 AMD is the "only viable a信計lternative" to Nvidia's AI民小 chips, at least for the time be光志ing.