在 GTC 大会上,作为分别站在硬件和算法顶端的人,英伟达首席科学家 Bill Dally 与谷歌首席科学家 Jeff Dean 的一场对话,向外界释放了一个明确的信号。 AI 的竞争重点变了。 单看模型参数、算法创新,已经不够。硬件、系统、能耗、推理速度,这些曾经属于基础设施的东西,正在成为决定性因素。 为什么这些会成为关键?今天的数据中心里,接近 90% 的电力花在推理上。做一次计算只需要几个...
在 GTC 大会上,作为分别站在硬件和算法顶端的人,英伟达首席科学家 Bill Dally 与谷歌首席科学家 Jeff Dean 的一场对话,向外界释放了一个明确的信号。 AI 的竞争重点变了。 单看模型参数、算法创新,已经不够。硬件、系统、能耗、推理速度,这些曾经属于基础设施的东西,正在成为决定性因素。 为什么这些会成为关键?今天的数据中心里,接近 90% 的电力花在推理上。做一次计算只需要几个飞焦耳,但把数据从内存搬过来,能耗高出上千倍。 而要让 AI 持续高效地运转,接下来比的是全栈能力有多扎实,以及把技术变成生产力的执行能力。 这场竞争,刚刚开始。 第一节|从问答到干活,AI 的角色变了 几年前,大模型能解对初中数学题就让人兴奋,但现在它跨过了一个真正的门槛。 过去的工作方式是“你问一句它答一句”,做完一步就停在原地。而现在,你只需要用自然语言交给它一个目标,它会自己拆解步骤、反复尝试、修正错误,并继续往下推进。它的工作时间尺度,也从几秒钟延长到了几个小时甚至几天。 不仅如此,AI 甚至开始能够改进自身的系统。 Jeff Dean 提到,你可以给模型下达一个指令,比如“请在这个方向上探索能提高性能的想法”。它会自己去跑 50 个实验,筛掉 40 个不行的,锁定 10 个有希望的,继续做后续测试。 以前要实现这种探索,你得写长串的代码告诉它去哪找、怎么找;现在,方向定好,它自己就会执行。 这背后的实质,是 AI 在不断地进行推理:接任务、出结果、再接下一个任务。跑 50 个实验,每个实验都要推理;持续迭代,每次迭代都要推理。推理速度决定了这些任务能跑多快。 而推理速度的致命瓶颈,在于延迟。 Bill Dally 指出,延迟越低,每一步等待越短,任务推进越快;延迟越高,AI 就会停滞,哪怕模型再强也无法持续工作。 训练阶段决定了模型“能学到什么”,但真正决定它“能干多少活”的,是推理阶段的效率。因此,衡量 AI 的核心标准也跟着变了:不再是模型准确率提升了多少,而是它一天到底能完成多少工作。 AI 正式从“辅助决策”进入“直接执行”阶段。 限制它的不再是模型本身的智商,而是算力、延迟、能耗这些支撑它运行的基础设施,以及背后把技术转化为生产力的投入能力。 第二节|90% 的电力,都花在让 AI 干活 AI 能力变了,钱的流向也跟着变了。从不计代价的“打地基”,变成...
It urges the court to set limits on how it can be invoked, saying it was not meant to be "used to distort or annihilate the rights and freedoms guaranteed by the Charter", or to reduce them to "des peaux de chagrin", that is, to shrivel them beyond recognition.
It urges the court to set limits on how it can be invoked, saying it was not meant to be "used to distort or annihilate the rights and freedoms guaranteed by the Charter", or to reduce them to "des peaux de chagrin", that is, to shrivel them beyond recognition.
imaginima/iStock via Getty Images What is interesting to me about AMD ( AMD ) is no longer the fact that they have become a legitimate competitor to AI technology which is obvious at this point but the way the company is perceived. The question is no longer how well Instinct GPUs sell, but whether AMD can deliver complete systems, secure its supply chain and translate that into sustained earnings ...
imaginima/iStock via Getty Images What is interesting to me about AMD ( AMD ) is no longer the fact that they have become a legitimate competitor to AI technology which is obvious at this point but the way the company is perceived. The question is no longer how well Instinct GPUs sell, but whether AMD can deliver complete systems, secure its supply chain and translate that into sustained earnings growth. We know the financials have already reflected the fact that the company is no longer simply a secondary player to AI technology. We know the stock is not euphoric and that the stock is sitting right in the middle of the expectations curve, where expectations are high but not so high as to expect perfection. To me, the next phase is no longer about how the chip performs relative to others but how the system performs relative to others. The Market Missed the Real Shift The most important event occurred on March 18 as AMD and Samsung ( SSNLF ) entered into an MOU which makes Samsung an aligned primary HBM4 supplier for the MI455X and a DDR5 partner for 6th Gen EPYC Venice, as well as a discussion for foundry services. While I do not think that memoranda should be considered comparable to revenue, this is an important event because it addresses a constraint that investors think of as a footnote. The constraint for infrastructure spend in the AI space is not just the acceleration device, it is the memory, the packaging, the thermal design, the interconnect and the timing. Samsung HBM4 is already described as being mass production ready, with speeds up to 13 Gbps and 3.3 TB/s bandwidth which tells me that AMD is attempting to de-risk supply and system readiness before the launch of their Helios. That is the behavior of a company that is preparing for platform wars, not just a company attempting to win a few benchmark wars. My mind always comes back to the simple fact. MI300 and MI350 made AMD relevant. Helios and Venice will determine if AMD becomes a harder competitor to...
imaginima/iStock via Getty Images What is interesting to me about AMD ( AMD ) is no longer the fact that they have become a legitimate competitor to AI technology which is obvious at this point but the way the company is perceived. The question is no longer how well Instinct GPUs sell, but whether AMD can deliver complete systems, secure its supply chain and translate that into sustained earnings ...
imaginima/iStock via Getty Images What is interesting to me about AMD ( AMD ) is no longer the fact that they have become a legitimate competitor to AI technology which is obvious at this point but the way the company is perceived. The question is no longer how well Instinct GPUs sell, but whether AMD can deliver complete systems, secure its supply chain and translate that into sustained earnings growth. We know the financials have already reflected the fact that the company is no longer simply a secondary player to AI technology. We know the stock is not euphoric and that the stock is sitting right in the middle of the expectations curve, where expectations are high but not so high as to expect perfection. To me, the next phase is no longer about how the chip performs relative to others but how the system performs relative to others. The Market Missed the Real Shift The most important event occurred on March 18 as AMD and Samsung ( SSNLF ) entered into an MOU which makes Samsung an aligned primary HBM4 supplier for the MI455X and a DDR5 partner for 6th Gen EPYC Venice, as well as a discussion for foundry services. While I do not think that memoranda should be considered comparable to revenue, this is an important event because it addresses a constraint that investors think of as a footnote. The constraint for infrastructure spend in the AI space is not just the acceleration device, it is the memory, the packaging, the thermal design, the interconnect and the timing. Samsung HBM4 is already described as being mass production ready, with speeds up to 13 Gbps and 3.3 TB/s bandwidth which tells me that AMD is attempting to de-risk supply and system readiness before the launch of their Helios. That is the behavior of a company that is preparing for platform wars, not just a company attempting to win a few benchmark wars. My mind always comes back to the simple fact. MI300 and MI350 made AMD relevant. Helios and Venice will determine if AMD becomes a harder competitor to...
Key Points Navitas saw its stock price rise 376% at its peak last year. However, analysts see the stock price receding over the next 12 months. IBM stock has great upside and is positioned to capitalize on the AI boom and quantum computing. 10 stocks we like better than International Business Machines › Navitas Semiconductor got a lot of attention last year as its stock price surged some 376% for ...
Key Points Navitas saw its stock price rise 376% at its peak last year. However, analysts see the stock price receding over the next 12 months. IBM stock has great upside and is positioned to capitalize on the AI boom and quantum computing. 10 stocks we like better than International Business Machines › Navitas Semiconductor got a lot of attention last year as its stock price surged some 376% for the year to more than $17 per share in late October. The chipmaker's stock price has fallen back to roughly $9 per share as of March 19, but it is still up 23% year to date and 250% over the past 12 months. Will AI create the world's first trillionaire? Our team just released a report on the one little-known company, called an "Indispensable Monopoly" providing the critical technology Nvidia and Intel both need. Continue » Navitas' meteoric rise was fueled by several factors. One of the major catalysts was a new partnership with Nvidia to supply it with its gallium nitride (GaN) and silicon carbide (SiC) chips for AI data centers. These Navitas chips are considered faster and more efficient than traditional silicon wafers and will be used in Nvidia's next-generation data center architecture, starting in 2027. Also, Navitas is pivoting from providing chips for consumer markets -- smart phones, PC, and electronics -- to bigger power markets, like data centers, electric vehicles, and industrial. Analysts expect to see revenue decline this year, due to the pivot, but bounce back in 2027 when the Nvidia contract kicks in. Navitas stock has a median price target of $8 per share, which would suggest a 9% decline in the stock price. While Navitas stock could certainly be a stellar long-term option, it is still not consistently profitable and faces uncertainties with its pivot. A more cautious investor may want to consider a less volatile AI stock, IBM (NYSE: IBM). IBM pivots toward AI and quantum computing IBM has made the transition from a computer hardware company to an AI powerh...
Nvidia (NVDA 3.17%) has proven its ability to deliver spectacular returns to investors, soaring more than 1,200% over the past five years. The reason for such gains is simple: In just a few short years, Nvidia has constructed an AI empire, selling a wide range of AI products, including the crown jewel, its graphics processing units (GPUs). The company, thanks to the speed and overall efficiency of...
Nvidia (NVDA 3.17%) has proven its ability to deliver spectacular returns to investors, soaring more than 1,200% over the past five years. The reason for such gains is simple: In just a few short years, Nvidia has constructed an AI empire, selling a wide range of AI products, including the crown jewel, its graphics processing units (GPUs). The company, thanks to the speed and overall efficiency of these chips, became the AI chip leader early on, and this continues as Nvidia launches innovation after innovation. All of this pushed earnings and the stock price to record levels, and valuation reached peak levels too. But, in recent times, general geopolitical and economic uncertainty, as well as certain concerns about the AI market, have weighed on Nvidia stock. And today, it trades at a surprisingly low level, at just 21x forward earnings estimates. Is the world's biggest AI stock actually a value play? Let's find out. Nvidia's focus on AI First, let's take a step back and consider Nvidia's position in the AI market today and what may be on the horizon. As mentioned, Nvidia has become a market giant. The company, originally focused on serving chips to the gaming market, began to focus on AI about a decade ago. Nvidia chief Jensen Huang saw the opportunity and pledged to design GPUs specifically to power AI. Huang's bet clearly was a wise one, as it allowed Nvidia to get in on this market before others and establish its leadership. The company also promised to update its chips on an annual basis to keep this prized position, and it's followed through on this: Over the past year and a half, it's launched Blackwell and Blackwell Ultra, and it's on track to release the Vera Rubin system later this year. Customers, led by tech giants such as Meta Platforms and Amazon, have flocked to Nvidia for chips and systems, helping earnings skyrocket. In the latest full year, revenue reached a record $215 billion and net income hit $120 billion. In previous years, Nvidia's chips were...
If you are looking for income stocks with high yields that can help set you up with a lifetime of reliable dividends, you'll want to focus on the businesses that back the yields. With yields of more than 5%, Realty Income (O 2.70%), Enterprise Products Partners (EPD +0.29%), and Verizon (VZ +1.28%) are all worth a deep dive today. 1. Realty Income is The Monthly Dividend Company Realty Income trad...
If you are looking for income stocks with high yields that can help set you up with a lifetime of reliable dividends, you'll want to focus on the businesses that back the yields. With yields of more than 5%, Realty Income (O 2.70%), Enterprise Products Partners (EPD +0.29%), and Verizon (VZ +1.28%) are all worth a deep dive today. 1. Realty Income is The Monthly Dividend Company Realty Income trademarked the nickname "The Monthly Dividend Company" to highlight the frequency of its dividend and, perhaps more notably, the importance of dividends to the company. It is built from the ground up to be reliable, with over three decades of annual dividend increases already in the books. The real estate investment trust (REIT) has an investment-grade credit rating, indicating a strong financial foundation. But that's just the starting point. It owns over 15,500 properties across the United States and Europe. It has exposure to retail and industrial assets, as well as a selection of more unique properties, like vineyards, casinos, and data centers. And the company's average lease length is 8.8 years, which provides stability to the rent roll if there is a recession. Even the most conservative investors will appreciate Realty Income and its attractive 5.1% dividend yield. Expand NYSE : O Realty Income Today's Change ( -2.70 %) $ -1.69 Current Price $ 60.95 Key Data Points Market Cap $57B Day's Range $ 60.63 - $ 62.60 52wk Range $ 50.71 - $ 67.94 Volume 414K Avg Vol 6.6M Gross Margin 48.73 % Dividend Yield 5.30 % 2. Enterprise Products Partners sidesteps commodity prices Enterprise Products Partners' 5.8% yield is supported by an energy business, which might worry some investors amid rising geopolitical tension in the Middle East. That's less of a worry than you may think because Enterprise's business is to move oil and natural gas around the world, collecting fees for the use of its vital North American energy infrastructure assets. The volume of energy moving through Enterpri...