 |
|
|
|||||
So far, Nvidia has provided the vast majority of the processors used to train and operate large AI models like the ones that underpin ChatGPT. Tech companies and AI labs dont like to rely too much on a single chip vendor, especially as their need for computing capacity increases, so theyre looking for ways to diversify. And so players like AMD and Huawei, as well as hyperscalers like Google and Amazon AWS, which just released its latest Trainium3 chip, are hurrying to improve their own flavors of AI accelerators, the processors designed to speed up specific types of computing tasks. Could the competition eventually reduce Nvidia, AIs dominant player, to just another AI chip vendor, one of many options, potentially shaking up the industrys technological foundations? Or is the rising tide of demand for AI chips big enough to lift all boats? Those are the trillion-dollar questions. Google sent a minor shockwave across the industry when it casually mentioned that it had trained its impressive new Gemini 3 Pro model entirely on its own Tensor Processing Units (TPUs)another flavor of AI accelerator chips (GPUs). Industry observers immediately wondered if the AI industrys broad dependence on Nvidia chips was justified. After all, theyre very expensive: A big part of the billions now being spent to build out AI computing capacity (data centers) is going to Nvidia chips. And Google TPUs are looking more like a Nvidia alternative. The company can rent TPUs in its own data centers, and its reportedly considering selling the chips outright to other AI companies, including Meta and Anthropic. A (paywalled) report from The Information in November said Google is in talks to sell or lease its GPUs so they can run in any companys data center. A Reuters report says Meta is in talks to spend billions on Googles TPUs starting in 2027, and may begin paying to run AI workloads on TPUs within Google data centers even sooner. Anthropic announced in October that it would use up to a million TPUs within Google data centers to develop its Claude models. Selling the TPUs outright would, technically, put Google in direct competition with Nvidia. But that doesnt mean that Google is gunning hard to steal Nvidias chip business. Google, after all, is a major buyer of Nvidia chips. Google may see selling TPUs to certain customers as an extension of selling access to TPUs running in its cloud. This makes sense if said customers are looking to do the types of AI processing that TPUs are especially good at, says IDC analyst Brandon Hoff. While Nvidias GPUs are workhorses capable of a wide range of work, most of the big-tech platform companies have designed their own accelerators that are purpose-built for their most crucial types of computing. Microsoft developed chips that are optimized for its Azure cloud services. Amazons Trainium chips are especially good at e-commerce-related tasks like product suggestion and delivery logistics. Googles TPUs are good at serving targeted ads across its platforms and networks. Thats something Google shares with Meta. They both do ads and so it makes sense that Meta wants to take a look at using Google’s TPUs, Hoff says. And its not just Meta. Most big tech companies use a variety of accelerators because they use machine learning and AI for a wide variety of tasks. Apple got some TPUs, got some of the AWS chips, of course got some GPUs, and they’ve been playing with what works good for different workloads, he adds. Nvidias big advantage has been that its chips are very powerfultheyre the reason that training large language models became possible. Theyre also great generalists, good for a wide variety of AI workloads. On top of that, theyre flexible, which is to say they can plug in to different platforms. For example, if a company wants to run its AI models on a mix of cloud services, theyre likely to develop those models to run on Nvidia chips because all the clouds use them. Nvidias flexibility advantage is a real thing; its not an accident that the fungibility of GPUs across workloads was focused on as a justification for increased capital expenditures by both Microsoft and Meta, analyst Ben Thompson wrote in a recent newsletter. TPUs are more specialized at the hardware level, and more difficult to program for at the software level; to that end, to the extent that customers care about flexibility, then Nvidia remains the obvious choice. However, vendor lock-in remains a big concern, especially as big tech companies and AI labs are sinking hundreds of billions of dollars into new data center capacity for AI. AI companies would prefer instead to use a mix of AI chips from different vendors. Anthropic, for one, is explicit about this: Anthropics unique compute strategy focuses on a diversified approach that efficiently uses three chip platformsGoogles TPUs, Amazons Trainium, and NVIDIAs GPUs, the company said in an October blog post. Amazons AWS says its Trainium3 chip is roughly four times faster than the Trainium2 chip it announced a year ago, and 40% more efficient. Because of the performance of Nvidia chips, many AI companies have standardized on CUDA, the Nvidia software layer that lets developers control how the GPUs work together to support their AI applications. Most of the engineers, developers, and researchers who work with large AI models know CUDA, which can cause another form of skills-based organizational lock-in. But now it may make sense for organizations to build whole new alternative software stacks to accommodate different kinds of chips, Thompson says. That they did not do so for a long time is a function of it simply not being worth the time and trouble; when capital expenditure plans reach the hundreds of billions of dollars, however, what is worth the time and trouble changes. IDC projects that the high demand for AI computing power isnt likely to abate very soon. We see that cloud service providers are growing quickly, but their spending will slow down, Hoff says. Beyond that, a second wave of demand may come from sovereign funds, such as Saudi Arabia, which is building the Humain AI hub, a large AI infrastructure complex that it will fund and control. Another wave of demand could come from large multinational corporations that want to build similar sovereign AI infrastructure, Hoff explains. There’s a lot of stuff in 2027 and 2028 that’ll keep driving demand. There are plenty of chipmaker challenges Nvidia stories out there, but the deeper one delves into the economic complexities and competitive dynamics of the AI chip market, much of the drama drains away. As AI finds more applications in both business and consumer tech, AI models will be asked to do more and more kinds of work, and each one will demand various mixtures of generalist or specialized chips. So while there is growing competitive pressure on Nvidia, theres still a lot ofgood reasons for players like Google and Amazon to collaborate with Nvidia. In the next two years, there is more demand than supply so almost none of that matters, says Moor Insights & Strategy chief analyst Patrick Moorhead. Moorhead believes that five years from now Nvidia GPUs will still retain their 70% market share.
Category:
E-Commerce
On November 19, Block Inc. held its first Investor Day in three years. Jack Dorsey, the company’s cofounder, chief executive, and “Block Head,” took to the stage and summarily posed what many investors and others in the industry were likely thinking. Our business is complicated, he said. We want to make it much easier to understand going forward. Dorseynotably clean-shavenproceeded to summarize the past few years at Block. The company is indeed much more complex now than when it was founded in 2009 as Square, named for the point-of-sale system that was the companys first product. Four years ago, it changed its name to Block, a much more fitting moniker given its increasingly multidimensional portfolio, which now includes not only Square but also Cash App, Afterpay, Tidal, Bitkey, and Proto. For Oakland, California-based Block, the growing pains were real as it has evolved from a single-product company to one that now facilitates payments (and buy-now-pay-later features) for both customers and merchants, has its hands in the crypto space, and even offers a streaming platform for musicians and creators. The numbers bear it out: After going public in 2015, Block saw its stock price peak in 2021 at more than $270. Like many other tech companies, Block has seen its shares fall from their pandemic-era highs. The stock is down roughly 26% in 2025 and the company fell short of Wall Street’s projections for its third-quarter earnings in November. But Block has been making some behind-the-scenes moves over the past few years to right the ship. A major philosophical change, key acquisitions, and a renewed focus on simplicity have Blocks leadership excited about the companys future. window.addEventListener("message",function(a){if(void 0!==a.data["datawrapper-height"]){var e=document.querySelectorAll("iframe");for(var t in a.data["datawrapper-height"])for(var r,i=0;r=e[i];i++)if(r.contentWindow===a.source){var d=a.data["datawrapper-height"][t]+"px";r.style.height=d}}}); A lesson in shape-shifting Changing from a Square to a Block required fundamental organizational and philosophical shifts, which have been the most important aspect of Blocks evolution. We decided to functionalize the company, says Owen Jennings, Blocks business lead. That meant making big internal shuffles and reorganizing how information moved between engineers, designers, sales staff, and executives. It also meant putting functional leads into positions where they could be most impactful, whether they were working on product development or sales strategies. We dissolved the business units and brought functional leaders to the top who reported directly to Jack, he says. He adds that the companys multiple business units were siloed and had different goals and models, which were leading to the wrong outcomes. What became evident was that Block needed to find ways to serve both merchants and customersusing its products to either transact (via Cash App) or process payments (via Square). The most obvious [thing] we could bring to the world was connecting the two worlds: consumers and sellers, Jennings says. But it wasnt happening based on the structure we had. Since the reorganization, “it feels like were one massive company,” Jennings adds, but those changes took time to implement. Functionalization happened within 18 months, says Nick Molnar, Blocks sales and marketing lead and the cofounder of Afterpay, who decided to stay with Block when it acquired Afterpay in early 2022. Molnar says that while he is a relative Block newbie (Jennings, by contrast, has been at the firm for more than a decade), he’s seen a notable shift at the company. Meanwhile, most people arent even seeing the full results yet. The back half of this year, youre seeing the work of the previous 18 months,” he says. Blocks leaders have also married the functional model to the Rule of 40, a metric common in the SaaS sector, which says that a companys growth rate and profit margin should sum up to 40%. Amrita Ahuja, Blocks foundational lead, says that prior to instituting the Rule of 40 framework, the company had expected that wed advance margins every yearwe wanted to share the trade-offs behind long-term growth and profitable long-term growth. So we reoriented the company from the inside out,” she added. “That was really a language we built for the company. It helped us move faster and become more efficient, and ensure investments were going to drive growth. The Rule of 40, paired with the new functional model, also allowed Block to reorient its larger focus on simplicitysomething it had gotten away from over the years as its business and structure have grown more complicated and convoluted. window.addEventListener("message",function(a){if(void 0!==a.data["datawrapper-height"]){var e=document.querySelectorAll("iframe");for(var t in a.data["datawrapper-height"])for(var r,i=0;r=e[i];i++)if(r.contentWindow===a.source){var d=a.data["datawrapper-height"][t]+"px";r.style.height=d}}}); Basic building blocks Ahuja first came to the company as Squares CFO in 2019. The thing that was striking to me, working at my first tech company, was the level of trust and transparency, she says. There was so much information, and everybody had access to it. But, naturally, things get more complicated as a company grows, as Square did when it contended with the pandemic and then morphed into Block. Square started with payments, then we built more than 30 products around it, Ahuja says. It was a similar situation with Cash App. The kernel was around social money, peer-to-peer transactions,” she adds. “Now we have built a dozen products around that. Over the years, its become increasingly important to get back to basics and “focus on the things that mean the most to our customers, Ahuja says, adding, “Weve already built a lot of depth and complexitynow its about making sure the right product gets surfaced at the right time for customers. That is exactly what Block is doing now. In recent months, Block has announced several new products, including new tools and features under its Square AI suite, Square Bitcoin, and Neighborhoods, a new feature for Cash App, which connects customers with local businesses. With crypto finding wider adoption and a friendly regulatory framework, and AI being basically everywhere and anywhere, developing and releasing these types of products clearly makes sense for Block. But Jennings says that Dorsey is not merely jumping on trends for the sake of doing so. Hes willing to be patient for a long period of time to the extent that he has conviction, and hes been proven right many times, Jennings says of Dorsey, who is also a cofounder of Twitter and, more recently, the competing social media platform Bluesky, although he’s no longer involved in either. “The power of Jack is that when he comes to the all-hands or presents the company strategy, its incredibly simple,” he adds, “and gets to the essence of what were trying to do.” The big question: Will it all pay off? A chip off the new Block Blocks leaders say that the pieces are in place for a sort of corporate renaissance. I believe that Block has the ingredients it needs to accelerate its growth, that flows through a really strong, profitable business thats growing in line with some of the best companies in the world, says Molnar. So even as it may seem like the companys been underachievingperhaps in terms of sagging stock and recent earnings missesthose on the inside say they are brimming with confidence. Were leading, and will continue to lead, says Jennings, who is particularly confident about Square Bitcoin, which offers no-fee Bitcoin payments for sellers around the world through its existing point-of-sale systems. He thinks Block is well-positioned to take advantage of the growing ubiquity of Bitcoin payments in the years ahead. As for Blocks broader goals? During its Investor Day 2025 presentations, the companys 2026 guidance showed expectations of nearly $12 billion in gross profit, an increase of 17% year-over-year. It also released, for the first time, a three-year financial outlook that lays out what Blocks leadership is expecting, a sign that Block is fully grown up out of its startup stage, and that it’s here for the long haul. By 2028, Block’s outlook shows, the company anticipates gross profit growth will be in the mid-teens, and that adjusted earnings per share growth will be somewhere around 30%, and on track for further revenue growth. That would mark quite a turnaround, but Block executives believe they have the team, product mix, and leadership to persevereeven if it takes some time. Jack is very good at knowing when to be patient and impatient, says Ahuja. From the first day I joined the company, there was a conversation about what his title should be: CEO or editor? Hes the editorhe’s the person who guides us in how we focus our efforts.” During his comments at Investor Day, Dorsey echoed Ahujas sentiment. Ive never felt more confident that we have all the tools, the structure, the team, and the people to prove this out,” he said.
Category:
E-Commerce
Internet infrastructure company Cloudflare on Friday said it was investigating an outage that took place in the morning that brought down several global websites including LinkedIn, Zoom and others, the second such crash to affect the company in less than three weeks.Cloudflare said the issue had been resolved, and that it was was “investigating issues with Cloudflare Dashboard and related APIs,” or application programming interface that allow software systems to communicate with each other.The company said the outage was not due to an attack. A change to how its firewall handles requests “caused Cloudflare’s network to be unavailable for several minutes this morning,” the company said.Users on social media platform X also reported problems accessing the website.Edinburgh airport had to shut down briefly on Friday morning. But the airport later said the outage was a localized issue that was not related to Cloudflare.In November, a Cloudflare outage affected users of everything from ChatGPT and the online game, “League of Legends,” to the New Jersey Transit system.Last month Microsoft had to deploy a fix to address an outage of their Azure cloud portal that left users unable to access Office 365, Minecraft and other services. The tech company wrote on its Azure status page that a configuration change to its Azure infrastructure caused the outage.Amazon also experienced a massive outage of its cloud computing service in October. This version has been updated to reflect that Edinburgh airport says its temporary shutdown was not related to the Cloudflare outage. Associated Press
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||