AI NewsAn exclusive tour of Amazon’s Trainium lab, the chip that’s won over Anthropic, OpenAI, even Apple

An exclusive tour of Amazon’s Trainium lab, the chip that’s won over Anthropic, OpenAI, even Apple

9:17 PM IST · March 22, 2026

An exclusive tour of Amazon’s Trainium lab, the chip that’s won over Anthropic, OpenAI, even Apple

Shortly after Amazon CEO Andy Jassy announced AWS’sgroundbreaking $50 billion investment dealwith OpenAI, Amazon invited me on a private tour of the chip development lab at the heart of the deal, at (mostly*) its own expense. Industry experts are watchingAmazon’s Trainium chip, created at that facility, for its implications for lower-cost AI inference and, potentially, a dent in Nvidia’s near monopoly. Curious, I agreed to go. My tour guides for the day were the lab’s director, Kristopher King (pictured below right) and director of engineering Mark Carroll (below left), as well as the team’s PR person who arranged the visit, Doron Aronson (pictured with yours truly later in the story). AWS has been Anthropic’s major cloud platform since the AI lab’s early days — a relationship significant enough to survive Anthropic later adding Microsoft as a cloud partner as well, and Amazon’s growing partnership with OpenAI. The OpenAI deal makes AWS the exclusive provider of the model maker’s new AI agent builder, Frontier, which could become an important part of OpenAI’s business if agents become as big as Silicon Valley thinks they will. We’ll see if that exclusivity stands exactly as announced. The Financial Timesreported this weekthat Microsoft may believe OpenAI’s deal with Amazon violates its own deal with OpenAI, namely with Redmond getting accessto all of OpenAI’s models and tech. What makes AWS so appealing to OpenAI? As part of this deal, the cloud giant has agreed to supply OpenAI with 2 gigawatts of Trainium computing capacity. This is a giant commitment, given that Anthropic and Amazon’s own Bedrock service are already consuming Trainium chips faster than Amazon can produce them. There are 1.4 million Trainium chips deployed across all three generations, and Anthropic’s Claude runs on over 1 million of the Trainium2 chips deployed, the company said. It’s worth noting that while Trainium was originally geared toward faster, cheaper model training (a bigger priority a couple of years ago), it’s now tuned and used for inference as well. Inference — the process of actually running an AI model to generate responses — is currently the biggest performance bottleneck in the industry. Case in point: Trainium2 handles the majority of the inference traffic onAmazon’s Bedrock service, which supports the building of AI applications by Amazon’s many enterprise customers and allows the apps to use multiple models. “Our customer base is just expanding as fast as we can get capacity out there,” King said. “Bedrock could be as big as EC2 one day,” he added, referring to AWS’s behemoth compute cloud service. Beyond offering an alternative to Nvidia’s backlogged, hard-to-acquire GPUs, Amazonsaysits new chips running on its new specialty Trn3 UltraServers cost up to 50% less to run for comparable performance than using classic cloud servers. Along withTrainium3, released in December, this AWS team also built new Neuron switches, and Carroll says that combo is transformative. “What that gives us is something huge,” Carroll said. The switches allow every Trainium3 chip to talk to every other chip in a mesh configuration, reducing latency. “That’s why Trainium3 is breaking all kinds of records,” particularly in “price per power,” he said. When trillions of tokens a day are involved, such improvements add up. In fact, Amazon’s chip team waslauded by Apple in 2024. In a rare moment of openness for the secretive company, Apple’s director of AI publicly described how it used another of the team’s chips — Graviton, a low-power, ARM-based server CPU and the first breakout chip this team designed. Apple also lauded Inferentia — a chip specifically designed for inference — and gave a nod to Trainium, which was new at the time. These chips represent the classic Amazon playbook: See what people want to buy, then build an in-house alternative that competes on price. The catch for chips, historically, has been switching costs. Applications written for Nvidia’s chips must be re-architected to work with others — a time-consuming process that discourages developers from switching. But the AWS chip team proudly told me that Trainium now supportsPyTorch, a popular open source framework for building AI models. That includes many of the ones hosted on Hugging Face, a vast library where developers share open source models. The transition, Carroll told me, requires “basically a one-line change, and then recompile, and then run on Trainium.” In other words, Amazon is attempting to chip away at Nvidia’s market dominance wherever possible. AWS has also this month announced apartnership with Cerebras Systems, integrating that company’s inference chip on servers running Trainium for what Amazon promises will be superpowered, low-latency AI performance. But Amazon’s ambitions go beyond the chips themselves. It also designs the server that hosts the chips. Besides the networking components, this team has designed “Nitro,” a hardware-software combo that provides virtualization tech (which allows many instances of software to run separately on the same server); new state-of-the-art liquid cooling technology; and the server sleds (pictured below) that host this gear. All of that is to control cost and performance. Amazon’s custom chip-designing unit was born when the cloud giantbought Israeli chip designer Annapurna Labsin January 2015 for about $350 million. So this team has now had more than 10 years designing chips for AWS. The unit has retained its Annapurna roots and name — its logo is everywhere in the office. This chip lab is located in a shiny, chrome-windowed building in Austin’s upscale “The Domain” district, a walkable area filled with shops and restaurants that’s sometimes calledAustin’s Silicon Valley. The offices have your classic tech corporate vibe: desks in cubicles, gathering spots, and conference rooms. But tucked away at the back of a high floor in the building is the actual lab, with sweeping views of the city. The shelving-filled lab, about the size of two large conference rooms, is a noisy industrial space thanks to the fans on the equipment. It looks like a cross between a high school shop class and a Hollywood set for a high-end lab, except the engineers are dressed in jeans, not white lab coats. Note that this is not where the chips are manufactured, so no white hazmat suits were necessary. The Trainium3 is a state-of-the-art 3-nanometer chip, produced by TSMC, arguably the leader in 3-nanometer manufacturing, with other chips produced by Marvell. But this is the room where the magic of the “bring-up” occurs. “A silicon bring-up is when you get the chip for the first time, and it’s like a big overnight party. You stay here, like a lock-in,” King explains. After 18 months of work, the chip is activated for the first time to verify it works as designed. The team even filmed some of the Trainium3 bring-upand posted it on YouTube. Spoiler alert: It’s never problem-free. For Trainium3, the prototype chip was originally air-cooled, like previous versions. The current chip is nowliquid-cooled,which offers energy advantages and was quite an engineering feat. During the bring-up, the dimensions for how the chip attached to the air-cooling heat sink were off, so the chip couldn’t be activated. Unfazed, the team “immediately got a grinder and just started grinding off the metal,” King said. Because they didn’t want the noise disrupting the bring-up pizza party atmosphere, they snuck off and did the grinding in a conference room. Staying up all night and solving problems “is what silicon bring-up is all about,” King said. The lab even has a welding station, where hardware lab engineer and master welder Isaac Guevara demonstrated welding tiny integrated circuit components through a microscope. This is such insanely difficult work that senior leader Carroll openly admitted he couldn’t do it, to the guffaws of Guevara and the rest of the engineers in the room. The lab also contains both custom-made and commercial tools for testing and analyzing issues with chips. Here’s signal engineer Arvind Srinivasan demonstrating how the lab tests each tiny component on the chip: But the star of the lab is an entire row showcasing each generation of the “sleds” the team designed. Sleds are the trays that house the Trainium AI chips, Graviton CPU chips, and supporting boards and components. Stack them together on a rack with the networking component, also custom-designed by this team, and you get the systems that are at the heart of Anthropic Claude’s success. Here’s the sled that was shown off during the AWS re:invent conference in December: I expected my guides to crow about the OpenAI deal during the tour. But they didn’t. The reticence could have been related to the aforementioned potential legal haze that might hang over the deal. But the sense I got was that these boots-on-the-ground engineers (who are currently designing the next version, Trainium4) haven’t had much chance to work with OpenAI yet. Their day-to-day work has so far been focused on Anthropic’s and Amazon’s needs. Currently, the biggest chunk of Trainium2 chips is deployed in Project Rainier — one of the world’s largest AI compute clusters — which went live in late 2025 with 500,000 chips. It’s used by Anthropic. But there was a wall monitor in the main office displaying a quote about how OpenAI will be using Trainium. The pride was there, if subtle. In addition to this lab, the team also has its own private data center for quality and testing purposes. A short drive away, it doesn’t run customer workloads, so it’s housed at a co-location facility, not an AWS data center. Security is tight: There are strict protocols to enter the building and to access Amazon’s area within. The data center’s cooling system is so loud that earplugs are mandatory, and the air is thick with the acrid smell of heated metal. It’s not a pleasant place for the average person to hang out. At this data center, there are rows and rows of servers filled with sleds that integrate all of Amazon’s newest custom chips: Graviton CPU, liquid-cooled Trainium3, Amazon Nitro, all happily computing away. The liquid runs on a closed system, meaning it is reused, which should also help reduce the environmental impact, the engineers said. Here’s what a current Trn3 UltraServer looks like: Multiple sleds are on top and bottom, with the Neuron switches in the middle. Hardware development engineer David Martinez-Darrow is seen here performing maintenance on a sled: While attention on the team has always been high, the scrutiny has really ratcheted up as of late. Amazon CEO Andy Jassy keeps a close eye on this lab, publicly bragging about its products like a proud dad. In December, he saidTrainium was already a multibillion-dollar business for AWSandcalled itone piece of AWS tech he’s most excited about. He alsogave the chip a shout-outwhen announcing the OpenAI agreement. The team feels the pressure, too. Engineers will work 24/7 for three to four weeks around each bring-up event to fix any issues so the chips can be mass-produced and put into data centers. “It’s very important that we get as fast as possible to prove that it’s actually going to work,” Carroll said. “So far, we’ve been doing really well.” *Disclosure: Amazon provided airfare and covered the cost of one night at a local hotel. Honoring itsLeadership Principle of Frugality, this was a back-of-the-plane middle seat and a modest room. TechCrunch picked up the other associated travel costs like Ubers and luggage fees. (Yes, I checked a bag for an overnight trip. I’m high maintenance that way.)

read more

Latest AI News

View All News →
There aren’t enough rockets for space data centers — Cowboy Space raised $275M to build them

There aren’t enough rockets for space data centers — Cowboy Space raised $275M to build them

The apparently insatiable demand for AI compute has data center entrepreneurs looking to the stars. There’s a key problem: There aren’t enough rockets to put data centers in orbit around Earth, and they’re too expensive. Most of the players are hoping that SpaceX’s Starship — expected to make its twelfth test flight as soon as this weekend— will solve the problem. But once the vehicle is operational it may be years before it is commercially available, given SpaceX’s internal satellite business. Thesame is truefor Blue Origin’s New Glenn rocket, which failed to deliver a satellite during its third launch in April. That leaves space data center schemes either targeting the mid 2030s, like Google Suncatcher, or preparing to start off doing edge processing tasks for space sensors, likeStarcloud. In theory, there’s a third way: “We’re standing up our own rocket program,” Baiju Bhatt, the CEO and founder of Cowboy Space Corporation, told TechCrunch. He expects the first launch before the end of 2028. Today, the company announced the closure of a $275 million Series B round at a post-money valuation of $2 billion, led by earlier backer Index Ventures, as a downpayment on that work. Breakthrough Energy Ventures, Construct Capital, IVP, and SAIC also participated. The company had previously raised $80 million from investors, including Index, Breakthrough Energy Ventures, Andreessen Horowitz, and New Enterprise Associates. Bhatt, a co-founder of online stock platform Robinhood, launched this startup in 2024 as Aetherflux, with plans to collect abundant solar energy in space and beam it down to Earth. The idea of space data centers led the company to pivot towards using its electricity while in orbit. Thepractical realitiesof that effort, in turn, led him to a rocket development program, and the company’s new name. Bhatt said he spoke to multiple launch providers to try and find a path where his company would only build satellites, but he couldn’t find enough launch capacity to truly scale an orbital data center business, or do so in a way where the unit economics could compete with terrestrial alternatives. "There's a lot of new rockets that are coming online, but as we look three, four years out, it's still very, very scarce, and I think that you're going to see a lot of the first party rocket providers actually specialize into their own payloads," Bhatt said. Of course, while bringing the rocket in-house is logical, it's also nuts. Only a handful of private companies in the West, mainly SpaceX, Rocket Lab and Arianespace, are consistently launching commercial rockets. Two others, Blue Origin and United Launch Alliance, have been struggling to drag their vehicles out of development hell for years. A number of startups, including Stoke Space, Firefly Aerospace, and Relativity Space, have worked for years and are still waiting to deliver operational systems. This evolution of the company will also bring Cowboy Space Corporation into direct competition with SpaceX and Blue Origin, the most advanced and well-funded players in the market. "The prize here, and the size of this market, is big enough that there's room for many players to succeed," Bhatt said "I see the demand for AI getting more and more acute, and I see the options on Earth getting more and more limited." One advantage, Bhatt argues, is the company's focus on this single market (data centers), and its unique design. Orbital rockets typically have a booster stage that flies the vehicle to the edge of space, and a second stage that carries the payload and delivers it to orbit. Cowboy Space plans to build its data centers directly into the second stage of its rocket. It's actually a bit of a throw-back: The first US satellite, Explorer 1, was built as the final stage of a rocket, filled with radio equipment and a few scientific instruments. Making the rocket purpose-built only to launch its data-center satellites should simplify the design process. The company expects each satellite to have a mass of 20,000 to 25,000 kilograms and to generate 1 MW of power for just under 800 onboard GPUs. That means its rocket would be slightly more powerful than the SpaceX's workhorse Falcon 9, though still smaller than its under-development Starship. Eventually, Bhatt says, he expects the booster to be reusable. Cowboy Space has hired veterans of the space industry, including former Blue Origin propulsion engineer Warren Lamont and former SpaceX launch director Tyler Grinne. The company also plans to build its own rocket engine, the most complex and expensive part of any launch vehicle. Cowboy Space is still working through key development needs, like facilities to test, manufacture and launch its rockets. The new vision comes with a new name for the startup, to emphasize its mission to "power humanity from the high frontier," although Bhatt admits "it gives me a reason to wear a cowboy hat and also grow this sick mustache."

5 hours ago

View

Digg tries again, this time as an AI news aggregator

Digg tries again, this time as an AI news aggregator

Digg is back from the dead. Again. Just months afterlaunching, therebootof Kevin Rose’s once-popular link-sharing siteshut down in March, as the company shifted course. Originally redesigned as acompetitorto the massive community forum site Reddit, the new Digg found that it wasn’t able to effectively manage the bot traffic invading its platform and hadn’t differentiated itself enough from the competition to make an impact. The startup laid off staff and said it was time to go back to the drawing board. Rose, a partner at True Ventures, returned to work full-time on a new version of Digg in April. On Friday evening, the founderprevieweda link to the newly redesignedDigg, which now looks nothing like a Reddit clone and more like the news aggregator it once was. a little project i've been hacking on:https://t.co/zTuwWy44lybugs expected. more topics soon. This time around, the site is focused on ranking news — specifically, AI news to start. In an email to beta testers, the company said the site’s goal is to “track the most influential voices in a space” and to surface the news that’s actually worth “paying attention to.” AI is the area it’s testing this idea with, but if successful, Digg will expand to include other topics. The email warned that the site was still raw and “buggy,” and was designed more to give users a first look than to serve as its public debut. On the current homepage, Digg showcases four main stories at the top: the most viewed story, a story seeing rising discussion, the fastest-climbing story, and one “In case you missed it” headline. Below that is a ranked list of top stories for the day, complete with engagement metrics like views, comments, likes, and saves. But the twist is that these metrics aren’t the ones generated on Digg itself. Instead, Digg is ingesting content from X in real-time to determine what’s being discussed, while also performing sentiment analysis, clustering, and signal detection to determine what matters most. As Roseremarked on X,when OpenAI CEO Sam Altman engages with a story about AI, it almost always sets off a chain reaction that includes deep discussion and propagation of that topic throughout X. The new Digg will be able to track that increased engagement. This might be something that’s interesting to data nerds, as it exposes the impact of X-based engagement with charts and graphs, and offers a way to track signal among what can, on X, often be a lot of noise. But it’s unclear whether there’s enough underlying value here for an everyday user, beyond seeing that yes, a@samatweet can make something go viral. The site also ranks the top 1,000 people involved in AI, as well as the top companies and the top politicians focused on AI issues. For those who don’t have time to spend on X tracking breaking AI news, Digg could prove a useful resource. But it’s not clear why people would regularly turn to Digg over their preferred news app, RSS reader, or even their X “For You” feed, if they wanted to catch up on what’s trending — especially because there isn’t currently any discussion happening on Digg’s site itself. Digg may also struggle when it moves on to other topics, as AI news is one of the few areas where discussion still heavily takes place on X. Other verticals don’t have the same traction, especially after Musk’s takeover of the site formerly known as Twitter gave rise to an ecosystem of competitors, which now includes Meta’s creator-focused Threads. Many non-tech-related discussions are now happening off X, or off the public internet entirely. However, if Digg does end up gaining steam, it could serve as a useful source of website traffic to publishers whose businesses have been decimated bydeclining clicksthanks to Google’s changing algorithms andthe impact of AI Overviews, the AI-generated summaries Google displays atop search results, which often answer users’ questions before they ever click through to a website.

5 hours ago

View

OpenAI Launches $4 Billion Deployment Company Backed by Top Investors

OpenAI Launches $4 Billion Deployment Company Backed by Top Investors

The Deployment Company will function as a standalone business unit while remaining closely connected to OpenAI’s research and product teams.

5 hours ago

View

Delhi-NCR Raises $1.7 Billion in Q1 2026 Driven by Large Funding Rounds

Delhi-NCR Raises $1.7 Billion in Q1 2026 Driven by Large Funding Rounds

Gurugram dominated the funding landscape, attracting 52% of all capital raised in the region, equivalent to $876 million.

9 hours ago

View