 |
|
|
|||||
Welcome to AI Decoded, Fast Companys weekly newsletter that breaks down the most important news in the world of AI. Im Mark Sullivan, a senior writer at Fast Company, covering emerging tech, AI, and tech policy. This week, Im focusing on the terms of Nvidias investment in OpenAI, in which the GPU maker gets guaranteed chip sales, an equity stake, and likely a product road map for years to come. I also look at the industrys fixation on huge models and the quiet appeal of small ones. Sign up to receive this newsletter every week via email here. And if you have comments on this issue and/or ideas for future ones, drop me a line at sullivan@fastcompany.com, and follow me on X (formerly Twitter) @thesullivan. Nvidia cements its power as AI infrastructure race begins Now its all about data centers and electricity. Big Tech companies are promising that AI models and apps are about to revolutionize business, and executives like OpenAI CEO Sam Altman say the greatest barrier to that happening is a dearth of data centers to run the models that businesses will soon need to operate. Big Tech companies are also challenged to find enough new energy sources to power and cool the massive data centers. Collectively, OpenAI, Amazon, Google, Meta, and Microsoft plan to spend more than $325 billion on data centers by the end of 2025, The New York Times reports. Anthropic said last year that it expects to spend $100 billion on these massive facilities over the next decade. The tech companies are now racing to plan and finance the new data centers. And this is creating some unique arrangements. Nvidia announced Monday it will invest $100 billion in OpenAI, which will buy about 2% equity in the company. But OpenAI will likely use most of that money to buy Nvidia GPUs, or graphics processing units, the chips that represent the greatest single capital expenditure of building a data center. [T]hese investments might be circular and raise related party concerns, as Nvidia may own shares in a customer that will likely use such funds to buy more Nvidia gear, writes Morningstar equity analyst Brian Colello in a research brief. (OpenAI struck a similar agreement with Microsoft when it took a $10 billion investment from the software giant, then used the money to buy its Azure cloud computing services.) Notably, the Nvidia investment will time the release of the funds according to the pace at which OpenAI buys the chips: Nvidia gets guaranteed chip sales and a 2% share of OpenAI. (As Bryn Talkington, managing partner at Requisite Capital Management, told CNBC: Nvidia invests $100 billion in OpenAI, which then OpenAI turns back and gives it back to Nvidia.). But it may be even better than that. Pitchbook AI and cybersecurity analyst Dimitri Zabelin believes Nvidia intends to plan the design of its future AI chips according to what it learns from OpenAIs infrastructure scale-up. That could be an invaluable feedback loop if all of the big AI companies follow OpenAIs lead in scaling up its infrastructure and developing compute-intensive AI products. Nvidia is consolidating control over the AI stack and reinforcing its position as the indispensable enabler of the sectors next phase, Zabelin says. OpenAI will likely buy between 4 and 5 million of Nvidias new Vera Rubin GPUs, which will require 10 gigawatts of power to run. They will likely be installed within the five new data centers the company just announced as part of its Stargate Project (revealed at the White House with partners SoftBank, Oracle, and MGX). OpenAI now expects that Stargate will secure the full $500 billion in planned investment to build new data centers, and do so by the end of this year, ahead of schedule. Betting big on big modelsnot smaller, safer ones Right now, a huge portion of the total value of the stock market is held up by AI hope, the promise that AI will bring dramatic new efficiencies to the way business is done. Maybe businesses will grow more profitable by moving faster, or maybe theyll do so by sloughing off human workers. Most likely both. The massive infrastructure investments of the Big Tech companies are all about supporting that transformation. The companies building the gigantic data centers are frontier model companies; their products are huge, generalist models, like OpenAIs GPT-5 and Googles Gemini, that have trillions of parameters and are very expensive to train and operate. Generalist models are built to possess a wide array of knowledgeeven a modicum of common sense about how the world worksthat can be leveraged for all kinds of tasks. Theyre trained with massive amounts of diverse data and web content. Its these frontier models that the AI companies hope will evolve to possess artificial general intelligence (AGI), or as much intelligence as most humans bring to most tasks, and then superintelligence, in which the model is far smarter than humans at almost any task. But many of the analysts and researchers Ive spoken to say that businesses usually need smaller models trained with a narrower set of (often proprietary) data that automate a specific set of tasks. They dont need to power their apps with a gigantic (and expensive) model that knows about 15th-century gold coins and can write poetry. Small models often dont need to run inside a dedicated data center, but are small enough to run on on-premise computers (in some cases, laptops or phones) or within a private cloud. With less exposure to wider networks, models that run on the edge devices are far less exposed to would-be hackers that might try to steal or poison corporate or personal data. But OpenAI and Google arent selling that. They offer access to frontier models via application programming interfaces (APIs) to developers and corporations. And its the massive frontier models that carry the greatest risks for society-level harms such as aiding in the building of a bioweapon or crashing economic systems. Some have worried that putting so much intelligence and computing power together in one place could create a supercomputer smart enough to crack open every cryptocurrency wallet on the blockchainwhich would cause economic chaos. Reducing the number of large frontier models (and tightly controlling their use) may be the only rational approach to protecting against the large-scale harms they might, in theory, inflict. Currently, as the big AI companies like OpenAI, Anthropic, Google, and Meta quickly and dramatically scale up their data centers and models, we are trusting them to keep the models from being used for harm. Can private, profit-driven companiessome of which are under great pressure to get to profitabilitycontrol intelligences far greater than our own? Lets hope so. More AI coverage from Fast Company: AI tools arent making much of a difference for companies Are companies calling themselves AI-first helping or hurting their own brands? This is how Gen Zers are AI-proofing their careers I gave ChatGPT $500 of real money to invest in stocks. Its picks surprised me Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.
Category:
E-Commerce
European automakers will save around 500-600 million euros ($585-700 million) a month dating back to Aug. 1 after the Trump administration implemented the U.S. end of its trade deal with the European Union, the EU’s top trade negotiator said Thursday. Trade Commissioner Maros Sefcovic said that the deal establishing a 15% tariff on most EU goods took effect with publication in the U.S. Federal Register instructing customs officials what to charge. That would reduce the tariff from a painfully high 27.5% rate set earlier by Trump. The reduction eases a major burden on EU automakers and is a chief selling point for the deal as presented by European Commission President Ursula von der Leyen. The deal is retroactive to Aug. 1 so “what we expect now is that the tariffs will be returned to the automakers as of the first of August, which is something like 500, 600 million euros per month,” Sefcovic said ahead of a meeting with Southeast Asian trade ministers in Kuala Lumpur, Malaysia. The 15% tariff is still much higher than tariffs from before Trump took office, which averaged in the single digits, and the trade deal has been criticized by business associations and some members of the European Parliament. Sefcovic said most member states supported the deal and that after detailed exchanges and presentations on the details he expected lawmakers would support it as well. He said it was “the best deal available” after difficult talks with Trump administration officials. “Any other alternative would be much worse,” he said. Trump threatened even higher rates during the talks. Associated Press
Category:
E-Commerce
The Barclays Center is taking its dressing rooms for touring artists to the next level, and they looked to local inspiration to decorate them. The Brooklyn venue, which hosts concerts and is home to the NBA’s Brooklyn Nets and WNBA’s New York Liberty, has upgraded six dressing rooms that now resemble Brooklyn brownstone apartments, complete with moody tones, soft-glow lighting, and high-end fixtures and finishes. [Photo: Barclays Center/BSE] The makeover is part of Barclays Center parent company BSE Global’s $100 million, five-year upgrade of the venue that’s still set to include improvements like a new fan zone and a new premium membership club. Laurie Jacoby, BSE Global’s chief entertainment officer, tells Fast Company the plan for the renovations is “to elevate the guest experience at every level, whether you’re a fan attending a concert or game, or an artist bringing your talent to our stage.” It’s especially important for artists, she says. [Photo: Barclays Center/BSE] “For touring artists, a dressing room is often their only connection to the city they’re performing in,” Jacoby says. “That’s why we designed our dressing rooms to capture the essence of a Brooklyn brownstone, featuring paneled walls, herringbone floors, warm lighting, and elegant finishes that create a cozy, residential atmospherea true home away from home right inside the arena.” [Photo: Barclays Center/BSE] BSE Global designed the dressing rooms in partnership with the Brooklyn Home Company, a real estate development and design firm that typically works on condos, townhomes, and vacation homes. [Photo: Barclays Center/BSE] The team designed each room to pay homage different Brooklyn neighborhoods around the stadium. Park Slope was inspired by diners and wine bars, Brooklyn Heights by red brick, Prospect by the Brooklyn Botanical Gardens, and Boerum Hill by the oxidization of copper. There’s also a classic brownstone-inspired room for Clinton Hill, and a green room with plants to represent Fort Greene. Construction began following the end of the 202425 NBA season, and the dressing rooms can seat up to 56 people. [Photo: Barclays Center/BSE] BSE Global wants the new Barclays Center to do more than impress rock bands, rappers, DJs, and pop stars. The team also hopes the elevated space will lead to elevated performances. [Photo: Barclays Center/BSE] “ur hope is that these new spaces provide comfort and hospitality for artists spending long hours on the road,” Jacoby says. “We know that when performers feel at home and relaxed, they deliver their best performanceswhich is ultimately a benefit to both fans and the overall experience.” Among the first artists who’ll use the new dressing rooms are Reneé Rapp, who’s bringing her Bite Me Tour to Brooklyn next month, and Tame Impala, who’s playing four nights there.
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||