 |
|
|
|||||
On Thursday, United Parcel Service (UPS) predicted downbeat 2025 revenue as it cut back service with its largest customer, Amazon, in order to focus on more profitable businesses. The decision will cut Amazons transported volumes by more than 50% by the second half of 2026. The unexpected announcement came with disappointing revenue results, with UPS failing to meet expectations for 2024. UPS’s Q4 revenue was $25.3 billion, slightly below the predicted $25.42 billion. This decision came at a time when UPS is struggling due to a decline in parcel demand following the e-commerce boom of the pandemic and an increase in shipments from discount online retailers such as Shein and Temu. UPS stock was down more than 14% as of Thursday afternoon, after falling during premarket trading. The decline was on track to break the current record for the stock’s worst day, which occurred in July 2024, as MarketWatch reports. Shares have lost half their value since 2022, though the company said that carrying less freight for Amazon will eventually boost its revenue per piece. Amazon is our largest customer, but its not our most profitable customer, CEO Carol Tomé told investors on a conference call Thursday. “Foundational changes” Amazon and its affiliates represented about 12% of UPS revenue in 2023, which is nearly all of its United States package business. The package carrier is hoping to gain volume from more profitable segments, such as healthcare-product shippers, small- and-medium-size businesses, and international markets. UPS expects revenue of about $89 billion for 2025, down from $91.1 billion in 2024. The parcel service has brought all of its UPS SurePost products in-house and will launch multiyear efficiency reimagine initiatives to drive about $1 billion in savings by rethinking its business design. The company aims to cut costs by closing buildings, reducing the size of its vehicle and aircraft fleets, and decreasing the size of its workforce. We are making business and operational changes that, along with the foundational changes weve already made, will put us further down the path to becoming a more profitable, agile, and differentiated UPS that is growing in the best parts of the market, said Tomé in the annual earnings report. In 2025, UPS expects average daily U.S volume to drop about 8.5% year-over-year while revenue per package is projected to increase by 6%.
Category:
E-Commerce
Welcome to AI Decoded, Fast Companys weekly newsletter that breaks down the most important news in the world of AI. You can sign up to receive this newsletter every week here. After a week of DeepSeek freakout, doubts and mysteries remain The Chinese company DeepSeek sent shockwaves through the AI and investment communities this week as people learned that it created state-of-the-art AI models using far less computing power and capital than anyone thought possible. The company then showed its work in published research papers and by making its models available to other developers. This raised two burning questions: Has the U.S. lost its edge in the AI race? And will we really need as many expensive AI chips as weve been told? How much computing power did DeepSeek really use? DeepSeek claimed it trained its most recent model for about $5.6 million, and without the most powerful AI chips (the U.S. barred Nvidia from selling its powerful H100 graphics processing units in China, so DeepSeek made do with 2,048 H800s). But the information it provided in research papers about its costs and methods is incomplete. The $5 million refers to the final training run of the system, points out Oregon State University AI/robotics professor Alan Fern in a statement to Fast Company. In order to experiment with and identify a system configuration and mix of tricks that would result in a $5M training run, they very likely spent orders of magnitude more. He adds that based on the available information its impossible to replicate DeepSeeks $5.6 million training run. How exactly did DeepSeek do so much with so little? DeepSeek appears to have pulled off some legitimate engineering innovations to make its models less expensive to train and run. But the techniques it used, such as Mixture-of-experts architecture and chain-of-thought reasoning, are well-known in the AI world and generally used by all the major AI research labs. The innovations are described only at a high level in the research papers, so it’s not easy to see how DeepSeek put its own spin on them. Maybe there was one main trick or maybe there were lots of things that were just very well engineered all over, says Robert Nishihara, cofounder of the AI run-time platform Anyscale. Many of DeepSeeks innovations grew from having to use less powerful GPUs (Nvidia H800s instead of H100s) because of the Biden Administrations chip bans. Being resource limited forces you to come up with new innovative efficient methods, Nishihara says. Thats why grad students come up with a lot of interesting stuff with far less resourcesits just a different mindset. What innovation is likely to influence other AI labs the most? As Anthropics Jack Clark points out in a recent blog post, DeepSeek was able to use a large model, DeepSeek-V3 (~700K parameters), to teach a smaller R1 model to be a reasoning model (like OpenAIs o1) with a surprisingly small amount of training data and no human supervision. V3 generated 800,000 annotated text samples showing questions and the chains of thought it followed to answer them, Clark writes. DeepSeek showed that after processing the samples for a time the smaller R1 model spontaneously began to think about its answers, explains Andrew Jardine, head of go-to-market at Adaptive ML. You just say here’s my problemcreate some answers to that problem and then based on the answers that are correct or incorrect, you give it a reward [a binary code that means good] and say try again, and eventually it starts going I’m not sure; let me try this new angle or approach or that approach wasnt the right one, let me try this other one and it just starts happening on its own. Theres some real magic there. DeepSeeks researchers called it an aha moment. Why havent U.S. AI companies already been doing what DeepSeek did? How do you know they havent? asks Jardine. We don’t have visibility into exactly the techniques that are being used by Google and OpenAI; we don’t know exactly how efficient the training approaches are. Thats because those U.S. AI labs dont describe their techniques in research papers and release the weights of their models, as DeepSeek did. There’s a lot of reason to believe they do have at least some of these efficiency methods already. It should come as no surprise if OpenAIs next reasoning model, o3, is less compute-intensive, more cost-effective, and faster than DeepSeeks models. Is Nvidia stock still worth 50X of earnings? Nvidia provides up to 95% percent of the advanced AI chips used to research, train, and run frontier AI models. The companys stock lost 17% of its value on Monday when investors interpreted DeepSeeks research results as a signal that fewer expensive Nvidia chips would be needed in the future than previously anticipated. Metas Yann LeCun says Mondays sell-off grew from a major misunderstanding about AI infrastructure investments. The Turing Award winner says that while DeepSeek showed that frontier models could be trained with fewer GPUs, the main job of the chips in the future will be during inferencethe reasoning work the model does when its responding to a users question or problem. (Actually, DeepSeek did find a novel way of compressing context window data so that less compute is needed during inference.) He says that as AI systems process more data, and more kinds of data, during inference, the computing costs will continue to increase. As of Wednesday night, the stock has not recovered. Did DeepSeek use OpenAI models to help train its own models? Nobody knows for sure, and disagreement remains among AI experts on the question. The Financial Times reports Wednesday that OpenAI believes it has seen evidence that DeepSeek did use content generated by OpenAI models to train its own models, which would violate OpenAIs terms. Distillation refers to saving time and money by feeding the outputs of larger, smarter models into smaller models to teach them how to handle specific tasks. Weve just experienced a moment when the open-source world produced some models that equaled the current closed-source offerings in performance. The real cost of developing the DeepSeek models remains an open question. But in the long run the AI companies that can marshal the most cutting-edge chips and infrastructure will very likely have the advantage as fewer performance gains can be wrung from pretraining and more computing power is applied at inference, when the AI must reason toward its answers. So the answers to the two burning questions raised above are probably not and likely yes. The DeepSeek breakthroughs could be good news for Apple The problem of finding truly useful ways of using AI in real life is becoming more pressing as te cost of developing models and building infrastructure mounts. One big hope is that powerful AI models will become so small and efficient that they can run on devices like smartphones and AR glasses. DeepSeeks engineering breakthroughs to create cheaper and less compute-hungry models may breathe new life into research on small models that live on edge devices. Dramatically decreased memory requirements for inference make edge inference much more viable, and Apple has the best hardware for exactly that, says tech analyst Ben Thompson in a recent Stratechery newsletter. Apple Silicon uses unified memory, which means that the CPU, GPU, and NPU (neural processing unit) have access to a shared pool of memory; this means that Apple’s high-end hardware actually has the best consumer chip for inference. Stability AI founder Emad Mostaque says that reasoning models like OpenAI’s o1 and DeepSeek’s R1 will run on smartphones by next year, performing PhD-level tasks with only 20 watts of electricityequivalent to the human brain. OpenAI releases an AI agent for government workers OpenAI this week announced a new AI tool called ChatGPT Gov thats designed specifically for use by U.S. government agencies. Since sending sensitive government data out through an API to an OpenAI server presents obvious privacy and security problems, ChatGPT Gov can be hosted within an agency’s own private cloud environment. [W]e see enormous potential for these tools to support the public sector in tackling complex challengesfrom improving public health and infrastructure to strengthening national security, OpenAI writes in a blog post. The Biden Administration in 2023 directed government agencies to find productive and safe ways to use new generative AI technology (Trump recently revoked the executive order). The Department of Homeland Security, for example, built its own AI chatbot, which is now used by thousands of DHS workers. OpenAI says 90,000 users within federal, state, and local government offices have already used the companys ChatGPT Enterprise product. More AI coverage from Fast Company: Microsoft posts 10% growth for Q4 as it plans to spend $80 billion on AI AI assistants for lawyers are a booming businesswith big risks Why we need to leverage AI to address global food insecurity Alibaba rolls out AI model, claiming its better than DeepSeek-V3 Want exclusive reporting and trend analysis on technology, business innovation, future of work, and design? Sign up for Fast Company Premium.
Category:
E-Commerce
This story was originally published by ProPublica. The icebreaker Aiviq is a gas guzzler with a troubled history. The ship was built to operate in the Arctic, but it has a type of propulsion system susceptible to failure in ice. Its waste and discharge systems werent designed to meet polar code, its helicopter pad is in the wrong place to launch rescue operations and its rear deck is easily swamped by big waves. On its maiden voyage to Alaska in 2012, the 360-foot vessel lost control of the Shell Oil drill rig it was towing, and Coast Guard helicopter crews braved a storm to pluck 18 men off the wildly lurching deck of the rig before it crashed into a rocky beach. An eventual Coast Guard investigation faulted bad decision-making by people in charge but also flagged problems with the Aiviqs design. But for all this, the same Coast Guard bought the Aiviq for $125 million late last year. The United States urgently needs new icebreakers in an era when climate change is bringing increased traffic to the Arctic, including military patrols near U.S. waters by Russia and China. That the first of the revamped U.S. fleet is a secondhand vessel a top Coast Guard admiral once said may, at best, marginally meet our requirements is a sign of how long the country has tried and failed to build new ones. Its also a sign of how much sway political donors can have over Congress. Edison Chouest, the Louisiana company that built the icebreaker, has contributed more than $7 million to state and national parties, to political action committees and super PACS, and to members of key House and Senate committees since 2012. Chouest spent most of that period looking to unload the vessel after Shell, its intended user, walked away. Members who received money from Chouest pressured the Coast Guard to rent or buy the Aiviq from the company. One U.S. representative from Alaska, where the ship will be stationed, told an admiral in a 2016 hearing that his services objections were bullshit. And there would be even tougher pressures to come. Its now been a dozen years since the Aiviq set out on its first mission to Alaska, long enough for its troubles to fade from public memory. The ship, though owned and operated by Chouest, was part of Shells Arctic fleet, designed for a specific role: as a tugboat that could tow Shells 250-foot-tall polar drill rig, the Kulluk, around the coast of Alaska and help anchor it in the waters of the Far North. At its christening ceremony in Louisiana, attended by Shell executives, U.S. Sen. Lisa Murkowski of Alaska and U.S. Rep. Steve Scalise of Louisiana, it was named after the Iupiaq word for walrus. As a journalist, Id been following the oil companys multibillion-dollar play in the warming Arctic with interest. One June morning in 2012, I got word that Shell was on the move near my Seattle home, so I sped to a narrow point in Puget Sound with a good view of passing traffic. It was sunny, the water calm. The Aiviq bobbed past with Kulluk in tow. The icebreakers paint blue at the time was fresh, its hull shiny. It looked capable. The problems began once the Aiviq was out of view. A Coast Guard report said that while the ship towed the Kulluk northward through an Arctic storm, waves crashed over its rear deck and poured into interior spaces, which investigators determined may have caused it to list up to 20 degrees to one side. The water damaged cranes, heaters and firefighting equipment, and the vents to the fuel system were submerged. On its way back from Alaskas Beaufort Sea two months later, the Aiviq suffered an electrical blackout, and one of its engines failed, necessitating a repair in Dutch Harbor in Alaskas Aleutian Islands. Then the Aiviq and Kulluk set out on a wintertime voyage back to Seattle. The National Weather Service issued a gale warning predicting 15-foot seas and 40-knot winds. The sailors aboard the Aiviq and Kulluk exchanged worried messages. The cable with which the Aiviq was towing the Kulluk came free two days later when a shackle broke. The icebreakers captain made a U-turn in heavy swells to hook up an emergency tow line, and water again poured over its deck and into the fuel vents. The Aiviqs four diesel engines soon began to fail, one after another. Although a Chouest engineer later testified that an unknown fuel additive must have caused the failures, Coast Guard investigators believe the likely cause was fuel contamination by seawater. They said the fuel systems design, which they described as substandard, made contamination more likely. The Aiviq and Kulluk were reattached but now, and for the next two days, adrift. Storms pushed them ever closer toward land. By the time the engines were repaired, it was too late. The Kulluk ran aground at an uninhabited island off Kodiak, Alaska, on New Years Eve. Shells Arctic dreams began to unravel. The oil company sold its drill rig off for scrap. (It did not respond to a request for comment.) And the Aiviq? A month after the accident, I visited Kodiak to report on what went wrong. I saw it anchored in the safety of a protected bay, an expensive, purpose-built ship now stripped of its purpose. Shell formally abandoned its Arctic efforts in 2015, after failing to find oil. The Aiviq eventually steamed back south. Chouest began looking around for someone to take the troubled icebreaker off its hands. The Coast Guard, which had criticized the ships role in the Kulluk accident, now became a potential customer. Traffic in the warming Arctic has surged as countries eye the regions natural resources, and it will grow all the more if the storied Northwest Passage melts enough to become a viable route for freight in the decades ahead. The number of ships in the High North increased by 37% from 2013 to 2023. Its the U.S. Coast Guards job to patrol these waters as part of an agreement with the Navy, projecting military strength while monitoring maritime traffic, enforcing fishing laws and rescuing vessels in distress. Although surface ice in the Arctic Ocean is shrinking on average, it can still form and move about the ocean unpredictably. A Coast Guard vessel needs to be able to cut through it to be a reliable presence. But the U.S. icebreaker fleet is deteriorating. The Coast Guard began raising alarms about the problem decades ago, starting with a study published in 1984. Russia, with its extensive northern coastline, now has over 40 large icebreakers, and more under construction. The United States has barely been able to keep two or three in service. An urgent Coast Guard report to Congress in 2010 highlighted what has become known as the icebreaker gap: If we didnt quickly start building new ships, our existing icebreakers could go out of commission before replacements were ready. The study called for at least six new icebreakers. Subsequent Coast Guard analysis has called for eight or nine. To date, the United States has built zero. Congress dragged its feet for years on funding icebreaker construction. But theCoast Guard also slowed progress with overly optimistic timelines, fuzzy cost estimates and a tendency to keep fiddling with new designs, according to a 2023 Government Accountability Office report. More than a decade in, construction on the first of the new ships has finally just begun. The Coast Guards latest cost estimate is $1 billion per icebreaker, while the Congressional Budget Office last year put it at $1.6 billion to $1.9 billion. Icebreakers have been the penultimate studied-to-death subject for 40 years, said Lawson Brigham, a former Coast Guard heavy icebreaker commander who has a doctorate from Cambridge University and has researched polar shipping since the 1980s. The longer the Coast Guard failed to build the ships it did want, the more pressure it faced to settle for one it didnt. Chouest seized the opportunity. The company invited Coast Guard officers to tour the Aiviq as early as 2016 and soon sent over a lease proposal. Canada rejected similar overtures that year. A middleman for Chouest promised Canadian lawmakers a fast-track polar icebreaker the Aiviq at less than one-third of the price of the permanent replacement. Also on offer were three smaller, Norwegian-built icebreakers. Canada bought those instead. The U.S. Coast Guards problem with the Aiviq, retired officers told ProPublica, was the ships design. Originally built for oil operations, it had a low, wet deck and a helipad near its bow, where it would be ill suited for launching rescue operations. Its direct-drive propulsion system was both less efficient and more likely to get jammed up in ice than the diesel-electric systems the Coast Guard used. I mean, on paper its an icebreaker, Adm. Paul Zukunft, the then-commandant of the Coast Guard, told Congress in 2017. But it hasnt demonstrated an ability to break ice. (Years later, in 2022 and 2023, the Aiviq would make two successful icebreaking trips to Antarctica under contract with the Australian government.) The service estimated it would take years and hundreds of millions of dollars to upgrade the Aiviqs features to near-standard for a Coast Guard icebreaker. Even then, it wouldnt be able to move forward through ice thicker than about 4.5 feet. The Coast Guards most immediate need was for heavy icebreakers, burlier ships that can handle missions in the Arctic as well as supply runs to the U.S. research station at McMurdo Sound, Antarctica. So how would the U.S. Coast Guard use the Aiviq beyond flag-waving and general presence in the near Arctic? According to Brigham, the former icebreaker captain and polar-shipping expert, No one that I know, no study that Ive seen, no one Ive talked to really knows. But it wasnt for the Coast Guard alone to turn down Chouests bargain offer. Members of Congress had their own ideas. The late U.S. Rep. Don Young represented Alaska, a state thousands of miles from Chouests home base in Louisiana. But as of 2016, when Chouest was looking to sell the Aiviq, Young had taken in hundreds of thousands of dollars in political contributions from the company so many donations in one year that he had once faced a congressional ethics investigation concerning Chouest money. (He was cleared.) Young became the most vocal of many congressional critics to publicly dress down the Coast Guard for resisting Chouests offering of the Aiviq. At a House hearing that July, he began grilling the Coast Guards second-in-command, Adm. Charles Michel, about a privately owned ship with a tremendous capability of icebreaking power. I know you have the proposal on your desk, he scolded Michel. It is an automatic no. Why? Sir, the admiral said, that vessel is not suitable for military service without substantial refit. Michels response sparked derision from Young. That is what I call, Young muttered, a bullshit answer. Michel, now retired, declined to comment on his exchange with Young. According to the representatives former chief of staff Alex Ortiz, Youngs frustration stemmed from the fact that the Coast Guard lacked the money to build an icebreaker from scratch but showed an unwillingness to accept the realities of that. Young and many other lawmakers also supported getting new icebreakers, but perfect had become the enemy of the good the Aiviq had to offer right away. I genuinely dont think that he was advocating for leasing the vessel just because of Chouests support, Ortiz said. Chouest, Youngs benefactor, is based in Cut Off, Louisiana. Its led by its founders billionaire son and has long provided ships for the oil and gas industry. At the time of the 2016 hearing, Chouest was relatively new to Coast Guard contracts. One of the companys affiliates would later take over the contract to build new heavy icebreakers, in 2022, making Chouest the supplier of both a ship the Coast Guard desired and the one it resisted. Chouest did not respond to questions for this article. More than 95% of Chouests $7 million in political contributions since 2012 has gone to Republicans, according to OpenSecrets, a nonprofit that tracks money from family members, employees and corporate affiliates. But when it comes to lawmakers who oversee the Coast Guard, Democrats also have been major recipients. The late Rep. Elijah Cummings of Maryland, head of the House Subcommittee on Coast Guard and Maritime Transportation for five years, received $94,700 in the decade before his 2019 death. Rep. John Garamendi of California, a longtime committee member, started taking Chouest donations in 2021 and has since received a total of $40,500. (Garamendis office acknowledged the recent donations but issued a statement saying he has for many years pushed the Coast Guard to build icebreakers expeditiously, particularly given the aging fleet and the national security imperative.) Alaska politicians are particular beneficiaries of Chouests largesse, second only to those from Louisiana. Chouests interests in the 49th state, beyond icebreakers, have included a 10-year contract to escort oil tankers through Alaskas Prince William Sound. Federal Elections Commission records show that Young, before his death in 2022, collected a career total of almost $300,000 from the company. Sen. Dan Sullivan has taken in at least $31,500, Sen. Lisa Murkowski $84,400. The year after Young swore at the Coast Guard admiral in public, Rep. Duncan D. Hunter of California brought up the issue once more at a different House hearing featuring a different admiral, Zukunft. Hunters total from Chouest would be $58,800 before he pleaded guilty to stealing campaign funds and stepped down in 2020. Icebreakers, Hunter said. Lets talk icebreakers. Hunter was backed up by Rep. Garret Graves of Louisiana, whose Chouest contributions now total $240,500. Admiral, I think every time youve come before this committee, this issue has come up, Graves said. We need to see some substantial progress. Weeks later at yet another hearing, Rep. John Carter of Texas, whose single biggest donor the previous election cycle was Edison Chouest at $33,700, pressed Zukunft again. Theres this commercial ship that has been offered Carter began. In the end, the advocates for Chouests ship prevailed. The Alaskans played a particular role. In 2022, after Youngs death, Sullivan helped author the Don Young Coast Guard Authorization Act, which included an approval for the service to buy aUnited States built available icebreaker. Sullivan, who would later be praised for leading a revolt against his Senate colleague Tommy Tubervilles blockade on promotions of military officers, also engaged in some quiet hardball. Until the country can complete a long-delayed near-Arctic port, icebreakers have been based in Seattle, where there are working shipyards and experienced contractors to do maintenance. But as a recent press release describes it, Sullivan put a hold on certain USCG promotions until the Coast Guard produced a long promised study on the homeporting of an icebreaker in Alaska. Last year, Sullivan, Murkowski and former Rep. Mary Peltola of Alaska announced that Congress had finally appropriated $125 million for the Aiviq. The Coast Guard took possession of the ship last month. (Murkowski and Peltola, along with Hunter, Graves and Carter, did not respond to requests for comment.) In a statement to ProPublica, a Sullivan spokesperson wrote that the senator has long advocated for the purchase of a commercially available icebreaker of the Coast Guards choosing but has never advocated for the purchase of the Aiviq specifically. The way Congress wrote the specifications for a United States built icebreaker, however, ensured there was only one the Coast Guard could choose: the Aiviq. The icebreaker’s new home based on the findings of the Coast Guards urgently completed port study will be Alaskas capital, Juneau. The city is facing what the Juneau Empire has called a crisis-level housing shortage, and it remains unclear how it will manage an influx of hundreds of sailors and family members. Juneau also lacks a shipyard. For repairs and upgrades, the Aiviq will have to travel hundreds or thousands of miles out of state. Former Coast Guard icebreaker captains were reluctant to criticize the purchase of the Aiviq when contacted by ProPublica, in part because it has taken impossibly long for the service to build the new heavy icebreakers it says it needs. Is the Coast Guard getting the Aiviq a bad thing? No, said Rear Adm. Jeff Garrett, a former captain of the Healy icebreaker. But is it the ideal resource? No. To reach the Arctic from Juneau, Garrett noted, the Aiviq will have to regularly cross the same storm-swept stretch of the Gulf of Alaska where it once lost the Kulluk. Lawson Brigham said he had questions about the Aiviq since its our tax dollars at work, but he granted that its bringing some capability into the Coast Guard at a time when were awaiting whenever the shipbuilder can get the first ship out, which is still unknown. Zukunft, who retired in 2018, stands by his past opposition to the Aiviq. I remain unconvinced, he wrote in response to questions from ProPublica, that it meets the operational requirements and design of a polar icebreaker that have been thoroughly documented by the Coast Guard. By acquiring the Aiviq, the Coast Guard runs the risk that those requirements can be compromised. In a statement, the Coast Guard described the purchase of the Aiviq as a bridging strategy and said the ship will be capable of projecting U.S. sovereignty in the Arctic and conducting select Coast Guard missions. The fuel vents that flooded during the Kulluk accident have since been raised, a Chouest engineer has testified. The Coast Guard did not respond to questions about the Aiviqs fuel consumption or whether its waste systems will comply with polar code. It did not say whether its helicopter deck will be moved aft for safer search-and-rescue operations. It confirmed that there will be no changes to the propulsion system. Initial modifications to the vessel will be minimal, the statement reads. The Aiviq will be put into service more or less as is. Last month, an amateur photographer spotted the Aiviq at a Chouest-owned shipyard in Tampa, Florida, and posted images online. It had been repainted, its hull now a gleaming Coast Guard icebreaker red. New lettering revealed that the ship has been renamed the Storis, after a celebrated World War II vessel that patrolled for 60 years in the Bering Sea and beyond. From a distance, the icebreaker looked ready to serve. The question is, said Brigham, What is this ship going to be used for? Thats been the question from Day 1. What the hell are we going to use it for? McKenzie Funk, ProPublica
Category:
E-Commerce
All news |
||||||||||||||||||
|
||||||||||||||||||