Latest news with #supercomputing
Yahoo
30-07-2025
- Business
- Yahoo
Hon Hai Technology Group (Foxconn) and TECO Announce Strategic Alliance Targeting AI Data Center Capabilities
Key markets get boost in new shares exchange TAIPEI, July 30, 2025 /PRNewswire/ -- Hon Hai Technology Group ("Foxconn") (TWSE:2317) and TECO Electric & Machinery Co Ltd Ltd ("TECO") on Wednesday announced a share exchange, strategic alliance that will strengthen their AI infrastructure capabilities and propel the two Taiwanese tech majors into key markets in the global super-computing race. The move brings together the strengths of Foxconn, the world's largest electronics manufacturing service provider and AI server producer, and TECO, a leader in industrial electro-mechanical engineering and green energy innovation. Seizing on the development of global AI data center (AIDC) construction towards standardization and modularization, the two companies will jointly explore AIDC business opportunities. Global customers will be able to tap comprehensive data center modular products, electromechanical engineering services, and cost-competitive, one-stop solutions. According to terms approved by both boards, respectively, TECO will take a 0.519% stake in Hon Hai Precision Industry Co Ltd, the formal name Foxconn trades on at the Taiwan Stock Exchange. In turn, Foxconn will own 10% in TECO. The latter will issue 237,644,068 new shares to Foxconn, and Foxconn will issue 72,481,441 new shares to TECO, implying a share exchange ratio of approx. 1 to 0.305. The no-cash transaction is targeted to complete in fourth quarter of this year, conditional on regulatory approvals. Foxconn Chairman Young Liu said, "Time-to-market is key in the global super-computing race. Modular design is gaining popularity. As AI data centers grow in size and demand ramps higher, teaming up with TECO means both companies are able to level up and rapidly deliver comprehensive, vertically-integrated solutions to our customers – the Tier-1 CSPs and hyperscalers." TECO Chairman Morris Li said, "Changing global dynamics are creating new opportunities for business and cooperation. The strategic partnership extends the two companies' cooperation in the fields of low-carbon smart factories and energy services, toward being a one-stop solution for data centers going forward." Target markets cover Taiwan, Asia, the Middle East and the US. Texas-based TECO-Westinghouse, a world leader in making electric motors, with its US manufacturing and local services, together with Foxconn's US manufacturing base, are in line with the companies' strategic aim to expand US manufacturing and reshape the global supply chain. About TECO here. About Foxconn here. View original content to download multimedia: SOURCE Hon Hai Technology Group (Foxconn) Error while retrieving data Sign in to access your portfolio Error while retrieving data Error while retrieving data Error while retrieving data Error while retrieving data
News.com.au
29-07-2025
- Automotive
- News.com.au
Elon Musk signs $25bn AI deal
Chief executive officer Elon Musk is doubling down on Tesla's AI future, signing a $US16.5bn ($25.3 billion) deal with Samsung Electronics to build its next generation AI chips in the United States. The agreement confirmed Monday by Musk will see Samsung produce Tesla's AI6 chip at its plant in Taylor, Texas. The AI6 chip is part of Tesla's internal Dojo supercomputing project, designed to process vast amounts of data from its global fleet. Samsung’s giant new Texas fab will be dedicated to making Tesla’s next-generation AI6 chip. The strategic importance of this is hard to overstate. Samsung currently makes AI4. TSMC will make AI5, which just finished design, initially in Taiwan and then Arizona. — Elon Musk (@elonmusk) July 28, 2025 It's a strategic risk at a volatile time. Last week, Tesla reported a 42 per cent drop in year-over-year operating income for Q2 2025, from USD $1.6 billion to $923 million, a concern for investors. Tesla shares rose 1.9 per cent in US premarket trading after the announcement and Samsung shared jumped 6.8 per cent, hitting its highest level in nearly a year. While the initial deal is valued around $25 billion, Musk added on X that the figure is 'just the bare minimum'. 'Actual output is likely to be several times higher,' Musk said in a post on X. WIN, WIN FOR TESLA, SAMSUNG The deal will be a win for Samsung and assist its Texas facility, which has never had a major client and faced delays in operation since being completed in 2022. Tesla is one of Samsung's most high-profile contracts in years and a win for the company who has lost ground in the global chipmaking race. Elon Musk has long insisted Tesla is 'an AI company as much as a car company,', and this new partnership will assist Tesla's goals for vertical integration. It will also support the US government's priorities to strengthen domestic production. 'Samsung's giant new Texas fab will be dedicated to making Tesla's next-generation AI6 chip. The strategic importance of this is hard to overstate. Samsung currently makes A14. TSMC will make the A15, which just finished design, initially in Taiwan and then Arizona,' Elon Musk said. Musk added that Tesla engineers will work closely with Samsung to improve production efficiency, and that he would personally oversee production lines. The AI6 chips are critical for Tesla's Full Self-Driving technology. While Tesla has begun beta-testing its 'Full Self-Driving v12' software with select users, critics and regulators have been sceptical of Musk's project. The AI6 chips will also likely assist robotaxi platforms and possibly Tesla's robotic projects, including a humanoid robot, Optimus.
Telegraph
17-07-2025
- Science
- Telegraph
China is preparing to steal the jobs of the future
They call it 'artificial general intelligence' (AGI) and it may be only a matter of a few years away. AGI is the imagined point at which computers finally achieve consciousness, surpass human intelligence and begin to self-improve across virtually all cognitive tasks at an accelerating rate. Sometimes referred to as the 'singularity', it's long been the stuff of science fiction, but many techies believe, such is the pace of current development, that it is on the verge of becoming a reality. Whoever gets there first, it is widely believed, will inherit the Earth, embedding their influence, ideology and systems of governance into world affairs for generations to come. It's a frightening as well as awe-inspiring prospect and it is one whose potentially transformational geopolitical consequences are only now starting to be more widely appreciated. And it's why the US and China are increasingly engaged in what can only be described as a new arms race – or space race – to develop and harness artificial superintelligence for economic and geopolitical superiority. In both jurisdictions, hundreds of billions of dollars a year are being poured into getting there first. Yet though the Trump administration is only too aware of the threat, its response is oddly backward-looking and counterproductive. Despite the apparent world lead the US has in supercomputing, there is a high chance it will end up losing the war.
Fast Company
15-07-2025
- Science
- Fast Company
Why 1995 was the year the internet grew up
The internet wasn't born whole—it came together from parts. Most know of ARPANET, the internet's most famous precursor, but it was always limited strictly to government use. It was NSFNET that brought many networks together, and the internet that we use today is almost NSFNET itself. Almost, but not quite: in 1995, the government that had raised the internet from its infancy gave it a firm shove out the door. Call it a graduation, or a coming of age. I think of it as the internet getting its first real job. In the early 1980s, the National Science Foundation sought to establish the United States as a leader in scientific computing. The plan required a fleet of supercomputers that researchers could readily use, a difficult feat when the computers routinely cost more than the buildings that housed them. Business computing had solved similar problems with time-sharing and remote terminals, and ARPANET had demonstrated that terminals could be connected to computers across the country using a packet-switching network. The Computer Science Network, or CSNET, was the NSF's first foray into wide area networking. It connected universities that didn't have defense contracts and, as a result, had been left out of ARPANET. With dozens of sites, CSNET was much smaller than ARPANET but proved that a group of universities could share computing resources. When the NSF funded five cutting-edge supercomputing centers in 1985, it planned to make them available to users over a similar network. The problem was that big computers invited big data: CSNET just wasn't fast enough for interactive work with large data sets, and it was falling further behind as traffic doubled about every two weeks. After a sluggish 56 Kbps pilot effort (about a thousand times slower than today's common broadband connections), the NSF contracted the University of Michigan to develop an all-new replacement based on MERIT—a Michigan inter-university network that had already started to expand its high-speed digital telephone and geostationary satellite links into other states. In 1987, the MERIT team brought on IBM and upstart long-distance carrier MCI, freshly invigorated by the antitrust breakup of their principal competitor and truly feeling their oats. They worked at a breakneck pace. In under a year, NSFNET connected the supercomputing centers and a half dozen regional networks at blistering T1 speeds: 1.5 Mbps—an almost 28-fold increase. Just after 8 p.m. on June 30, 1988, Hans-Werner Braun, the project's co-principal investigator, sent an email to the NSFNET mailing list to announce these new high-capacity links—among the fastest long-distance computer connections ever deployed—with typical scientific understatement: 'The NSFnet Backbone has reached a state where we would like to more officially let operational traffic on.' Braun's email 'received little notice at the time,' the NSF wrote in a 2008 announcement. But 'those simple words announced the birth of the modern Internet.' NSFNET was a runaway success. Besides its massive capacity, the network maintained an open door for interconnection. Overseas academic computer networks established peer connections with NSFNET, and in 1989 the federal government opened two Federal Internet Exchanges that routed traffic between NSFNET, ARPANET, and other government networks. The superior speed of NSFNET meant that these exchanges served mostly to bring NSFNET to federal users, and ARPANET's fate was sealed. The military network, birthplace of many internet technologies, was deemed obsolete and decommissioned the next year. At the turn of the 1990s, NSFNET had become the Internet: the unified backbone by which regional and institutional networks came together. NSFNET never stopped growing. It was a remarkable problem: at every stage, NSFNET traffic grew faster than anticipated. During 1989 alone, traffic increased by five times. The state of the art T1 links were overwhelmed, demanding a 1991 upgrade to 45 Mbps T3 connections. To manage the rapidly expanding infrastructure, the original NSFNET partners formed Advanced Network and Services (ANS). ANS was an independent nonprofit that could be called the first backbone ISP, the service provider that service providers themselves connected to. The popularity of this new communications system was not limited to government and academia. Private industry took note as well. During the 1980s, 'online services' had sprouted: companies like CompuServe, PlayNet, and AOL that are often considered early ISPs but were, in fact, something else. Online services, for both businesses and consumers, were walled gardens. They descended from time-sharing systems that connected users to a single computer, providing only a 'curated' experience of software provided by the online service itself. The internet, in the tradition of ARPANET and especially NSFNET, was very different. It was a collection of truly independent networks, autonomous systems, with the freedom to communicate across geographical and organizational boundaries. It could feel like chaos, but it also fostered innovation. The internet offered possibilities that the online services never could. Douglas Van Houweling, director of the MERIT office, called NSFNET's university origin 'the only community that understands that great things can happen when no one's in charge.' At first, it was contractors who took their business to the internet. ARPANET had always been strictly for government business, but still, companies with the privilege of ARPANET connections found it hard not to use them for other work. Despite prohibitions, ARPANET users exchanged personal messages, coordinated visits, and even distributed the first spam. NSFNET's much wider scope, welcoming anyone with a nexus to research or education, naturally invited users to push the limits further. Besides, the commercial internet was starting to form. CERN engineer Tim Berners-Lee had invented HTML and, along with it, the World Wide Web. In 1993, NCSA—one of the same NSF supercomputing centers that NSFNET was built to connect—released Mosaic, the first popular web browser. Early private ISPs, companies like PSINet and Cerfnet, started out as regional academic networks (New York and California's). There was obvious business interest, and for cash-strapped academic networks paying customers were hard to turn down. NSFNET went into business on its own, with ANS establishing its own for-profit commercial subsidiary called ANS CO+RE. The term 'internet backbone' still finds use today, but in a less literal sense. NSFNET truly was the spine of the early 1990s internet, the only interconnection between otherwise disparate networks. It facilitated the internet's growth, but it also became a gatekeeper: NSF funding came with the condition that it be used for research and education. NSFNET had always kept a somewhat liberal attitude towards its users' online activities, but the growth of outright for-profit networks made the conflict between academia and commerce impossible to ignore. Several commercial ISPs established their own exchange, an option for business traffic to bypass NSFNET, but it couldn't provide the level of connectivity that NSFNET did. Besides, ANS itself opposed fragmentation of the internet and refused to support direct interconnection between other ISPs. In 1992, a series of NSFNET policy changes and an act of Congress opened the door to business traffic on a more formal basis, but the damage was done. A divide had formed between the internet as an academic venture and the internet as a business, a divide that was only deepened by mistrust between upstart internet businesses and incumbent providers ANS, IBM, and MCI. The network was not the only place that cracks formed. Dating back to ARPANET, a database called the Domain Name System maintained a mapping between numeric addresses and more human-friendly names. While DNS was somewhat distributed, it required a central organization to maintain the top level of the hierarchy. There had been different databases for different networks, but consolidation onto NSFNET required unifying the name system as well. By 1993, all of the former name registries had contracted the work to a single company called Network Solutions. At first, Network Solutions benefited from the same federal largesse as NSFNET. Registry services were funded by government contracts and free to users. Requests came faster and faster, though, and the database grew larger and larger. In 1995, Network Solutions joined the ranks of the defense industrial complex with an acquisition by SAIC. Along with the new owner came new terms: SAIC negotiated an amendment to the NSF contracts that, for the first time, introduced a fee to register a domain name. Claiming a name on the internet would run $100 per two years. By then, commercial ISPs had proliferated. Despite policy changes, NSFNET remained less enthusiastic about commercial users than academic ones. Besides, traffic hadn't stopped growing, and improved routing technologies meant the network could scale across multiple routes. The internet became competitive. MCI, benefiting from their experience operating NSFNET links, had built its own backbone network. Sprint, never far behind MCI, had one too. ANS reorganized their assets, placing much of their backbone infrastructure under their commercial operations. Government support of the increasingly profit-driven internet seemed unwise and, ultimately, unnecessary. In April of 1995, the internet changed: NSF shut down the NSFNET backbone. The government funded, academically motivated core of the internet was replaced by a haphazard but thriving interconnection of commercial ventures. ANS, now somewhat lost for purpose, stepped out into the new world of internet industry and sold its infrastructure to AOL. Network Solutions became embroiled in a monopoly controversy that saw DNS reorganized into a system of competitive private registrars. Modems became standard equipment on newly popular personal computers, and millions of Americans dialed into a commercial ISP. We built communities, businesses, and the shape of the 21st century over infrastructure that had been, just years before, a collection of universities with an NSF grant. The internet, born in the 1960s, spent its young adult years in the university. It learned a lot: the policies, the protocols, the basic shape of the internet, all solidified under the tutelage of research institutions and the NSF. And then, the internet graduated. It went out, got a job, and found its own way. Just where that way leads, we're still finding out. The super-early-rate deadline for Fast Company's Most Innovative Companies Awards is Friday, July 25, at 11:59 p.m. PT. Apply today.
National Post
08-07-2025
- Business
- National Post
SiPearl: Final closing of €130m Series A with Cathay Venture (Taiwan), EIC Fund and France 2030
Article content Tape-out of Rhea1, the most complex processor ever designed in Europe which will equip JUPITER, the 1 st European exascale supercomputer Largest Series A in the European fabless semiconductor industry including €32m in new funding to close Series A and prepare the launch of Series B Strengthening ties with Taiwan's semiconductor ecosystem to complement Europe's sovereignty in supercomputing and AI Article content MAISONS-LAFFITTE, France — SiPearl, the company building European high-performance energy-efficient processors for supercomputing and AI has achieved the definitive closing of its €130m Series A with a third tranche of €32m. Article content Seed funded by the European Union, SiPearl was launched in January 2020 under the auspices of the European Processor Initiative consortium which aims to foster the return of high-performance energy-efficient processor technologies in Europe. Since then, the company has fulfilled its mission by building a world-class processor team of 200 employees in France, Spain, and Italy, set up its own sovereign infrastructure with data centres in France equipped with servers and emulators dedicated to semiconductor design. SiPearl has completed the conception of the most complex processor ever designed in Europe, Rhea1. With 80 Arm Neoverse V1 cores, Rhea1 is composed of more than 61bn transistors. Several weeks ago, Rhea1 taped-out and was handed off to the world's leading foundry, TSMC in Taïwan, for start of manufacturing. Article content Rhea1 will equip the CPU cluster of JUPITER, the first European exascale supercomputer which is owned by EuroHPC JU and operated by Jülich Supercomputing Centre (Germany). Key component of iconic European collaborative projects, Rhea1 will help ensure Europe's technological sovereignty, independence, and competitiveness. Article content SiPearl's total Series A rounds is the largest Series A in the European fabless semiconductor industry. Its third tranche of funding is backed by 2 existing investors, EIC Fund and The French State, via French Tech Souveraineté which is part of France 2030 led by the General Secretariat for Investment, and a new investor, a major private equity firm, the Taiwanese Cathay Venture for its first investment in France. Article content This latest Series A funding round supports the industrialization phase of Rhea1. It also accelerates R&D activities for the launch of next-generation processors that will meet the needs of supercomputing and new market segments, such as data centres, AI, and enterprises, ahead of the launch of Series B in a few weeks. Article content Article content Article content Article content Article content Article content
