Microsoft's next-gen AI chip production delayed to 2026, The Information reports
June 27 (Reuters) - Microsoft's (MSFT.O), opens new tab next-generation Maia AI chip is facing a delay of at least six months, pushing its mass production to 2026 from 2025, The Information reported on Friday, citing three people involved in the effort.
When the chip, code-named Braga, goes into production, it is expected to fall well short of the performance of Nvidia's (NVDA.O), opens new tab Blackwell chip that was released late last year, the report said.
Microsoft had hoped to use the Braga chip in its data centers this year, the report said, adding that unanticipated changes to its design, staffing constraints and high turnover were contributing to the delay.
Microsoft did not immediately respond to a Reuters request for comment.
Like its Big Tech peers, Microsoft has focused heavily on developing custom processors for artificial intelligence operations and general purpose applications, a move that would help reduce the tech giant's reliance on pricey Nvidia chips.
Cloud rivals Amazon (AMZN.O), opens new tab and Alphabet's (GOOGL.O), opens new tab Google have both raced to develop chips in-house, customized for their specific needs with the goal of improving performance and reducing costs.
Microsoft had introduced the Maia chip in November 2023, but has lagged its peers in ramping it up to scale.
Google, meanwhile, has seen success with its custom AI chips - called Tensor Processing Units - and in April unveiled its seventh-generation AI chip designed to speed the performance of AI applications.
Amazon in December also unveiled its next-generation AI chip Trainium3 that is set to be released late this year.
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Reuters
2 hours ago
- Reuters
Nvidia insiders sold over $1 billion in stock amid market surge, FT reports
June 29 (Reuters) - Nvidia (NVDA.O), opens new tab insiders sold over $1 billion worth of company stock in the past year, with a notable uptick in recent trading activity as executives capitalize on surging investor interest in artificial intelligence, the Financial Times reported on Sunday. More than $500 million of the share sales took place this month as the California-based chips designer's share price climbed to an all-time high, the report said. Jensen Huang, Nvidia's chief executive, started selling shares this week for the first time since September, the SEC filing showed, opens new tab. Nvidia's stock hit a record on Wednesday, and the chipmaker reclaimed the crown as the world's most valuable company after an analyst said the chipmaker was set to ride a "Golden Wave" of artificial intelligence. Its latest gains reflect the U.S. stock market's return to the "AI trade" that fueled massive gains in chip stocks and related technology companies in recent years on optimism about the emerging technology. Nvidia did not immediately respond to a Reuters request for comment. Nvidia's shares have rebounded over 60% from their closing low on April 4, when Wall Street was reeling from President Donald Trump's global tariff announcements. U.S. stocks, including Nvidia, have recovered on expectations the White House will reach trade deals to soften the tariffs.
The National
3 hours ago
- The National
AI alone just won't wash – people must be in the pipeline
I needed to spin up a very quick solution to the problem of getting 2025 tourism information in Tiree tidied up into a single usable place which was, crucially, easily updatable. This might surprise some folks, who remain convinced that by talking about the challenges of tourism, and second homes, I am single-handedly trying to destroy the industry. Nothing could be further from the truth – the goal is to try and do tourism better, to the benefit of our communities – but that's a column for another day. READ MORE: 'Completely unprecedented': BBC cuts live feed for Kneecap Glastonbury performance In this instance, I am wearing my Trust comms hat, and trying to ensure that people visiting Tiree get all the information they need at their fingertips. I want to make sure that we are promoting all local businesses, and that we are clearly communicating the key things we want people to know – like how to use passing places, and when and how dogs should be controlled. This is an important part of making tourism sustainable, and beneficial. It should also give the visitor a better experience. Getting that information out to as many people as possible seems like something that technology should be able to simplify. After all, in 2025, isn't everything solved by applying 'AI'? In short, no – but it can help if you know what you are doing. AI doesn't know where we live. It doesn't know what is still open or not open, it has no idea about the realities of visiting an island with no cash machine and cranky crofters (I include myself in that demographic), and it has not a scooby about the vagaries of island life. The problem of getting the details right is not one that's going to be solved by AI and it's not one that's going to be solved by guides that are based on people sucking information off the internet and turning it into a money maker. It's a problem that needs to be solved by people. We understand the difference between information and knowledge. Computers do not. (Image: Unsplash) A disclaimer: I am writing this before the app is launched, and in the full knowledge that I might end up with egg all over my face as the entire concept flops. Such is life. Let's take this information guide as an example of when and how 'AI' is particularly useful. Why am I insisting on putting it in quotes? Because it is not true AI. Not even close. The AI tools we are using are just clever computer programmes with fancy names. Most of the companies producing 'AI' solutions are simply reclothing the emperor. The emperor has had many outfits. Remember Dropbox's early days? It presented itself as seamless cloud magic, but behind the curtain, there was a bloke manually moving files between servers – a classic piece of human-powered sleight of hand. We've seen that before, and now it's happening again with AI. Take once a Microsoft-backed, billion-valuation startup, it claimed its 'AI' assistant Natasha could build apps just by chatting. In reality, however, it leaned on around 700 human engineers in India to do the coding while calling it AI-powered – a textbook case of 'AI-washing'. It's not that it didn't work – it did, thanks to real people – but the magic was all in the marketing, not the algorithm. So back to the practical side of building this thing. The island already has a very good website, but staff at the Trust find it hard to update because the backend of the website is needlessly complicated, and requires them to set aside time to refresh themselves on how to do it every time. When it came to the data, I could have sent AI off to gather all the up-to-date details for businesses in Tiree, to scour the web for the important stuff people need to know and to compile it into a guide. But that would have been a deeply stupid course of action. The internet is full of information about Tiree – some of it right, a lot of it not. There are business websites and social media profiles updated on an ad hoc basis, glossy magazine features with variable accuracy and out-of-date attempts at exactly what I am doing. If you want a good laugh, there's an 'AI'-written guide to Tiree that's so wildly inaccurate it's worth buying for the giggle. (Image: Getty Images) AI doesn't think independently. It draws conclusions from the data it's given. If that data is wrong, so are the results. In this case, the data is far from sound. We would have had to check everything anyway. So we did the data entry manually – copied content, cross-referenced with social media and filled in what we could from public sources. We did the bit AI cannot do – we verified. Content gathering, though, was the easy part. The harder bits were organising the information and making it super simple to update. For users, it had to be laid out in a way that made sense to both visitors and the community. What do we want people to know? What are they likely to miss? What matters here, in Tiree? These are not questions AI can answer. You need to live here, or listen to the people who do. So I asked around, took advice and tried to reflect the priorities of the island. None of that could have come from a chatbot. That part still takes people. Always has. It was the updating part that I was most interested in. If you have an office tech geek with the time to update opening hours in an inscrutable content management system, that's great but those are few and far between. Updating needed to be quick and effortless. I asked ChatGPT how best to do it. It came up with a series of suggestions – including an app builder that runs off a spreadsheet. I had never heard of it, but I was interested. I ended up using it to build the new information app. And the whole thing does indeed run off a Google spreadsheet. To change an opening time, all anyone needs to do is update a spreadsheet cell. That's it. In this case, AI earned its keep by giving me a quick solution I didn't know existed. It offered up an app builder that runs off a spreadsheet, and that turned out to be exactly what I needed. But it only gave the right suggestion because I already understood the 'why' and the 'who'. The software itself also describes what it does as 'AI'. What it actually did was ask me what I wanted, suggest a suitable template and walk me through connecting a spreadsheet and mapping fields to cells. As a front-end person, databases are not my bag, so this worked really well. But let's not pretend it was thinking. It was responding to instructions. So yes – AI helped me build an app. But the purpose, the content, the priorities? That came from here. It didn't speak to the people who live on the island, or consider what makes life easier for visitors and locals alike. I did that. The tool was helpful because the hard part – the understanding – had already been done. Only the season will tell whether other people agree with my understanding.
Times
9 hours ago
- Times
Don't dismiss Elon Musk's Doge so fast: we can learn from its failure
History won't be kind to Elon Musk's Department of Government Efficiency (Doge). As failed, expensive experiments go, it's up there with HS2 or 'Tay,' Microsoft's much-hyped AI chatbot that, within hours of launching in 2016, morphed spectacularly into a racist troll. Doge initially pledged to slice $2 trillion from federal spending — a bravado-fuelled ambition that was swiftly halved and then repeatedly whittled down until landing at a relatively underwhelming $150 billion. Even this revised sum relies on some fairly questionable assumptions and shaky accounting. Some independent commentators suggest the real savings hover close to zero, especially once the anticipated tsunami of lawsuits — or the burden of haphazardly dismantled departments limping on dysfunctionally — are factored in. The Department of Government Efficiency will, ironically, be remembered for its inefficiency. Yet, oddly enough, I remain rather enamoured with the concept. Just as high-speed rail doesn't have to degenerate into a bloated money pit bereft of trains, and AI chatbots needn't transform into spiteful bigots, the basic idea behind Musk's ill-fated initiative holds merit. The execution may have flopped, but the underlying model of applying a private sector mindset to government spending and bureaucracy deserves resurrection. History is full of promising ideas that tanked the first time around. Bubble Wrap, for example, originally intended as textured wallpaper, languished unsold until IBM adopted it in the 1960s for protecting computer components. The billions of Post-it Notes sold annually began life inauspiciously in 1968 as a glue deemed too weak for aerospace engineering, only to find new purpose in the 1980s. One of Doge's many glaring oddities was that, despite being overseen by one of the world's most successful corporate CEOs, its cost-cutting approach was remarkably un-corporate. Most jobs — even those which are poorly performed, vulnerable to automation, or submerged in bureaucratic sludge —were originally created for a valid reason. This explains why corporate belt-tightening almost always ends rather than begins with job cuts. CEOs and CFOs typically look at expenses before wielding the axe, scrutinising discretionary spending first — travel, events, equipment — and cutting back hours or trimming temporary hires. Full-time roles are usually the last domino to fall. Musk flipped that sequence, firing up the chainsaw without any evident due diligence or sober analysis. This wasn't the sort of disruption or iconoclastic thinking for which Silicon Valley has become famed, just muddled recklessness. Over here, taxpayers' money continues to be splurged on an array of baffling pursuits. This year, these have included a £99 million initiative devoted to teaching overseas families to 'cook with electricity', including in places where basic infrastructure — and electrical cooking appliances — are lacking. • Outgoing head of Reform's 'Doge' urges party to avoid Musk's mistakes Then there's the £1 million grant to the Open University, which made the news last month, funding a two-year project to encourage students to 'touch as a mole' and feel 'like a bee'. Given the same amount would pay for 20 police constables for a year, I'm fairly certain that wastefulness could be identified without requiring a Musk-style wrecking ball approach. Right now, injecting disciplined private-sector thinking into governmental budgeting feels not just sensible, but necessary. Rachel Reeves's recent spending review has dramatically opened the purse strings, with departmental budgets growing by 2.3 per cent — the neck end of an additional £200 billion allocated to daily public-sector operations. I don't find the spending itself inherently problematic. Rather, I worry about who within the current government has earned genuine credibility managing substantial sums prudently or, indeed, could be trusted to implement rigorous cost controls to offset the increased largesse. In this regard, I believe there are plenty of highly talented figures from the world of business, with proven track records, who would bring more experience, expertise and, in all likelihood, results than the government could muster. • Fraser Nelson: Elon Musk's Doge debacle has done us all a favour How about someone like Sir Terry Leahy, who streamlined Tesco? His brand of operational efficiency relied heavily on automation, data analytics, and smart technology as well as lean logistics. Or Sir Stuart Rose, under whose leadership M&S developed more prudent supply chain management, stricter inventory controls and a better ability to negotiate more favourable terms with suppliers? Perhaps once Dame Carolyn McCall steps down from ITV? Diplomatic and a good communicator, she has credentials in turning around, and modernising, businesses in the public eye. Musk might have inadvertently given efficiency a bad name but his version of Doge should be regarded as a flawed prototype rather than proof of a dud concept. Having had a helpful case study of 'how not to do it', perhaps it's time for someone to do it properly. Seema Shah is chief global strategist at Principal Asset Management
