Is the cloud the wrong place for AI?
The enterprise software playbook seemed clear: everything moves to the cloud eventually. Applications, databases, storage: they all followed the same inevitable arc from on-premises to software-as-a-service.
But with the arrival and boom of artificial intelligence, we're seeing a different story play out, one where the cloud is just one chapter rather than the entire book.
AI systems
AI workloads are fundamentally different beasts than the enterprise applications that defined the cloud migration wave. Traditional software scales predictably, processes data in batches, and can tolerate some latency.
AI systems are non-deterministic, require massive parallel processing, and often need to respond in real-time. These differences reshape the entire economic equation of where and how you run your infrastructure.
Take the challenge of long-running training jobs. Machine learning models don't train on schedule; they train until they converge. This could be hours, days, or weeks. Cloud providers excel at providing infrastructure at short notice, but GPU capacity at hyperscalers can be hard to get without a 1 year reservation.
The result is either paying for guaranteed capacity you might not fully use, or risking that your training job gets interrupted when using spot instances to reduce costs.
Then there's the inference challenge. Unlike web applications that might see traffic spikes during Black Friday, AI services often need to scale continuously as customer usage grows.
The token-based pricing models that govern large language models make this scaling unpredictable in ways that traditional per-request pricing never was. A single customer query might consume 10 tokens or 10,000, depending on the complexity of the response and the size of the context window.
Hybrid approaches
The most intriguing development involves companies discovering hybrid approaches that acknowledge these unique requirements rather than abandoning the cloud. They're using on-premises infrastructure for baseline, predictable workloads while leveraging cloud resources for genuine bursts of demand.
They're co-locating servers closer to users for latency-sensitive applications like conversational AI. They're finding that owning their core infrastructure gives them the stability to experiment more freely with cloud services for specific use cases.
This evolution is being accelerated by regulatory requirements that simply don't fit the cloud-first model. Financial services, healthcare, and government customers often cannot allow data to leave their premises.
For these sectors, on-premises or on-device inference represents a compliance requirement rather than a preference. Rather than being a limitation, this constraint is driving innovation in edge computing and specialized hardware that makes local AI deployment increasingly viable.
Infrastructure strategies
The cloud providers aren't standing still, of course. They're developing AI-specific services, improving GPU access, and creating new pricing models. But the fundamental mismatch between AI's resource requirements and traditional cloud economics suggests that the future won't be a simple rerun of the SaaS revolution.
Instead, we're heading toward a more nuanced landscape where different types of AI workloads find their natural homes. Experimentation and rapid prototyping will likely remain cloud-native. Production inference for established products might move closer to owned infrastructure. Training runs might split between cloud spot instances for cost efficiency and dedicated hardware for mission-critical model development.
The approach represents a step toward infrastructure strategies that match the actual needs of AI systems rather than forcing them into patterns designed for different types of computing
The most successful AI companies of the next decade will likely be those that think beyond the cloud-first assumptions and build infrastructure strategies as sophisticated as their algorithms.
We've featured the best cloud storage.
This article was produced as part of TechRadarPro's Expert Insights channel where we feature the best and brightest minds in the technology industry today. The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/news/submit-your-story-to-techradar-pro
Hashtags
Try Our AI Features
Explore what Daily8 AI can do for you:
Comments
No comments yet...
Related Articles
Yahoo
25 minutes ago
- Yahoo
Global M&A hits $2.6 trillion peak year-to-date, boosted by AI and quest for growth
By Emma-Victoria Farr and Amy-Jo Crowley LONDON (Reuters) - Global dealmaking has reached $2.6 trillion, the highest for the first seven months of the year since the 2021 pandemic-era peak, as a quest for growth in corporate boardrooms and the impact of a surge in AI activity has overcome the uncertainty caused by U.S. tariffs. The number of transactions to August 1 is 16% lower than the same time last year, but their value is 28% higher, according to Dealogic data, boosted by U.S. megadeals valued at more than $10 billion. They include Union Pacific Corp's proposed $85 billion acquisition of small rival Norfolk Southern and OpenAI's $40 billion funding round led by Softbank Group. The upsurge will be a relief to bankers who began the year with expectations the administration of U.S. President Donald Trump would lead to a wave of consolidation. Instead, his trade tariffs and geopolitical uncertainty made companies pause until renewed confidence in corporate boardrooms and the U.S. administration's anti-trust agenda changed the mood. "What you're seeing in terms of deal rationale for transactions right now is that it's heavily growth-motivated, and it's increasing," Andre Veissid, EY Global Financial Services Strategy and Transactions Leader, told Reuters. "Whether it's artificial intelligence, the change in the regulatory environment, we see our clients not wanting to be left behind in that race and that's driving activity." Compared with August 2021, when investors, rebounding from pandemic lockdowns drove the value of deals to $3.57 trillion, this year's tally is nearly a $1 trillion, or 27%, lower. Still deal-makers at JPMorgan Chase JP Morgan Chase have said there is more to come, with companies pursuing bigger deals in the second half of the year as executives adapt to volatility. "People have got used to the prevailing uncertainty, or maybe the unpredictability post-U.S. election is just more predictable now," Simon Nicholls, co-head of Slaughter and May Corporate and M&A group, said. Nigel Wellings, Partner at Clifford Chance said the market was moving beyond tariffs. "Boardrooms are seeing the M&A opportunity of a more stable economic environment and positive regulatory signals. But it is not a frothy market." FROM HEALTH TO TECH While the healthcare sector drove M&A in the years after the pandemic, the computer and electronics industry has produced more takeover bids in the U.S. and the United Kingdom in the last two years, according to Dealogic. Artificial intelligence is expected to drive more dealmaking. M&A activity has increased around data centre usage, such as Samsung's $1.7 billion acquisition of Germany's FlaktGroup, a data centre cooling specialist. Palo Alto Networks $25 billion deal for Israeli cybersecurity peer CyberArk was the largest deal in Europe, Middle East and Africa so far this year as rising AI-driven threats push companies to adopt stronger defences. Private equity, which had been sitting on the sidelines, has once again been active, with Sycamore Partners' $10 billion deal to take private Walgreens Boots Alliance and a sweetened $6.4 billion offer from Advent for UK scientific instrument maker Spectris. The U.S. was the biggest market for M&A, accounting for more than half of the global activity. Asia Pacific's dealmaking doubled over the same year to date period last year, outpacing the EMEA region.
Engadget
27 minutes ago
- Engadget
ChatGPT will now remind you to take breaks, following mental health concerns
OpenAI has announced that ChatGPT will now remind users to take breaks if they're in a particularly long chat with AI. The new feature is part of OpenAI's ongoing attempts to get users to cultivate a healthier relationship with the frequently compliant and overly-encouraging AI assistant. The company's announcement suggests the "gentle reminders" will appear as pop-ups in chats that users will have to click or tap through to continue using ChatGPT. "Just Checking In," OpenAI's sample pop-up reads. "You've been chatting for a while — is this a good time for a break?" The system is reminiscent of the reminders some Nintendo Wii and Switch games will show you if you play for an extended period of time, though there's an unfortunately dark context to the ChatGPT feature. The "yes, and" quality of OpenAI's AI and it's ability to hallucinate factually incorrect or dangerous responses has led users down dark paths, The New York Times reported in June — including suicidal ideation. Some of the users whose delusions ChatGPT indulged already had a history of mental illness, but the chatbot still did a bad job of consistently shutting down unhealthy conversations. OpenAI acknowledges some of those shortcomings in its blog post, and says that ChatGPT will be updated in the future to respond more carefully to "high-stakes personal decisions." Rather than provide a direct answer, the company says the chatbot will help users think through problems, offer up questions and list pros and cons. OpenAI obviously wants ChatGPT to feel helpful, encouraging and enjoyable to use, but it's not hard to package those qualities into an AI that's sycophantic. The company was forced to rollback an update to ChatGPT in April that lead the chatbot to respond in ways that were annoying and overly-agreeable. Taking breaks from ChatGPT — and having the AI do things without your active participation — will make issues like that less visible. Or, at the very least, it'll give users time to check whether the answers ChatGPT is providing are even correct.
Fast Company
27 minutes ago
- Fast Company
What's really stopping workers from using AI isn't what you think
From Hollywood to Big Tech, major industries across the U.S. are increasingly going all-in on AI workflow tools, and they're expecting employees to follow suit. Late last month, Business Insider reported that Microsoft has started evaluating some employees on their AI fluency, factoring their competency with AI tools into metrics like performance reviews. But in spite of the growing workplace incentive to adopt AI tools, some employees are actively resisting AI uptake—and their reasons make more sense than you might think. According to a new study conducted by a team of researchers at Peking University and The Hong Kong Polytechnic University, an emerging phenomenon is actively deterring employees from picking up AI tools, even at companies where doing so is strongly encouraged. Dubbed the 'competence penalty,' this bias leads to AI users being seen as less competent by their peers—regardless of actual performance. It's a perception gap that's especially damaging for women in technical roles. The background The researchers' study was conducted at an unnamed leading tech company. In an article written for the Harvard Business Review (HBR), the study's authors explain that this company had previously rolled out a state-of-the-art AI coding assistant to its developers, which was promised to 'boost productivity significantly.' Still, 12 months later, only 41% of the nearly 30,000 surveyed engineers had even tried the coding assistant. Subscribe to the Daily newsletter. Fast Company's trending stories delivered to you every day Privacy Policy | Fast Company Newsletters Adoption also varied based on employees' identities. Just 39% of engineers 40 and older were using the tool, alongside a meager 31% of female engineers. That's not for lack of trying on the company's part, either: Rather than throwing their employees into the AI deep end without guidance (a prevalent issue as AI workflow tools become more common), this company offered dedicated AI teams, adoption incentives, and free training. So, researchers set out to understand what was going wrong. The competence penalty To get to the bottom of this lackluster adoption pattern, the study's authors established an experiment with 1,026 engineers from the same company. The engineers were given a snippet of Python code to evaluate. While the code was the exact same for every participant, each was told that it was created under different conditions—including with or without AI and by a male or female engineer. The results showed that, when participants believed a fellow engineer had used AI to write their code, they rated that engineer's competence 9% lower on average. The competence penalty's severity was also dependent on the reported gender of the engineer. If they were described as male, there was only a 6% competence reduction, compared to 13% for those described as female. Further, the reviewer's own identity and stance on AI had an impact on how they rated others. Engineers who hadn't adopted AI themselves were most critical of AI-users, and male non-adopters penalized female AI-users 26% more harshly than their male AI-using counterparts. Through a follow-up study of 919 engineers, the researchers found that many employees were actually innately aware of this competence penalty, and were avoiding AI usage as a result. advertisement 'Those who most feared competence penalties in the tech industry—disproportionately women and older engineers—were precisely those who adopted AI least,' the study's authors write. 'The very groups who might benefit most from productivity-enhancing tools felt they couldn't afford to use them.' 'Women often face extra scrutiny' The study's findings offer a strong counterpoint to the oft-repeated sentiment that AI tools might even the proverbial playing field at work, presenting a one-size-fits-all solution by making everyone more productive. 'Our results suggest that this is not guaranteed and in fact the opposite could be true,' the authors write. 'In our context, which is dominated by young males, making AI equally available increased bias against female engineers.' These results could help explain patterns that have already been observed in AI uptake. According to recent research conducted by Harvard Business School associate professor Rembrand Koning, women are adopting AI tools at a 25% lower rate than men, on average. In an article for Fast Company earlier this month, Kamales Lardi, author of the book Artificial Intelligence For Business, noted that, 'In my experience, women often face extra scrutiny over their skills, capabilities, and technical prowess. There may be a deep-rooted concern that leveraging AI tools may be perceived as cutting corners or reflect poorly on the users' skill level.' How leaders should prepare for the competence penalty Companies like the one in the study shouldn't give up on implementing new AI tools, especially given that agentic AI is predicted to play a huge role in the future of work. Instead, leaders should use this data to put more AI adoption guardrails in place. In their analysis for HBR, the study's authors offer several main steps for managers to consider: Map your organization's penalty hotspots. Leaders should focus on identifying teams where the AI competence penalty might be highest, including those with more women and older engineers reporting to male non-adopters. Monitoring these teams might help to understand where and how the competence penalty is playing out. Convert the influential skeptics. Because non-adopters are the harshest critics of AI users, influential skeptics can have a major impact on the whole team. The study's authors suggest that breaking this cycle requires the skeptics to see respected colleagues successfully using AI without professional consequence. Redesign evaluations to remove the signal. Based on the study's results, flagging a product as 'made with AI' can negatively impact performance reviews. 'The solution is straightforward: Stop signalling AI use in performance evaluations until your culture is ready,' the authors write.
