Tech
Why building big AIs costs billions – and how Chinese startup DeepSeek dramatically changed the calculus
Ambuj Tewari, University of Michigan
State-of-the-art artificial intelligence systems like OpenAI’s ChatGPT, Google’s Gemini and Anthropic’s Claude have captured the public imagination by producing fluent text in multiple languages in response to user prompts. Those companies have also captured headlines with the huge sums they’ve invested to build ever more powerful models.
An AI startup from China, DeepSeek, has upset expectations about how much money is needed to build the latest and greatest AIs. In the process, they’ve cast doubt on the billions of dollars of investment by the big AI players.
I study machine learning. DeepSeek’s disruptive debut comes down not to any stunning technological breakthrough but to a time-honored practice: finding efficiencies. In a field that consumes vast computing resources, that has proved to be significant.
Where the costs are
Developing such powerful AI systems begins with building a large language model. A large language model predicts the next word given previous words. For example, if the beginning of a sentence is “The theory of relativity was discovered by Albert,” a large language model might predict that the next word is “Einstein.” Large language models are trained to become good at such predictions in a process called pretraining.
Pretraining requires a lot of data and computing power. The companies collect data by crawling the web and scanning books. Computing is usually powered by graphics processing units, or GPUs. Why graphics? It turns out that both computer graphics and the artificial neural networks that underlie large language models rely on the same area of mathematics known as linear algebra. Large language models internally store hundreds of billions of numbers called parameters or weights. It is these weights that are modified during pretraining. https://www.youtube.com/embed/MJQIQJYxey4?wmode=transparent&start=0 Large language models consume huge amounts of computing resources, which in turn means lots of energy.
Pretraining is, however, not enough to yield a consumer product like ChatGPT. A pretrained large language model is usually not good at following human instructions. It might also not be aligned with human preferences. For example, it might output harmful or abusive language, both of which are present in text on the web.
The pretrained model therefore usually goes through additional stages of training. One such stage is instruction tuning where the model is shown examples of human instructions and expected responses. After instruction tuning comes a stage called reinforcement learning from human feedback. In this stage, human annotators are shown multiple large language model responses to the same prompt. The annotators are then asked to point out which response they prefer.
It is easy to see how costs add up when building an AI model: hiring top-quality AI talent, building a data center with thousands of GPUs, collecting data for pretraining, and running pretraining on GPUs. Additionally, there are costs involved in data collection and computation in the instruction tuning and reinforcement learning from human feedback stages.
All included, costs for building a cutting edge AI model can soar up to US$100 million. GPU training is a significant component of the total cost.
The expenditure does not stop when the model is ready. When the model is deployed and responds to user prompts, it uses more computation known as test time or inference time compute. Test time compute also needs GPUs. In December 2024, OpenAI announced a new phenomenon they saw with their latest model o1: as test time compute increased, the model got better at logical reasoning tasks such as math olympiad and competitive coding problems.
Slimming down resource consumption
Thus it seemed that the path to building the best AI models in the world was to invest in more computation during both training and inference. But then DeepSeek entered the fray and bucked this trend.
Their V-series models, culminating in the V3 model, used a series of optimizations to make training cutting edge AI models significantly more economical. Their technical report states that it took them less than $6 million dollars to train V3. They admit that this cost does not include costs of hiring the team, doing the research, trying out various ideas and data collection. But $6 million is still an impressively small figure for training a model that rivals leading AI models developed with much higher costs.
The reduction in costs was not due to a single magic bullet. It was a combination of many smart engineering choices including using fewer bits to represent model weights, innovation in the neural network architecture, and reducing communication overhead as data is passed around between GPUs.
It is interesting to note that due to U.S. export restrictions on China, the DeepSeek team did not have access to high performance GPUs like the Nvidia H100. Instead they used Nvidia H800 GPUs, which Nvidia designed to be lower performance so that they comply with U.S. export restrictions. Working with this limitation seems to have unleashed even more ingenuity from the DeepSeek team.
DeepSeek also innovated to make inference cheaper, reducing the cost of running the model. Moreover, they released a model called R1 that is comparable to OpenAI’s o1 model on reasoning tasks.
They released all the model weights for V3 and R1 publicly. Anyone can download and further improve or customize their models. Furthermore, DeepSeek released their models under the permissive MIT license, which allows others to use the models for personal, academic or commercial purposes with minimal restrictions.
Resetting expectations
DeepSeek has fundamentally altered the landscape of large AI models. An open weights model trained economically is now on par with more expensive and closed models that require paid subscription plans.
The research community and the stock market will need some time to adjust to this new reality.
Ambuj Tewari, Professor of Statistics, University of Michigan
This article is republished from The Conversation under a Creative Commons license. Read the original article.
STM Daily News is a vibrant news blog dedicated to sharing the brighter side of human experiences. Emphasizing positive, uplifting stories, the site focuses on delivering inspiring, informative, and well-researched content. With a commitment to accurate, fair, and responsible journalism, STM Daily News aims to foster a community of readers passionate about positive change and engaged in meaningful conversations. Join the movement and explore stories that celebrate the positive impacts shaping our world.
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
Tech
When ‘Head in the Clouds’ Means Staying Ahead
Head in the Clouds: Cloud is no longer just storage—it’s the intelligent core of modern business. Explore how “cognitive cloud” blends AI and cloud infrastructure to enable real-time, self-optimizing operations, improve customer experiences, and accelerate enterprise modernization.
Last Updated on February 7, 2026 by Daily News Staff
When ‘Head in the Clouds’ Means Staying Ahead
(Family Features) You approve a mortgage in minutes, your medical claim is processed without a phone call and an order that left the warehouse this morning lands at your door by dinner. These moments define the rhythm of an economy powered by intelligent cloud infrastructure. Once seen as remote storage, the cloud has become the operational core where data, AI models and autonomous systems converge to make business faster, safer and more human. In this new reality, the smartest companies aren’t looking up to the cloud; they’re operating within it. Public cloud spending is projected to reach $723 billion in 2025, according to Gartner research, reflecting a 21% increase year over year. At the same time, 90% of organizations are expected to adopt hybrid cloud by 2027. As cloud becomes the universal infrastructure for enterprise operations, the systems being built today aren’t just hosted in the cloud, they’re learning from it and adapting to it. Any cloud strategy that doesn’t account for AI workloads as native risks falling behind, holding the business back from delivering the experiences consumers rely on every day. After more than a decade of experimentation, most enterprises are still only partway up the curve. Based on Cognizant’s experience, roughly 1 in 5 enterprise workloads has moved to the cloud, while many of the most critical, including core banking, health care claims and enterprise resource planning, remain tied to legacy systems. These older environments were never designed for the scale or intelligence the modern economy demands. The next wave of progress – AI-driven products, predictive operations and autonomous decision-making – depends on cloud architectures designed to support intelligence natively. This means cloud and AI will advance together or not at all.The Cognitive Cloud: Cloud and AI as One System
For years, many organizations treated migration as a finish line. Applications were lifted and shifted into the cloud with little redesign, trading one set of constraints for another. The result, in many cases, has been higher costs, fragmented data and limited room for innovation. “Cognitive cloud” represents a new phase of evolution. Imagine every process, from customer service to supply-chain management, powered by AI models that learn, reason and act within secure cloud environments. These systems store and interpret data, detect patterns, anticipate demand and automate decisions at a scale humans simply cannot match. In this architecture, AI and cloud operate in concert. The cloud provides computing power, scale and governance while AI adds autonomy, context and insight. Together, they form an integrated platform where cloud foundations and AI intelligence combine to enable collaboration between people and systems. This marks the rise of the responsive enterprise; one that senses change, adjusts instantly and builds trust through reliability. Cognitive cloud platforms combine data fabric, observability, FinOps and SecOps into an intelligent core that regulates itself in real time. The result is invisible to consumers but felt in every interaction: fewer errors, faster responses and consistent experiences.Consumer Impact is Growing
The impact of cognitive cloud is already visible. In health care, 65% of U.S. insurance claims run through modernized, cloud-enabled platforms designed to reduce errors and speed up reimbursement. In the life sciences industry, a pharmaceuticals and diagnostics firm used cloud-native automation to increase clinical trial investigations by 20%, helping get treatments to patients sooner. In food service, intelligent cloud systems have reduced peak staffing needs by 35%, in part through real-time demand forecasting and automated kitchen operation. In insurance, modernization has produced multi-million-dollar savings and faster policy issuance, improving both customer experience and financial performance. Beneath these outcomes is the same principle: architecture that learns and responds in real time. AI-driven cloud systems process vast volumes of data, identify patterns as they emerge and automate routines so people can focus on innovation, care and service. For businesses, this means fewer bottlenecks and more predictive operations. For consumers, it means smarter, faster, more reliable services, quietly shaping everyday life. While cloud engineering and AI disciplines remain distinct, their outcomes are increasingly intertwined. The most advanced architectures now treat intelligence and infrastructure as complementary forces, each amplifying the other.Looking Ahead
This transformation is already underway. Self-correcting systems predict disruptions before they happen, AI models adapt to market shifts in real time and operations learn from every transaction. The organizations mastering this convergence are quietly redefining themselves and the competitive landscape. Cloud and AI have become interdependent priorities within a shared ecosystem that moves data, decisions and experiences at the speed customers expect. Companies that modernize around this reality and treat intelligence as infrastructure will likely be empowered to reinvent continuously. Those that don’t may spend more time maintaining the systems of yesterday than building the businesses of tomorrow. Learn more at cognizant.com. Photo courtesy of ShutterstockCulver’s Thank You Farmers® Project Hits $8 Million Donation MilestoneLink: https://stmdailynews.com/culvers-thank-you-farmers-project-hits-8-million-donation-milestone/
Author
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
The Knowledge
Beneath the Waves: The Global Push to Build Undersea Railways
Undersea railways are transforming transportation, turning oceans from barriers into gateways. Proven by tunnels like the Channel and Seikan, these innovations offer cleaner, reliable connections for passengers and freight. Ongoing projects in China and Europe, alongside future proposals, signal a new era of global mobility beneath the waves.
For most of modern history, oceans have acted as natural barriers—dividing nations, slowing trade, and shaping how cities grow. But beneath the waves, a quiet transportation revolution is underway. Infrastructure once limited by geography is now being reimagined through undersea railways.
Undersea rail tunnels—like the Channel Tunnel and Japan’s Seikan Tunnel—proved decades ago that trains could reliably travel beneath the ocean floor. Today, new projects are expanding that vision even further.
Around the world, engineers and governments are investing in undersea railways—tunnels that allow high-speed trains to travel beneath oceans and seas. Once considered science fiction, these projects are now operational, under construction, or actively being planned.
Undersea Rail Is Already a Reality
Japan’s Seikan Tunnel and the Channel Tunnel between the United Kingdom and France proved decades ago that undersea railways are not only possible, but reliable. These tunnels carry passengers and freight beneath the sea every day, reshaping regional connectivity.
Undersea railways are cleaner than short-haul flights, more resilient than bridges, and capable of lasting more than a century. As climate pressures and congestion increase, rail beneath the sea is emerging as a practical solution for future mobility.
What’s Being Built Right Now
China is currently constructing the Jintang Undersea Railway Tunnel as part of the Ningbo–Zhoushan high-speed rail line, while Europe’s Fehmarnbelt Fixed Link will soon connect Denmark and Germany beneath the Baltic Sea. These projects highlight how transportation and technology are converging to solve modern mobility challenges.
The Mega-Projects Still on the Drawing Board
Looking ahead, proposals such as the Helsinki–Tallinn Tunnel and the long-studied Strait of Gibraltar rail tunnel could reshape global affairs by linking regions—and even continents—once separated by water.
Why Undersea Rail Matters
The future of transportation may not rise above the ocean—but run quietly beneath it.
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
CES 2026
Inside the Computing Power Behind Spatial Filmmaking: Hugh Hou Goes Hands-On at GIGABYTE Suite During CES 2026
Inside the Computing Power Behind Spatial Filmmaking: Hugh Hou Goes Hands-On at GIGABYTE Suite During CES 2026
Spatial filmmaking is having a moment—but at CES 2026, the more interesting story wasn’t a glossy trailer or a perfectly controlled demo. It was the workflow.
According to a recent GIGABYTE press release, VR filmmaker and educator Hugh Hou ran a live spatial computing demonstration inside the GIGABYTE suite, walking attendees through how immersive video is actually produced in real-world conditions—capture to post to playback—without leaning on pre-rendered “best case scenario” content. In other words: not theory, not a lab. A production pipeline, running live, on a show floor.
A full spatial pipeline—executed live
The demo gave attendees a front-row view of a complete spatial filmmaking pipeline:
- Capture
- Post-production
- Final playback across multiple devices
And the key detail here is that the workflow was executed live at CES—mirroring the same processes used in commercial XR projects. That matters because spatial video isn’t forgiving. Once you’re working in 360-degree environments (and pushing into 8K), you’re no longer just chasing “fast.” You’re chasing:
- System stability
- Performance consistency
- Thermal reliability
Those are the unsexy requirements that make or break actual production days.
Playback across Meta Quest, Apple Vision Pro, and Galaxy XR
The session culminated with attendees watching a two-minute spatial film trailer across:
- Meta Quest
- Apple Vision Pro
- Newly launched Galaxy XR headsets
- Plus a 3D tablet display offering an additional 180-degree viewing option
That multi-device playback is a quiet flex. Spatial content doesn’t live in one ecosystem anymore—creators are being pulled toward cross-platform deliverables, which adds even more pressure on the pipeline to stay clean and consistent.
Where AI fits (when it’s not the headline)
One of the better notes in the release: AI wasn’t positioned as a shiny feature. It was framed as what it’s becoming for a lot of editors—an embedded toolset that speeds up the grind without hijacking the creative process.
In the demo, AI-assisted processes supported tasks like:
- Enhancement
- Tracking
- Preview workflows
The footage moved through industry-standard software—Adobe Premiere Pro and DaVinci Resolve—with AI-based:
- Upscaling
- Noise reduction
- Detail refinement
And in immersive VR, those steps aren’t optional polish. Any artifact, softness, or weird noise pattern becomes painfully obvious when the viewer can look anywhere.
Why the hardware platform matters for spatial workloads
Underneath the demo was a custom-built GIGABYTE AI PC designed for sustained spatial video workloads. Per the release, the system included:
- AMD Ryzen 7 9800X3D processor
- Radeon AI PRO R9700 AI TOP GPU
- X870E AORUS MASTER X3D ICE motherboard
The point GIGABYTE is making is less “look at these parts” and more: spatial computing workloads demand a platform that can run hard continuously—real-time 8K playback and rendering—without throttling, crashing, or drifting into inconsistent performance.
That’s the difference between “cool demo” and “reliable production machine.”
The bigger takeaway: spatial filmmaking is moving from experiment to repeatable process
By running a demanding spatial filmmaking workflow live—and repeatedly—at CES 2026, GIGABYTE is positioning spatial production as something creators can depend on, not just test-drive.
And that’s the shift worth watching in 2026: spatial filmmaking isn’t just about headsets getting better. It’s about the behind-the-scenes pipeline becoming stable enough that creators can treat immersive production like a real, repeatable craft—because the tools finally hold up under pressure.
Source:PRNewswire – GIGABYTE press release
Welcome to the Consumer Corner section of STM Daily News, your ultimate destination for savvy shopping and informed decision-making! Dive into a treasure trove of insights and reviews covering everything from the hottest toys that spark joy in your little ones to the latest electronic gadgets that simplify your life. Explore our comprehensive guides on stylish home furnishings, discover smart tips for buying a home or enhancing your living space with creative improvement ideas, and get the lowdown on the best cars through our detailed auto reviews. Whether you’re making a major purchase or simply seeking inspiration, the Consumer Corner is here to empower you every step of the way—unlock the keys to becoming a smarter consumer today!
https://stmdailynews.com/category/consumer-corner
Discover more from Daily News
Subscribe to get the latest posts sent to your email.
