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.
aerospace
Boom Supersonic Update 2026: Overture Progress, XB-1 Milestones, and What’s Next
Boom Supersonic’s 2026 update: XB-1 test success, Overture production timeline, funding progress, and the challenges facing the return of commercial supersonic travel.
By STM Daily News Staff
The race to bring back commercial supersonic travel is accelerating once again, led by Boom Supersonic, a Colorado-based aerospace company aiming to succeed where Concorde left off. As of 2026, the company has achieved meaningful technical milestones—but still faces significant financial, regulatory, and industrial hurdles.
Here’s a comprehensive look at where Boom stands today, and what it means for the future of high-speed air travel.
XB-1 Demonstrator Completes Historic Test Program
Boom’s experimental aircraft, the XB-1, has successfully completed its flight test campaign, marking a critical step toward validating the company’s supersonic technology.
- Achieved multiple supersonic flights in 2025
- Demonstrated aerodynamic stability and performance
- Tested “boomless cruise” capabilities to reduce sonic disturbances
The XB-1 program served as a scaled demonstrator for the company’s flagship commercial jet, proving that modern materials, software, and engine integration can support efficient supersonic flight.
With testing complete, the aircraft is expected to be preserved as a prototype, representing a turning point in private-sector aerospace innovation.
Overture: Boom’s Commercial Supersonic Jet
The centerpiece of Boom’s vision is the Overture, a next-generation supersonic passenger aircraft designed to carry between 60 and 80 passengers at speeds approaching Mach 1.7.
Current projected timeline:
- Prototype rollout: Targeted for 2026
- First flight: Expected around 2027
- Commercial service entry: Late 2020s (estimated 2029–2030)
Unlike Concorde, which catered primarily to elite travelers, Boom aims to position Overture with business-class pricing, potentially expanding access to faster global travel.
The aircraft is also being designed with sustainability in mind, including compatibility with sustainable aviation fuel (SAF).
Funding and Financial Momentum
In recent developments, Boom Supersonic secured an additional $100 million in funding, reinforcing investor confidence in the company’s long-term vision.
However, building a supersonic passenger aircraft remains one of the most capital-intensive challenges in aviation. Continued fundraising and strategic partnerships will be essential as the company moves from prototype to production.
Boomless Cruise: A Potential Game-Changer
One of Boom’s most significant innovations is its focus on “boomless cruise,” a method of flying supersonically without producing an audible sonic boom on the ground.
If proven viable at scale, this technology could influence regulatory changes—particularly in the United States, where overland supersonic flight is currently restricted.
The ability to fly faster-than-sound over land would unlock major domestic routes, dramatically reducing travel times between cities like New York and Los Angeles.
Manufacturing Challenges and Delays
Despite technical progress, Boom’s manufacturing ambitions face uncertainty. A planned production facility in North Carolina has experienced delays, raising questions about when large-scale assembly will begin.
Scaling production from prototype to commercial aircraft remains one of the most difficult phases of any aerospace program, requiring supply chain coordination, workforce development, and regulatory alignment.
Industry Skepticism Remains
While Boom has secured interest from major airlines, skepticism persists within the aviation industry.
Key concerns include:
- Certification complexity and regulatory approval timelines
- Operational costs versus ticket pricing
- Long-term demand for supersonic travel
Even airline executives have expressed cautious optimism, with some suggesting the project’s success remains uncertain.
The Bigger Picture: A Defining Decade for Supersonic Travel
Boom Supersonic has moved beyond concept and into real-world testing, demonstrating that modern supersonic flight is technically achievable.
However, the next phase—bringing Overture to market—will determine whether supersonic passenger travel becomes a viable industry once again or remains an ambitious experiment.
If successful, Boom could redefine global travel times. If not, it will join a long list of bold aerospace ventures that struggled to overcome economic reality.
Sources and External Links
- Boom Supersonic – Year in Review
- XB-1 Aircraft Overview
- Overture Aircraft Specifications
- Funding Announcement
- Industry Perspective
Dive into “The Knowledge,” where curiosity meets clarity. This playlist, in collaboration with STMDailyNews.com, is designed for viewers who value historical accuracy and insightful learning. Our short videos, ranging from 30 seconds to a minute and a half, make complex subjects easy to grasp in no time. Covering everything from historical events to contemporary processes and entertainment, “The Knowledge” bridges the past with the present. In a world where information is abundant yet often misused, our series aims to guide you through the noise, preserving vital knowledge and truths that shape our lives today. Perfect for curious minds eager to discover the ‘why’ and ‘how’ of everything around us. Subscribe and join in as we explore the facts that matter. https://stmdailynews.com/the-knowledge/
Space and Tech
I’ve fired one of America’s most powerful lasers – here’s what a shot day looks like
A lead scientist takes you inside the Texas Petawatt at UT Austin, where hours of careful alignment and safety checks build to a single, breath-holding laser shot that briefly creates star-like conditions in a vacuum chamber.
Ahmed Helal, The University of Texas at Austin
If you walk across the open yard in front of the Physics, Math and Astronomy building at the University of Texas at Austin, you’ll see a 17-story tower and a huge L-shaped building. What you won’t see is what’s underneath you. Two floors below ground, behind heavy double doors stamped with a logo that most students have never noticed, sits one of the most powerful lasers in the United States.
I was the lead laser scientist on the Texas Petawatt, or TPW as we called it, from 2020 to 2024. Texas Petawatt, which is currently closed due to funding cuts, was a government-funded research center where scientists from across the country applied for time to use specialized equipment. It was part of LaserNetUS, a Department of Energy network of high-power laser labs.
This type of laser takes a tiny pulse of light, stretches it out so it doesn’t blast optics to pieces, and amplifies it until, for a brief instant, it carries more power than the entire U.S. electrical grid. Then it compresses the pulse back to a trillionth of a second to create a star in a vacuum chamber.
On a typical shot day, the target might be a piece of metal foil thinner than a human hair, a jet of gas or a tiny plastic pellet – each designed to answer a different scientific question.
Scientists from across the country applied for time on TPW to study everything from the physics of stellar interiors and fusion energy to new approaches for cancer treatment.
Most people hear about petawatt lasers and picture something out of a movie. A “shot day” is actually hours of quiet, repetitive work followed by about 10 seconds where nobody breathes.
I now work as a research scientist at the University of Texas-Austin, studying the interaction of lasers with different materials, but a typical shot day during my time running TPW would look like this:
7 a.m.
I arrive two hours before the first scheduled shot. I put on my gown, boots and hairnet and step into the cold clean room. The laser doesn’t just turn on. You coax it awake.
I start with the oscillator, a small box that generates the first seed of light. I write down the parameters that define how the laser will behave during the shot: energy, center frequency, vacuum pressure in the tubes, cooling water level and flow. At this stage, they are fixed regardless of the experiment. The laser must perform the same way every time before the science can begin. Then I fire up the pump laser that will amplify this tiny pulse from nanojoules to about half a joule.
The system needs at least 30 minutes to stabilize. During that time, I check alignment through every pinhole and every camera along the beam path. A slight misalignment at this stage isn’t just a problem; it can be catastrophic – a mispointed beam at full power can burn through optics that take months to source and replace, setting the entire laser back.
Building the beam
Once the system is warmed up, I send the beam into the first amplifier: a glass rod surrounded by bright flash lamps that pump light into the glass – like charging a battery. With each pass, the beam absorbs energy from the glass and grows stronger. Then the beam travels into a larger rod, where it makes four passes, picking up more energy each time until it reaches about 12 joules, roughly the energy of a ball thrown hard across a room.
This process alone takes the better part of an hour, most of it spent checking and confirming alignment and energy at each stage.
I expand the beam and send it through the final stage: the disk amplifiers. Two amplifiers, each consisting of two massive 30-centimeter glass disks, are pumped by a huge bank of flash lamps powered by capacitor banks – essentially giant batteries that store electrical energy and release it in a sudden burst. They are so large that they have their own room on a separate floor. Fast optical shutters between each stage act as gates, controlling exactly when and where the beam travels.
The shot
When the experimental team confirms that the target is in position, it asks me to prepare for a system shot. I run through the long checklist. We test the shutters and switch to system shot mode. Every monitor in the facility changes to display the same message – “System Shot Mode” – and flashes red.
I lean into the microphone at the control desk, a vintage piece that looks like it belongs in a World War II radio room, and announce that we’re going into a system shot. Then I open the compressor beam dump: a heavy glass plate that normally blocks the beam from reaching the target. It takes about two minutes to move.
“Sweeping, sweeping for a system shot.”
The announcement goes out over speakers across the facility. I grab a small interlock key, put on my laser safety goggles and head downstairs. I walk a specific pattern through every room, checking that nobody is still inside. As I go, I lock each door with the key. If anyone opens one of those doors after I’ve locked them, the entire shot sequence aborts.
Back in the control room, I sit down and start charging the capacitor banks. At this point, there’s no going back except for an emergency shutdown, and that means losing the shot and waiting for everything to cool down.
“Charging.”
The room goes silent. Everyone’s eyes are on the monitors. Nobody talks.
I typically will share a glance with the researcher whose project the shot is for – today it’s Joe, a visiting scientist from Los Alamos National Lab, who designed the target we’re about to vaporize. He’s gripping his coffee cup like it owes him money. I turn back to the console.
“Charge complete. Firing system shot in three, two, one. Fire.”
I press the button. A loud thud rolls through the building as all that stored energy dumps into the beam. The monitors freeze, capturing everything at the moment of the shot: beam profiles, spectra, diagnostics – these metrics provide a full picture of exactly how the laser performed and whether the shot was clean. Downstairs, in the vacuum chamber, a spot smaller than a human hair just reached temperatures measured in millions of degrees.
I lean back in my chair and start recording laser parameters as everyone exhales. A radiation safety officer heads down first to check readings around the target chamber before anyone else can enter. The experimental team follows to collect data.
Sometimes it all works perfectly. Sometimes a shutter fails to open and you lose the shot.
For example, one afternoon in 2023, we’d spent three hours preparing for a high-priority shot. Target aligned. Capacitors charged. I pressed the button and heard nothing. A shutter had failed somewhere in the chain. The monitors stayed frozen, showing black. Nobody said anything. I wrote SHOT FAILED in the logbook and started the hourlong cooldown sequence. That’s the part they don’t show in movies: sitting in silence, waiting to try again. We got the shot four hours later.
This anticipation is all part of the job: hours of patience for 10 seconds you never quite get used to. Everything happens underneath a campus where thousands of people walk above, unaware that for a fraction of a second, a tiny point of matter hotter than the surface of the Sun just existed below their feet.
Ahmed Helal, Research Scientist, The University of Texas at Austin
This article is republished from The Conversation under a Creative Commons license. Read the original article.
The science section of our news blog STM Daily News provides readers with captivating and up-to-date information on the latest scientific discoveries, breakthroughs, and innovations across various fields. We offer engaging and accessible content, ensuring that readers with different levels of scientific knowledge can stay informed. Whether it’s exploring advancements in medicine, astronomy, technology, or environmental sciences, our science section strives to shed light on the intriguing world of scientific exploration and its profound impact on our daily lives. From thought-provoking articles to informative interviews with experts in the field, STM Daily News Science offers a harmonious blend of factual reporting, analysis, and exploration, making it a go-to source for science enthusiasts and curious minds alike. https://stmdailynews.com/category/science/
Science
New Glenn’s Third Mission Set for April 19 as Blue Origin Advances Commercial Space Capabilities
CAPE CANAVERAL, Fla. — Blue Origin has confirmed the launch window for the third mission of its heavy-lift New Glenn rocket, marking another step forward in the company’s expanding role in commercial spaceflight.
New Glenn’s Third Mission
Launch Details and Timeline
The mission is scheduled to lift off no earlier than Sunday, April 19, 2026, from Launch Complex 36 at Cape Canaveral Space Force Station. The two-hour launch window opens at 6:45 a.m. EDT (10:45 UTC) and closes at 8:45 a.m. EDT (12:45 UTC).
Viewers can follow the mission through a live webcast hosted by Blue Origin, beginning approximately 30 minutes before liftoff.
Mission Payload: Expanding Space-Based Connectivity
At the heart of the mission is the deployment of the BlueBird 7 satellite, developed by AST SpaceMobile. The satellite is designed to enhance a growing direct-to-smartphone broadband network, an emerging technology aimed at delivering connectivity to standard mobile devices without the need for ground-based towers.
BlueBird 7 will contribute to expanding network capacity and is expected to support initial service rollout plans targeted for 2026. The broader initiative reflects a significant shift in how satellite infrastructure could complement terrestrial telecom systems, particularly in underserved or remote regions.
Reusability Milestone: Booster Returns Again
A key feature of this mission is the planned reuse of New Glenn’s first-stage booster, “Never Tell Me The Odds.” The booster previously demonstrated a successful launch and landing during the rocket’s second mission in November, underscoring Blue Origin’s commitment to reusable rocket technology—a cornerstone of cost reduction and operational efficiency in modern spaceflight.
If successful, this mission will further validate the reliability of the New Glenn system and strengthen its competitiveness in a market increasingly shaped by reusable launch vehicles.
Industry Context: Competing in a Rapidly Evolving Market
The New Glenn program represents Blue Origin’s answer to heavy-lift launch demands, positioning the company alongside major players such as SpaceX. As satellite constellations grow in scale and ambition, reliable and cost-effective launch services have become a critical component of the global space economy.
The inclusion of commercial payloads like BlueBird 7 highlights the increasing collaboration between aerospace firms and telecommunications providers, signaling a future where space-based infrastructure plays a central role in everyday connectivity.
Looking Ahead
With its third mission, New Glenn continues to build momentum as a next-generation launch platform. The combination of reusable hardware, commercial partnerships, and advanced payload capabilities places this launch among the most closely watched developments in the 2026 spaceflight calendar.
For ongoing updates, mission tracking, and live coverage, audiences can follow Blue Origin across its digital platforms or visit its official website.
Source
Blue Origin Official Announcement – New Glenn Third Mission
Related External Links
- Learn More About Blue Origin’s New Glenn Rocket
- AST SpaceMobile – Space-Based Cellular Broadband Network
- Cape Canaveral Space Force Station Information
- NASA Overview of Low Earth Orbit (LEO) Operations
Explore the latest in innovation, AI, gadgets, startups, and digital trends in STM Daily News’ Techsection.
