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.
Tech
Professional gaming is brutal on the body – inside the growing field of treating esports injuries
Professional gaming can cause serious physical and mental strain, including wrist injuries, back pain, vision problems and sleep disturbances. As esports grows, specialists are developing new prevention and rehabilitation strategies to protect players.
Erica D. Henn, Temple University
Chinese esports player Jian “Uzi” Zi-hao was more than a star “League of Legends” player: He was so talented that his team structured its entire strategy around his abilities, funneling resources to him and another teammate, Shi “Ming” Sen-Ming, so the duo could lead the team to victory. Over the course of his career, he earned more than US$500,000 in tournament prize money.
And yet in 2020, Jian announced his retirement, his storied career cut short by chronic wrist and arm injuries.
He was 25 years old.
Video games have a reputation for keeping people glued to the couch. But for those who compete against the best players in the world, gaming can be surprisingly physically and mentally taxing.
The strain of ‘climbing the ladder’
The best esports players often have grinding schedules. They’re expected to regularly practice, participate in scrimmages and compete in official matches. Even at the recreational level, esports players cite the mental toll: playing for hours a day, every day, to hone their skills so they can “climb the ladder.”
A 2021 study of Portuguese esports players found that 37% experienced anxiety and depression, and 45% experienced sleep disturbances. The most common physical injuries in esports generally fall under eye problems and musculoskeletal issues.
Prolonged screen time leads to many types of vision problems in gamers, from the more common eye fatigue to a decreased ability to focus. Furthermore, the light emitted from screens appears to contribute to sleep disturbances by disrupting the release of melatonin, the hormone that helps regulate your sleep-wake cycle.
Many esport players – professional or otherwise – also battle hand and wrist pain, with repetitive button-smashing causing injuries such as carpal tunnel syndrome, tendonitis and “gamer’s thumb,” which arises from overuse or irritation of the tendons around the thumb and wrist. Many of the overuse injuries seen among video game players are also familiar to assembly-line workers, whose jobs can involve similarly repetitive movements.
Then there are the back injuries. Players can remain seated for three or more hours without a break, and this prolonged sitting can take a toll on the lower back and spine.
The sedentary nature of esports has also led to a lesser-known – sometimes fatal – injury called deep vein thrombosis: a blood clot, often in the leg, that can become life-threatening if it travels to the lungs. In 2011, British gamer Chris Staniforth – who would play for as long as 12 hours at a time – died of the condition.
Preventative measures
As more esports injury research has been published, more treatment and prevention strategies have emerged.
Top esports teams now have physical therapists, performance psychologists, athletic trainers and even massage therapists on staff to optimize the performance and recovery of their players.
Teams often incorporate group exercise activities to both build rapport among players and reduce the risk of injuries. During competition, proper positioning of the spine and limbs has become an essential injury prevention strategy. For example, selecting a chair that encourages an upright posture can reduce pain and injury risk, especially when gamers couple ergonomics with an exercise program that centers on functional strength, mobility and stretching of the upper limbs.
Looking to the stage
When esports began gaining mainstream popularity in the 1990s and early 2000s, its proponents were eager to draw comparisons to traditional sports and athletic competition. The parallels helped establish esports’ legitimacy and gave non-gamers a familiar framework for understanding the competition.
As esports became a big business and a lucrative career path, players and teams hired a web of support staff – trainers, coaches and therapists – that mirrored the structure of professional sports. In this vein, a lot of esports injury research has pulled from the training methods of traditional sports.
However, as a scholar of exercise science, I think injury treatment and prevention strategies could be further improved by seeing esports competitors as more like musicians and dancers than football players and basketball players.
Performance optimization and injury research on professional performing artists has existed for centuries, and I think it represents a valuable, untapped resource. That’s because the physical and mental stresses experienced during musical performance have a lot in common with esports competition: long stretches of sitting; small, dexterous hand movements; and performing without the real-time input of a coach.
For example, biomechanics research has found similar patterns of forearm muscle fatigue among esports players and piano players. However, no studies to date have directly compared the two groups.
And what if the interest in joint hypermobility or hand size among performing artists were translated to esport populations? Could popular piano warm-up exercises be effective for esports athletes who use keyboards?
Even research on sports like car racing might offer valuable insights. As with gamers, many people overlook how physically demanding car racing can be – and yes, that includes sitting for extended periods of time.
Prevention and rehabilitation techniques continue to improve. Even Jian, the player who retired in 2020, returned to play for a few splits, or partial seasons, in 2022 and 2023.
In July 2023, “League of Legends” star Lee “Faker” Sang-hyeok was relegated to the bench due to cubital tunnel syndrome, an injury that emerges from arm and hand overuse.
Through a treatment plan that included changing his gaming posture and intensive physical therapy, Lee was able to return to play just a month later. He went on to win three consecutive world championships, with his support team helping prevent recurrence of injury throughout each season.
Thanks to his rehabilitation, Lee’s fans will be able to follow his hunt for his fourth “League of Legends” World Championship, which kicks off in October 2026 in the United States.
Sienna Cinti assisted with the research and writing of this article.
Erica D. Henn, Assistant Professor of Kinesiology and Exercise Science, Temple University
This article is republished from The Conversation under a Creative Commons license. Read the original article.
📰 Enjoying STM Daily News? Join the conversation!
💬 Leave a comment, share your thoughts, and subscribe to our newsletter for the latest stories, updates, and “News You Can Use This Moment!” delivered to your inbox.
Stay connected with STM Daily News!
News
Joby Aviation and Toyota kick off manufacturing alliance to scale electric air taxi production
Joby Aviation and Toyota launch a joint venture to improve productivity, quality, and cost as they prepare to scale electric air taxi production.
Joby Aviation and Toyota Motor Corporation have launched the initial phase of a strategic manufacturing alliance aimed at accelerating commercial production of electric air taxis—an early step the companies say is designed to make “air mobility for all” a practical, everyday reality.
Announced June 30, 2026, the partnership formalizes a new joint venture that will combine Joby’s electric aviation development with Toyota’s production systems and operational expertise. The near-term focus: building the groundwork for commercial production while pushing improvements in productivity, quality, and cost—key factors as the industry moves from prototypes to scaled manufacturing.
What the joint venture is designed to do
According to the companies, the alliance will initially concentrate on:
- Establishing the foundation for commercial production capability
- Advancing manufacturing excellence with an emphasis on productivity, quality, and cost
- Supporting expansion of Joby’s production capacity as it works toward aircraft certification and prepares for anticipated demand
The announcement positions Toyota’s manufacturing playbook—known globally for lean production and continuous improvement—as a lever to help Joby move from development into repeatable, high-quality output at scale.
Why it matters: eVTOLs need scale, not just flight tests
Electric vertical take-off and landing (eVTOL) aircraft have become one of the most closely watched bets in next-generation transportation, but the path to viable air taxi services depends on more than successful test flights. Certification timelines, supply chain readiness, and the ability to produce aircraft consistently (and affordably) are often what separates promising technology from commercial reality.
By forming a joint venture focused on manufacturing readiness, Joby and Toyota are signaling that the next competitive frontier is industrialization—how quickly and reliably eVTOL aircraft can be built to meet safety standards and market demand.
Related Links for Further reading
- Joby Aviation (official): https://www.jobyaviation.com
- Joby Investor Relations / News (official updates & filings): https://ir.jobyaviation.com
- Toyota Newsroom (official): https://www.toyotanewsroom.com
- Toyota Global (corporate overview): https://global.toyota/en
- FAA Advanced Air Mobility / Air Taxis (context): https://www.faa.gov/air-taxis
What executives are saying
Joby founder and CEO JoeBen Bevirt emphasized the long-running relationship between the companies, calling the joint venture a reflection of shared confidence in the opportunity ahead.
“Toyota has been by Joby’s side for nearly a decade, providing invaluable guidance and support as we built the foundation for manufacturing our aircraft,” Bevirt said. “Together, we share a vision of making aerial mobility an everyday reality.”
Toyota Motor Corporation Chairman Akio Toyoda framed air mobility as an extension of the company’s broader mission.
“Since our founding, we’ve been guided by the philosophy of providing mobility for all,” Toyoda said, adding that Toyota views air mobility as “a natural extension of that philosophy—from the ground into the sky.”
About the companies
Joby Aviation (NYSE: JOBY) is a California-based transportation company developing an all-electric eVTOL air taxi. The company intends to operate its own air taxi service in cities worldwide and sell aircraft to other operators and partners.
Toyota (NYSE: TM) has operated in North America for nearly 70 years and says it is focused on sustainable, next-generation mobility through Toyota and Lexus brands. Toyota reports nearly 64,000 employees in North America, 14 manufacturing plants, and more than 1,800 dealerships. The company also noted that its North Carolina plant began assembling automotive batteries for electrified vehicles in 2025.
What to watch for next
For readers tracking the air taxi sector, the next milestones will likely center on:
- Details on how the joint venture will be structured operationally
- Updates on Joby’s certification progress and production ramp timelines
- Signs of how manufacturing improvements translate into cost reductions and throughput
- Additional agreements or expanded collaboration as the alliance progresses
While the companies highlighted expected benefits, they also noted the usual forward-looking risks—such as regulatory certification timelines, market conditions, and the ability to finalize additional agreements.
Source: Toyota Motor North America / PRNewswire (June 30, 2026)
📰 Enjoying STM Daily News? Join the conversation!
💬 Leave a comment, share your thoughts, and subscribe to our newsletter for the latest stories, updates, and “News You Can Use This Moment!” delivered to your inbox.
Stay connected with STM Daily News!
Consumer Corner
The Evolution of Retail Technology: Connecting Consumers to Valuable Product Information
The Evolution of Retail Technology: Connecting Consumers to Valuable Product Information
(Feature Impact) For more than 50 years, traditional universal product codes (UPCs), better known as barcodes, have automated checkout, powered retail and kept the world’s products moving one scan at a time.
Watch this video to learn more
Now it’s time for the next chapter. Brands and retailers are transitioning to QR codes powered by GS1 to enhance everyday shopping experiences, unlock more information and empower customers to make informed purchase decisions with a simple smartphone scan – while still going “beep” at the register.
For decades, UPC barcodes simply provided the price of an item, but today’s shoppers are looking for more information. This retail-labeling transformation will include advanced QR codes that unlock information about ingredients, allergens, freshness, product origin, sustainability details, recipes and more. Retailers have set a 2027 target to accept these QR codes at checkout, which can help them better operate and serve their customers in numerous ways, such as preventing recalled products from being sold.
They can also help reduce food waste, save consumers money and help people make smarter purchases by encouraging shoppers to have a richer experience with the products they’re putting in their carts.
Visit gs1us.org/smarter to discover more about the future of shopping and checkout.
SOURCE:
📰 Enjoying STM Daily News? Join the conversation!
💬 Leave a comment, share your thoughts, and subscribe to our newsletter for the latest stories, updates, and “News You Can Use This Moment!” delivered to your inbox.
Stay connected with STM Daily News!
