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.
FOX Corporation says it has reached a definitive agreement to acquire Roku in a cash-and-stock deal valued at $160 per share, putting Roku at roughly $22 billion in enterprise value. On paper, it’s a classic “content meets platform” move: FOX brings premium live programming (sports, news, entertainment) and Tubi, while Roku brings the connected TV operating system, The Roku Channel, and a direct relationship with more than 100 million streaming households.
For STM Daily News readers, the big question isn’t the stock math—it’s the practical one: what changes for people who already own a Roku device or use Roku’s service? Here’s what the companies are saying, what’s likely, and what to watch as the deal heads toward a targeted close in the first half of 2027.
The headline: FOX wants the “front door” to streaming
Roku isn’t just a streaming stick. It’s the home screen millions of people see every day—the place where apps are discovered, promoted, and monetized. FOX is betting that pairing its live content (especially sports and news) with Roku’s platform and ad tech creates a scaled media-and-technology business with stronger reach and advertising power.
FOX and Roku also emphasized that Roku will continue operating as an “open, partner-friendly platform,”and that FOX content will remain widely distributed. That’s an important promise—because Roku’s value depends on being a neutral platform that works with everyone.
What this could mean for Roku owners (the consumer view)
1) Your Roku device should keep working—no “sudden shutdown” expected
Nothing in the announcement suggests existing Roku players or Roku TVs will stop functioning. In most acquisitions like this, the priority is stability: keep devices running, keep accounts intact, keep app availability broad. Roku’s installed base is the asset.
What to watch for: changes to software update cadence, account terms, or how the home screen is organized.
2) Expect tighter FOX + Roku integration (and more promotion)
If FOX owns Roku, it can promote FOX properties more aggressively across the Roku interface—think:
- More prominent placement for Tubi and The Roku Channel
- Faster paths to live FOX events (sports, breaking news)
- Bundled sign-ups or simplified authentication
This could be convenient for viewers who already watch FOX content. It could also feel like “more FOX everywhere” if the home screen starts prioritizing FOX-owned services.
What to watch for: whether Roku’s home screen recommendations become noticeably more FOX-heavy.
3) Advertising could get smarter—and more intense
Both companies highlighted reach, engagement, and monetization. Roku’s first-party data and ad platform are a major part of the appeal. FOX’s live sports and news are premium ad environments. Put together, the combined company will likely push for:
- More advanced ad targeting and measurement across streaming
- More ad inventory tied to live events
- Stronger cross-promotion between linear TV and streaming
What to watch for: ad load (how many ads you see), frequency (how often you see the same ad), and new ad formats.
4) The Roku Channel and Tubi could become a bigger “free TV” hub
Roku already operates The Roku Channel, and FOX owns Tubi—two major free, ad-supported streaming services (FAST). A combined strategy could mean:
- More shared content pipelines
- Expanded live channels
- A clearer “free streaming” destination inside the Roku ecosystem
What to watch for: whether the services stay distinct or begin to merge features, libraries, or branding.
5) App availability is the make-or-break issue
Roku’s strength comes from being the platform where all the major services want to be. If partners believe the platform is no longer neutral, negotiations can get tense.
FOX and Roku say they intend to keep Roku open and partner-friendly. That’s a signal to streaming services, device makers, and advertisers: “we’re not closing the ecosystem.”
What to watch for: any public disputes over app placement, revenue share, data access, or carriage terms.
What the deal terms tell us (and why it matters)
FOX says it expects the deal to be accretive to free cash flow per share by the second full year after closing and targets about $400 million in run-rate cost synergies, with additional revenue upside. Translation: there will be pressure to streamline operations and increase monetization.
FOX also plans to fund the cash portion with new debt and cash on hand, with a pro forma net leverage expectation of about 2.8x (including partial credit for synergies). That kind of financing structure typically increases the importance of predictable cash generation—often from advertising and platform economics.
Timeline: nothing changes overnight
The transaction still needs shareholder approvals and U.S. and non-U.S. regulatory approvals, and the companies expect to close in the first half of 2027. That means the Roku experience you have today is likely to remain largely the same in the near term.
Bottom line: convenience vs. control
For consumers, this deal is a tug-of-war between two outcomes:
- Convenience: easier access to FOX content, stronger free streaming options, and a more integrated experience.
- Control: more aggressive promotion, more advertising optimization, and potential shifts in platform neutrality.
If you’re a Roku owner, the best move right now is simple: keep an eye on interface changes and terms-of-service updates as the deal progresses. The “what to watch for” items above will be the early signals of whether this becomes a viewer-friendly upgrade—or a more tightly monetized streaming front door.
What to watch for next
- Regulatory review updates and any conditions attached to approval
- How FOX positions Tubi vs. The Roku Channel
- Any changes to Roku’s partner relationships (major app negotiations)
- New product announcements tied to live sports/news streaming
Source (press release):
Fox Corporation via PRNewswire — “FOX CORPORATION TO ACQUIRE ROKU, INC.” (June 15, 2026)
Related external links (as referenced in the release):
- SEC filings portal: https://www.sec.gov
- FOX Investor Relations: https://investor.foxcorporation.com/
- Roku Investor Relations: https://www.roku.com/investor
- Fox Corporation (company site): https://www.foxcorporation.com/
STM Daily News will continue tracking what this acquisition means for cord-cutters, connected TV users, and the future of streaming discovery.
Automotive
EPA removal of vehicle emissions limits won’t stop the shift to electric vehicles, but will make it harder, slower and more expensive
The EPA’s move to rescind the 2009 “endangerment finding” and roll back vehicle emissions limits won’t stop the shift to electric vehicles—but it will slow adoption, raise costs, and increase climate and public health harms.
Alan Jenn, University of California, Davis
The U.S. government is in full retreat from its efforts to make vehicles more fuel-efficient, which it had been prioritizing, along with state governments, since the 1970s.
The latest move came on Feb. 12, 2026, when President Donald Trump and the Environmental Protection Agency issued a new rule rescinding the landmark “endangerment finding,” and reversing various emissions limits on cars and trucks. The 2009 finding stated that greenhouse gases pose a threat to public health and welfare. If the new rule stands up in court and is not overruled by Congress, it would undo a key part of the long-standing effort to limit greenhouse gas emissions from vehicles.
As a scholar of how vehicle emissions contribute to climate change, I know that the science behind the endangerment finding hasn’t changed. If anything, the evidence has grown that greenhouse gas emissions are warming the planet and threatening people’s health and safety. Heat waves, flooding, sea-level rise and wildfires have only worsened in the decade and a half since the EPA’s ruling.
Regulations over the years have cut emissions from power generation, leaving transportation as the largest source of greenhouse gas emissions in the U.S.
The scientific community agrees that vehicle emissions are harmful and should be regulated. The public also agrees, and has indicated strong preferences for cars that pollute less, including both more efficient gas-burning vehicles and electric-powered ones. Consumers have also been drawn to electric vehicles thanks to other benefits such as performance, operation cost and innovative technologies.
That is why I believe the EPA’s move will not stop the public and commercial transition to electric vehicles, but it will make that shift harder, slower and more expensive for everyone.
Putting carmakers in a bind
The most recent EPA rule about vehicle emissions was finalized in 2024. It set emissions limits that can realistically only be met by a large-scale shift to electric vehicles.
Over the past decade and a half, automakers have been building up their capability to produce electric vehicles to meet these fleet requirements, and a combination of regulations such as California’s zero-emission-vehicle requirements have worked together to ensure customers can get their hands on EVs. The zero-emission-vehicle rules require automakers to produce EVs for the California market, which in turn make it easier for the companies to meet their efficiency and emissions targets from the federal government. These collectively pressure automakers to provide a steady supply of electric vehicles to consumers.
The new EPA move would undo the 2024 EPA vehicle-emissions rule and other federal regulations that also limit emissions from vehicles, such as the heavy-duty vehicle emissions rule.
The possibility of a regulatory reversal puts automakers into a state of uncertainty. Legal challenges to the EPA’s shift are all but guaranteed, and the court process could take years.
For companies making decade-long investment decisions, regulatory stability matters more than short-term politics. Disrupting that stability undermines business planning, erodes investor confidence and sends conflicting signals to consumers and suppliers alike.
A slower roll
The Trump administration has taken other steps to make electric vehicles less attractive to carmakers and consumers.
The White House has already suspended key provisions of the Inflation Reduction Act that provided tax credits for purchasing EVs and halted a US$5 billion investment in a nationwide network of charging stations. And Congress has retracted the federal waiver that allowed California to set its own, stricter emissions limits. In combination, these policies make it hard to buy and drive electric vehicles: Fewer, or no, financial incentives for consumers make the purchases more expensive, and fewer charging stations make travel planning more challenging.
Overturning the EPA’s 2009 endangerment finding would remove the legal basis for regulating climate pollution from vehicles altogether.
But U.S. consumer interest in electric vehicles has been growing, and automakers have already made massive investments to produce electric vehicles and their associated components in the U.S. – such as Hyundai’s EV factory in Georgia and Volkswagen’s Battery Engineering Lab in Tennessee.
Global markets, especially in Europe and China, are also moving decisively toward electrifying large proportions of the vehicles on the road. This move is helped in no small part due to aggressive regulation by their respective governments. The results speak for themselves: Sales of EVs in both the European Union and China have been growing rapidly.
But the pace of change matters. A slower rollout of clean vehicles means more cumulative emissions, more climate damage and more harm to public health.
The EPA’s move seeks to slow the shift to electric vehicles, removing incentives and raising costs – even though the market has shown that cleaner vehicles are viable, the public has shown interest, and the science has never been clearer. But even such a major policy change can’t stop the momentum of those trends.
This is an updated version of an article originally published Aug. 5, 2025.
Alan Jenn, Associate Professor of Civil and Environmental Engineering, University of California, Davis
This article is republished from The Conversation under a Creative Commons license. Read the original article.
Stay ahead of the curve with STM Daily News’ Tech section, featuring the latest on innovation, consumer technology, digital trends, startups, AI, and the stories shaping how we live and work.
Consumer Corner
How agentic AI is changing the way contractors assess storm damage
Hours after a storm, teams of professionals need to move quickly to assess the damage. That work used to take days of site visits and time-consuming analyses. Now, a roofing contractor can pull up a color-coded map of every roof over 15 years old within two miles with agentic tools.
How agentic AI is changing the way contractors assess storm damage
(Tiffany Miller) Hours after a storm, teams of professionals need to move quickly to assess the damage. That work used to take days of site visits and time-consuming analyses.
Now, a roofing contractor can pull up a color-coded map of every roof over 15 years old within two miles with agentic tools. An insurance claims manager can see which homes have the worst damage and need to be inspected first. A government assessor can map which neighborhoods were hit hardest. As hurricane season opens June 1, Eagleview Horizon, an agentic AI engine, shows how predictive AI is giving these professionals a head start.
Why the trades are under pressure
Billion-dollar weather and climate disasters in the U.S. have nearly tripled since the 2000s, now averaging 19 per year, according to NOAA. The trades are paying for it. Insurance carriers are required to handle more claims. Contractors race to keep up with demand. And government planners assess wider damage zones due to population increases. For insurers, that pressure is already showing. AM Best reported that the homeowners insurance segment had its toughest first quarter in five years, driven by January wildfires in California and tornado outbreaks across the Midwest, Southern and Plains states. That hit a property and casualty insurance market where U.S. premiums crossed $1.05 trillion in 2024, according to S&P Global Market Intelligence.
For the majority of the 100,000-plus contractors in the $25 billion U.S. roofing industry, the work hasn’t changed much. They still spend hours canvassing neighborhoods street by street. Adjusters drive door-to-door. Government planners wait for damage reports to come in. As storm seasons grow more intense and claims volume rises, that model is starting to crack.
A shift across multiple industries
AI can find information faster, but newer systems can also turn a question into a workflow.
For early adopters, that is changing more than storm response. Commercial roofers can use these tools to pinpoint structural changes and complete annual inspections more safely. Infrastructure managers can track changes over time to flag maintenance needs. Property managers can identify which assets in large portfolios are approaching maintenance risk before they fail.
What may matter most is who can do this work. Until recently, sophisticated property analysis required trained specialists and could take weeks. Now any contractor, adjuster, assessor or planner can begin with a plain question and get a clearer view of where to act first.
What this means for storm season
Trades will keep facing issues caused by more storms, and time will tell whether new tools can spread fast enough to meet the overwhelming demands of the industry.
Photo courtesy of Shutterstock
SOURCE:
Eagleview
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
