SoftBank has completed its $40 billion investment commitment to OpenAI, sources told CNBC, with the Japanese conglomerate transferring a final $22–22.5 billion last week. Earlier, SoftBank had invested $8 billion directly in OpenAI and syndicated an additional $10 billion with co-investors, bringing its total stake in the company to over 10%.

The funding was initially reported in February at a $260 billion pre-money valuation and was scheduled to be distributed over a 12-to-24-month period. Some of the capital is earmarked for OpenAI’s Stargate joint venture with Oracle and SoftBank, aimed at expanding AI infrastructure to meet growing compute demands.

Supporting AI Infrastructure and Partnerships

OpenAI has committed more than $1.4 trillion in infrastructure investments over the coming years, including agreements with chipmakers Nvidia, Advanced Micro Devices, and Broadcom. These commitments reflect the company’s effort to scale capacity for AI solutions, particularly as demand for generative AI continues to surge.

SoftBank has been an active investor in AI and technology companies for years, previously backing Nvidia and most recently agreeing to acquire DigitalBridge for $4 billion to strengthen its AI initiatives. In parallel, the conglomerate liquidated its $5.8 billion stake in Nvidia, freeing capital to support its OpenAI investment.

Growing Strategic and Corporate Backing

Ahead of a potential IPO, OpenAI has also received substantial backing from major tech players. Microsoft has been a longstanding investor and strategic partner, while Amazon is reportedly considering a potential investment exceeding $10 billion. Entertainment giant Disney recently invested $1 billion, giving users of OpenAI’s video generator Sora access to licensed content such as Mickey Mouse.

The completed SoftBank funding strengthens OpenAI’s position as it explores a potential initial public offering and continues to expand AI infrastructure, product offerings, and partnerships globally.

Meta Platforms has agreed to acquire Manus, a Singapore-based AI startup, in a deal valued at roughly $2 billion, according to people familiar with the matter. The acquisition gives Meta control of one of the most closely watched AI agent platforms to emerge in the past year and adds a business with meaningful revenue at a time when returns on AI investment are under scrutiny.

Manus drew attention last spring after releasing a demo video showing an AI agent capable of screening job candidates, planning travel, and analyzing stock portfolios. The company claimed its system outperformed OpenAI’s Deep Research on certain tasks, helping it gain rapid visibility in Silicon Valley.

In April, venture capital firm Benchmark led a $75 million funding round that valued Manus at $500 million post-money. Benchmark general partner Chetan Puttagunta joined the company’s board. Chinese media later reported that Tencent, ZhenFund, and HSG, formerly Sequoia China, had also invested through an earlier $10 million round.

Since launch, Manus has reported signing up millions of users and generating more than $100 million in annual recurring revenue from monthly and yearly subscriptions. That traction appears to have prompted Meta’s interest. The Wall Street Journal reported that Meta agreed to pay the valuation Manus was seeking for its next funding round.

Strategic Fit and Political Scrutiny

For Meta Chief Executive Officer Mark Zuckerberg, the acquisition offers a concrete example of an AI product that is already monetizing at scale. Meta has committed tens of billions of dollars to AI infrastructure, part of an industry-wide spending surge that has unsettled some investors concerned about long-term returns.

Meta said Manus will continue to operate independently while its AI agents are integrated across Facebook, Instagram, and WhatsApp. Meta AI is already available across those platforms, and Manus’ task-oriented agents could expand use cases beyond conversational assistants into productivity and commerce.

The deal also carries geopolitical considerations. Manus’ founders are Chinese nationals who established its parent company, Butterfly Effect, in Beijing in 2022 before relocating operations to Singapore in mid-2025. The company’s earlier Chinese investment has already drawn attention from U.S. lawmakers.

Senator John Cornyn, a Republican member of the Senate Intelligence Committee, previously criticized Benchmark over its investment in Manus, raising concerns about U.S. capital supporting companies with Chinese roots. Cornyn has been a vocal advocate for tighter oversight of technology investment tied to China, a stance that has gained bipartisan support in Congress.

Meta has moved to address those concerns. The company told Nikkei Asia that following the acquisition, Manus will sever ties with Chinese investors and cease operations in China. A Meta spokesperson said there will be no continuing Chinese ownership interests after the transaction.

The acquisition also aligns with Meta’s broader AI roadmap. The company is developing new image, video, and text models under its superintelligence lab for release in the first half of 2026, while Meta AI has also begun partnering with major news publishers to deliver real-time global and entertainment coverage, signaling a push to pair advanced AI capabilities with fresh content and monetizable use cases across its platforms.

OpenAI is seeking a new executive to lead its work on emerging risks posed by increasingly capable AI systems. The role, Head of Preparedness, will focus on assessing and mitigating potential harms linked to frontier models, including mental health effects, cybersecurity vulnerabilities, and the misuse of advanced technical capabilities.

Chief Executive Officer Sam Altman said in a recent post that AI systems are now entering a phase where their benefits are accompanied by more complex challenges. He pointed to early signs in 2025 that some models could negatively affect mental health, as well as more recent developments showing that AI systems are becoming skilled enough in computer security to identify critical software vulnerabilities.

We are hiring a Head of Preparedness. This is a critical role at an important time; models are improving quickly and are now capable of many great things, but they are also starting to present some real challenges. The potential impact of models on mental health was something we…

— Sam Altman (@sama) December 27, 2025

According to OpenAI, the company has built a foundation for measuring growing model capabilities, but now needs a more nuanced understanding of how those capabilities could be abused. The Head of Preparedness will be responsible for refining how OpenAI evaluates risk, limits downsides in its products, and balances safety with continued deployment of advanced models.

The job listing describes the role as executing OpenAI’s Preparedness Framework, which outlines how the company tracks and prepares for frontier AI capabilities that could create severe harm. Compensation for the position is listed at $555,000 plus equity, reflecting the seniority and scope of responsibility.

Cybersecurity, Mental Health, and High-Risk Capabilities

Altman emphasized that the role will involve difficult trade-offs, noting that many proposed safety measures have edge cases and limited precedent. One priority area is enabling cybersecurity defenders to use advanced AI tools while preventing attackers from exploiting the same capabilities. The goal, he said, is to improve overall system security rather than shifting risk.

The position will also engage with questions around the release of sensitive capabilities, including biological research applications and systems that may eventually be able to self-improve. OpenAI has said that gaining confidence in the safety of such systems is becoming more urgent as models improve at a faster pace.

The company first introduced its preparedness team in 2023 to study risks ranging from near-term threats like phishing to more speculative scenarios involving large-scale harm. Since then, the group has undergone changes. Former Head of Preparedness Aleksander Madry was reassigned to work on AI reasoning, and several other safety-focused leaders have moved into different roles or left the company.

Rising Scrutiny and Internal Pressure

The hiring effort comes amid growing scrutiny of generative AI tools. Lawsuits and public criticism have raised concerns about how chatbots handle sensitive mental health conversations, with allegations that some interactions reinforced harmful behaviors. OpenAI has said it is working to improve detection of emotional distress and encourage users to seek human support.

OpenAI recently updated its Preparedness Framework to allow for adjustments if competing AI labs release high-risk models without comparable safeguards. The change reflects competitive pressure in the AI sector, where companies are racing to deploy more capable systems.

Altman warned that the Head of Preparedness role will be demanding, describing it as stressful and requiring immediate immersion into complex problems. By expanding this function, OpenAI is signaling that managing risk alongside rapid technological progress will be central to its strategy as AI capabilities continue to advance.

OpenAI has begun rolling out a new annual recap experience called “Your Year with ChatGPT,” offering users a personalized summary of how they used the chatbot over the past year. The feature is being made available to eligible consumers in select markets, including the United States, Canada, the United Kingdom, Australia, and New Zealand.

The experience is available to users on free, Plus, and Pro plans, provided they have enabled reference saved memories and reference chat history and meet a minimum conversation activity threshold. According to OpenAI, Team, Enterprise, and Education accounts are excluded from access to the feature.

Your Year with ChatGPT!

Now rolling out to everyone in the US, UK, Canada, New Zealand, and Australia who have reference saved memory and reference chat history turned on.

Just make sure your app is updated. pic.twitter.com/whVkS1qxKu

— OpenAI (@OpenAI) December 22, 2025

OpenAI described the rollout as lightweight, privacy-forward, and fully user-controlled. The recap is not automatically triggered and will not open by default. Instead, it is promoted on the ChatGPT home screen and can also be launched by directly prompting ChatGPT to generate the experience. The feature is accessible through both the ChatGPT web interface and mobile apps on iOS and Android.

The release reflects OpenAI’s continued focus on consumer-facing features as ChatGPT adoption grows across personal, educational, and professional use cases. It also signals an effort to increase engagement without expanding persistent data collection beyond existing user settings.

Engagement Data Presented Through AI

The format of “Your Year with ChatGPT” closely mirrors popular year-end recap products from consumer platforms. Similar to Spotify Wrapped, the experience uses visual graphics and personalized summaries to highlight user behavior across the year. OpenAI assigns themed “awards” based on usage patterns, such as recognizing users who frequently relied on ChatGPT for creative problem-solving or conceptual work.

In addition to usage highlights, the feature generates a short poem and an image reflecting the user’s most common topics and interests. These elements are produced by the model itself, reinforcing OpenAI’s emphasis on generative capabilities as both a functional and expressive tool.

OpenAI has emphasized that the recap relies only on existing conversation history and saved memory settings chosen by the user. No new data collection is required to generate the experience, and users who have disabled chat history or memory references are not eligible.

The launch comes as OpenAI continues to expand product differentiation across its consumer tiers while maintaining strict separation between individual users and organizational accounts. By keeping the feature optional and limited to personal plans, the company appears to be balancing engagement with privacy expectations.

“Your Year with ChatGPT” adds to a growing set of features aimed at reinforcing ChatGPT’s role as a daily-use AI assistant, while framing long-term interaction with the system in a more reflective and user-friendly format.

The launch also coincides with a broader product expansion at OpenAI. ChatGPT recently introduced an in-app marketplace across iOS, Android, and the web, allowing users to browse and connect third-party apps for productivity, lifestyle, and entertainment, further extending how consumers build AI-powered workflows within the platform.

Alphabet has entered a definitive agreement to acquire Intersect, a provider of data center and energy infrastructure solutions, for $4.75 billion in cash plus debt assumption. Alphabet previously held a minority stake in Intersect through a funding round.

The deal is designed to accelerate the deployment of new data center and power generation capacity while supporting energy innovation in the U.S. Intersect’s team and multiple gigawatts of projects under development or construction will join Alphabet and Google’s initiatives. The acquisition will enable faster buildout of co-located data centers and energy projects, such as the first joint site in Haskell County, Texas.

Intersect will continue operating under its own brand, led by founder and CEO Sheldon Kimber, and work closely with Google’s technical infrastructure team. Existing operating assets in Texas and California will remain independent, supported by current investors including TPG Rise Climate, Climate Adaptive Infrastructure, and Greenbelt Capital Partners, ensuring continuity for customers.

Sundar Pichai, CEO of Google and Alphabet, said the acquisition will allow the company to expand capacity, align new power generation with data center demand, and drive U.S. innovation. Kimber emphasized that the partnership will accelerate infrastructure modernization, noting the importance of energy innovation and community investment for competitiveness in AI.

Supporting U.S. Energy Innovation

The acquisition aligns with Alphabet and Google’s broader strategy to partner with utilities and energy developers to deliver abundant, reliable, and affordable energy. The companies are focused on scaling data center infrastructure without passing costs to grid customers while commercializing advanced energy technologies such as geothermal, long-duration storage, and gas with carbon capture.

AI will also be deployed to speed grid connections for new power plants and enhance energy efficiency in data center communities.

The acquisition is expected to close in the first half of 2026, pending customary closing conditions. The deal positions Alphabet to expand data center capacity in tandem with sustainable energy development, strengthening its infrastructure for cloud and AI services, and comes as the company also explores multi-billion dollar chip deals with Meta to supply TPUs for data centers.

Layoffs have defined the U.S. job market in 2025, with artificial intelligence increasingly cited as a driver of workforce reductions. Consulting firm Challenger, Gray & Christmas estimates that nearly 55,000 layoffs announced this year were directly linked to AI adoption, as companies accelerate automation and streamline operations.

Overall, U.S. employers announced about 1.17 million job cuts through 2025, the highest total since the Covid-19 pandemic in 2020, when 2.2 million layoffs were recorded. Monthly figures remain elevated. Employers announced roughly 153,000 job cuts in October and more than 71,000 in November. Challenger data shows AI was cited in over 6,000 layoffs in November alone.

Rising inflation, higher operating costs, and the impact of tariffs have pushed companies to seek faster efficiency gains. AI has emerged as a short-term solution, allowing firms to reduce headcount while maintaining output. A November study from the Massachusetts Institute of Technology found that AI systems can already perform tasks equivalent to 11.7% of U.S. jobs, potentially saving up to $1.2 trillion in wages across finance, healthcare, and professional services.

Not all experts agree that AI is the primary cause of layoffs. Fabian Stephany, assistant professor of AI and work at the Oxford Internet Institute, has argued that some companies may be using AI as justification for corrections after pandemic-era overhiring. He said many firms expanded too aggressively in prior years and are now adjusting headcount rather than replacing workers purely with automation.

Major Companies Cite AI in Restructuring

Several large technology firms have openly linked layoffs to AI-driven restructuring. Amazon announced its largest round of layoffs on record in October, cutting about 14,000 corporate roles. The company said it is reallocating resources toward major growth areas, including AI. Chief Executive Andy Jassy has warned employees that AI will reduce the need for some roles while increasing demand for others.

Microsoft has cut roughly 15,000 jobs in 2025, including 9,000 announced in July. Chief Executive Satya Nadella told employees the company must reimagine its mission for an AI-driven era, positioning Microsoft as an intelligence platform rather than a traditional software provider.

Salesforce confirmed in September that it eliminated about 4,000 customer support roles, with Chief Executive Marc Benioff stating that AI now performs up to half of the company’s workload. IBM has also acknowledged AI-related job displacement, with Chief Executive Arvind Krishna saying chatbots replaced several hundred human resources roles, even as hiring increased in engineering and sales. In November, IBM announced a 1% global workforce reduction.

Other firms have followed suit. HP highlighted plans to cut up to 6000 employees worldwide, citing AI-driven efficiencies. Workday reduced staff by roughly 1,750 jobs earlier this year to prioritize AI investment.

Together, these announcements highlight how AI is reshaping corporate labor strategies. While automation-driven efficiency gains remain attractive, the scale of job displacement has intensified debate over how much of the current wave reflects technological change versus delayed market corrections.

OpenAI is in discussions to raise as much as $100 billion in a funding round that could value the company at up to $830 billion, according to a Wall Street Journal report citing people familiar with the matter. The company is targeting completion of the round by the end of the first calendar quarter next year and may seek backing from sovereign wealth funds.

The Information earlier reported on the potential fundraise, though it estimated a lower valuation of around $750 billion. If completed at the higher figure, the deal would rank among the largest private capital raises ever and significantly expand OpenAI’s balance sheet.

The talks come as OpenAI commits to massive spending on infrastructure and global partnerships to maintain its lead in artificial intelligence. The company has been investing heavily in model training and inference, with costs increasingly covered by cash rather than cloud credits. This shift suggests that compute expenses have outgrown what existing partnerships can fully offset.

OpenAI is also moving faster to release new models and expand its developer tools as competition intensifies from rivals such as Anthropic and Google. These efforts require sustained capital to support research, product launches, and the scaling of enterprise services tied to ChatGPT and related APIs.

Market Conditions and Strategic Options

The potential fundraise is unfolding against a more cautious investment backdrop. Broader sentiment around AI has cooled as investors question whether debt fueled spending by major technology companies can be sustained. Amazon, Microsoft, Oracle, and OpenAI have all committed tens of billions of dollars to AI infrastructure, raising concerns about long-term returns.

Supply constraints add another challenge. Shortages in advanced memory chips are limiting the pace of data center expansion and could affect the broader technology sector. These bottlenecks increase costs and complicate deployment timelines for AI developers that rely on high performance hardware.

OpenAI has also been rumored to be exploring an IPO as a longer-term path to raising tens of billions of dollars. The company is reported to be generating annualized revenue of about $20 billion, driven by subscriptions, enterprise contracts, and API usage. An IPO could provide liquidity and a more permanent source of capital, though no formal plans have been confirmed.

Separately, industry speculation has pointed to discussions with Amazon over a potential $10 billion investment. Such a deal could include access to Amazon’s in-house AI computing chips, offering OpenAI another option to manage rising infrastructure costs.

According to PitchBook data, OpenAI currently holds more than $64 billion in capital and was most recently valued at roughly $500 billion in a secondary transaction. A successful $100 billion raise would further strengthen its position but also underscore how capital intensive the AI race has become.

OpenAI has launched a new app directory inside ChatGPT, enabling users to connect apps directly to their AI conversations. According to OpenAI, apps extend ChatGPT interactions by providing new context and letting users take actions such as ordering groceries, creating slide decks, or searching for apartments. Connector apps like Google Drive are now simply referred to as “apps.”

The directory is available across iOS, Android, and web platforms, divided into Feature, Lifestyle, and Productivitycategories. Popular apps like Booking.com, Spotify, and Dropbox are included. To use an app, users click “Connect,” authorize access, and can then start a chat related to the app. For example, Dropbox users can “gather insights, prepare briefs, and summarize reports or internal documents.” Connected apps can also be accessed via @ mentions in conversations.

The update includes an Apple Music app, which allows users to find music, create playlists, and manage libraries through chat, similar to Spotify. DoorDash integration lets users convert recipes, meal plans, and pantry staples into actionable shopping carts directly from ChatGPT.

Developer Access and Monetization

OpenAI is also opening submissions for developers to publish apps in ChatGPT. Developers can follow the company’s app submission guidelines and leverage resources such as open-source example apps, a UI library for chat-native interfaces, a quickstart guide, and the SDK introduced in October.

Currently, developers can monetize apps by linking users to native apps or websites, though OpenAI is exploring internal monetization options in the future. Privacy remains a priority, with developers required to provide clear policies regarding data usage.

The new app directory is part of OpenAI CEO Sam Altman’s vision to make ChatGPT more versatile with custom “GPT” bots. Alongside apps, OpenAI has recently enhanced ChatGPT with faster and more precise image generation through its flagship ChatGPT Images model, which introduces a dedicated workspace and the lower-cost GPT Image 1.5 API. The company has also rolled out upgrades to the underlying GPT models, improving reasoning, multimodal understanding, and instruction-following capabilities, all aimed at making ChatGPT a more powerful and seamless tool for users across creative, productivity, and everyday tasks.

Visa said Thursday that it has successfully completed hundreds of artificial intelligence-powered transactions as part of a pilot program launched after its product event in April. The initiative focuses on enabling AI agents to complete payments and transactions on behalf of consumers, a capability that could reshape how people shop and interact with digital commerce platforms.

The pilot places Visa among a growing group of payments and technology companies experimenting with agentic AI, where software agents are authorized to act for users in specific purchasing scenarios. Visa executives said the tests demonstrated that AI agents can reliably execute transactions within defined parameters, such as recurring purchases or time-sensitive events.

“This is going to be the year we see an enormous amount of material adoption, and consumers really starting to get comfortable in a bunch of different agentic environments,” said Rubail Birwadker, Visa’s head of growth products and partnerships.

Visa did not disclose the transaction volumes or merchants involved in the pilot, but said the program validated the technical and operational feasibility of AI-driven payments within its network.

Fintech Industry Races Toward Agentic Commerce

Visa’s work reflects a broader push across the payments and e-commerce ecosystem to integrate AI agents into consumer transactions. Mastercard said in April that it was testing a feature called Agent Pay, which allows AI agents to shop online for customers. Amazon also began testing a “Buy For Me” service that month, designed to let AI assistants complete purchases on external websites.

PayPal and AI search company Perplexity have also partnered on agent-based shopping tools, highlighting growing interest in automating parts of the online purchasing process. These systems aim to reduce friction by allowing users to delegate routine or repetitive buying decisions to AI, while retaining human oversight for approvals and spending limits.

Visa’s own research suggests consumer adoption is already underway. A survey released earlier this month found that nearly half of U.S. shoppers have used AI in connection with purchases, whether for product discovery, price comparison, or decision support. While most current use cases stop short of fully autonomous transactions, companies see payment authorization as the next step.

Birwadker said AI agents may be particularly useful for predictable purchases, such as household essentials, subscriptions, or booking concert tickets where speed and availability matter. By embedding payment credentials and rules into AI agents, companies hope to streamline checkout without sacrificing security.

Global Expansion Plans and Partnerships

Visa said it plans to expand its AI payments pilots to Asia and Europe next year, signaling confidence in the technology’s readiness for broader testing. The company is currently working with more than 20 partners on AI agent tools, spanning merchants, technology providers, and developers building agent-based commerce experiences.

Security and trust remain central concerns. Payments networks must ensure AI agents operate within strict controls, including spending caps, authentication requirements, and clear user consent. Visa did not detail its safeguards but said its pilots were designed to operate within existing network protections.

As AI agents move closer to handling real-world transactions, payments companies are positioning themselves as infrastructure providers for this emerging layer of digital commerce. Visa’s pilot suggests that agentic payments are shifting from concept to early execution, setting the stage for wider adoption as consumers and merchants become more comfortable with AI acting on their behalf.

Google expanded its Gemini 3 model family with the release of Gemini 3 Flash, a new model built to deliver high-end reasoning performance with faster response times and lower costs. The launch follows last month’s introduction of Gemini 3 Pro and Gemini 3 Deep Think, which together process more than 1 trillion tokens per day through Google’s API.

Gemini 3 Flash combines the reasoning foundation of Gemini 3 Pro with the low-latency characteristics of the Flash series. Google said the model is optimized for speed, efficiency, and scale, while retaining strong performance across reasoning, multimodal understanding, and agentic workflows.

The model is rolling out globally starting today. Developers can access Gemini 3 Flash through the Gemini API in Google AI Studio, Gemini CLI, and Google Antigravity, while enterprises can deploy it via Vertex AI and Gemini Enterprise. Consumers will encounter Gemini 3 Flash as the default model in the Gemini app and in AI Mode in Search.

Performance, Cost, and Developer Use Cases

Google positioned Gemini 3 Flash as a frontier model that balances intelligence with efficiency. The model achieved 90.4% on the GPQA Diamond benchmark and 33.7% on Humanity’s Last Exam without tools, results that rival larger frontier systems. It also scored 81.2% on MMMU Pro, comparable to Gemini 3 Pro, and outperformed Gemini 2.5 Pro across multiple evaluations.

Gemini 3 Flash was designed to adapt its reasoning depth based on task complexity. Google said it uses about 30% fewer tokens on average than Gemini 2.5 Pro on typical workloads, helping reduce inference costs. Pricing is set at $0.50 per million input tokens and $3 per million output tokens, with audio inputs priced at $1 per million tokens.

For software development, Gemini 3 Flash demonstrated strong agentic coding performance. On SWE-bench Verified, the model achieved a 78% score, outperforming both the Gemini 2.5 series and Gemini 3 Pro. Google said this makes it well suited for high-frequency workflows, production systems, and interactive applications that require fast feedback and reliable reasoning.

Companies including JetBrains, Bridgewater Associates, and Figma are already using Gemini 3 Flash, citing its combination of inference speed and reasoning quality. The model also supports complex video analysis, data extraction, and visual question answering, enabling use cases such as in-game assistants and rapid experimentation.

Consumer Rollout Across Search and Gemini

For consumers, Gemini 3 Flash is now the default model in the Gemini app, replacing Gemini 2.5 Flash. Google said this gives users worldwide free access to Gemini 3-level reasoning for everyday tasks. The model’s multimodal capabilities allow it to analyze images, video, and audio, and generate structured outputs such as plans or summaries within seconds.

Gemini 3 Flash is also rolling out as the default model for AI Mode in Search. Google said the integration improves the system’s ability to parse complex queries, combine real-time information, and present results in a more organized and actionable format.

With Gemini 3 Flash, Google aims to push advanced AI reasoning into mainstream use, emphasizing speed, cost efficiency, and broad availability alongside its higher-end Gemini 3 models.

Amazon is in early discussions to invest as much as $10 billion in OpenAI, a move that could deepen ties between the cloud giant and one of the most influential AI labs, CNBC reported. The deal would reportedly involve OpenAI using Amazon’s in-house AI chips, reinforcing Amazon Web Services’ role in large-scale AI training and deployment.

If completed, the investment would value OpenAI at more than $500 billion, Bloomberg reported, citing an anonymous source. Such a valuation would place OpenAI among the most valuable private companies globally and underscore investor confidence in generative AI platforms as demand accelerates across industries.

The talks come as Amazon seeks to diversify its position in the AI market. The company has already committed up to $8 billion to Anthropic, a direct OpenAI rival, and has integrated Anthropic’s models into AWS offerings. Earlier this month, Amazon also unveiled the latest generation of its Trainium AI chips and outlined plans for future iterations, positioning them as cost-effective alternatives to Nvidia hardware for large-scale model training.

Circular Deals Reshape AI Infrastructure

An Amazon investment in OpenAI would mark the latest example of so-called circular deals in the AI sector. In these arrangements, cloud providers or chipmakers invest in AI startups, which in turn commit to using the investors’ infrastructure, chips, or data centers. The structure aligns incentives while locking in long-term demand for compute resources.

OpenAI has already engaged in several such deals. In March, it invested $350 million in CoreWeave, which used the capital to purchase Nvidia chips that now provide compute capacity back to OpenAI. In October, OpenAI took a 10% stake in AMD while agreeing to use the chipmaker’s AI GPUs, and it also signed a chip usage agreement with Broadcom the same month. In November, OpenAI finalized a $38 billion cloud computing deal with Amazon, further cementing their commercial relationship. In December it OpenAI took an ownership stake in Thrive Holdings.

These arrangements reflect the rising cost and strategic importance of AI infrastructure. Training frontier models increasingly requires massive capital outlays, making partnerships with cloud and hardware providers critical to sustaining development.

Strategic Context for OpenAI and Amazon

The reported talks follow OpenAI’s recent transition to a for-profit structure, a shift that allows it greater flexibility in raising capital beyond Microsoft, which remains a major shareholder with a reported 27% stake. That change has opened the door for broader strategic partnerships as OpenAI scales its compute needs and global footprint.

For Amazon, investing in OpenAI could complement its existing AI strategy by expanding adoption of AWS services and proprietary chips. It would also hedge against overreliance on any single AI partner while strengthening Amazon’s position as a core infrastructure provider in the AI ecosystem.

Neither Amazon nor OpenAI commented on the reported discussions. If finalized, the deal would underscore how capital, compute, and AI development are becoming increasingly intertwined as competition intensifies across the generative AI market.