25 Predictions for 2025 (Update)

At the beginning of the year, we put together a collection of theses that could potentially materialize during 2025. “25 Predictions for 2025”, we called it. And as we’re about to cross the one-third mark for the year, an update examining their respective news flow is now befitting.

In this research piece, we’ve summarized hundreds of news items and assessed whether our theses are strengthening or weakening. Safe to say, the current warp-speed technological progress in society overall is no doubt mirrored. AI, general-purpose robots, autonomous vehicles, protectionism, AI-generated music, and simulation of the future, etc. – are all taking substantial strides.

---

Event on May 5th

On May 5th (15.30-17.30) in Stockholm, we are very proud to be hosting an event together with Epicenter, on how AI will transform society.

We'll have an intriguing panel of AI experts, including Peter Kurzwelly (from AI Sweden), Ludvig Strand (Axel Johnson), and our own CEO Daniel Isaacs.

🚀 If you wish to join, e-mail christopher@sircular.io, and he'll send you an invite, or send him a DM.

---

Update:

25 Predictions for 2025

In just four months, many of our prediction from the start of the year (not all of course) have moved from theses, to dawning realities. Autonomous vehicles have surged, with Waymo surpassing 250,000 weekly paid rides and expanding into more cities, while Zoox, Nuro, Tesla, etc., are taking big strides. General-purpose humanoid robots from Figure began full-time work at BMW's South Carolina plant, marking a critical transition from prototypes to practical tools.

Furthermore, AI-generated music saw a significant leap with Amazon integrating Suno into its Alexa+ platform, despite intense legal battles over copyright. Meanwhile, OpenAI stirred the market by entering talks to acquire AI-coding startup Windsurf, a deal potentially valued at billions, underscoring the explosive growth of AI tools designed to empower solo entrepreneurs.

Finally, protectionism escalated dramatically with US imposing tariffs, the world’s first fully AI-made movie was made, simulations of the future is materializing led by SandboxAQ and their Large Quantitative Models (LQMs), and companies like Clarivate and Google intensified efforts to revolutionize the field of research.

Download this research update in PDF or keep reading

Enjoy,

Christopher Lyrhem

Chief Future Officer at Sircular

---

Prediction #1)

Autonomous Mobility Starts Proliferating

Update: Strong Signals for Autonomy (especially for Waymo)

The past months have seen substantial progress across multiple key players in the autonomous vehicle sector. Waymo, Tesla, Wayve, Nuro, and Zoox all featured significant news, including expansion into new domestic and international cities, major software releases (Tesla FSD 13.2), crucial OEM partnerships (Wayve/Nissan), capital raising (Nuro), and increasing ride volumes (Waymo).

While Waymo's progress toward the 500k weekly trip thesis requires further acceleration, the overall trend of proliferation, market activity, and technological advancement is strong and accelerating.

News flow so far in 2025:

In late April, Waymo announced it had surpassed 250,000 weekly paid rides across its service areas. While this represents significant growth (reportedly 25x in roughly two years), it indicates that achieving the initially predicted 500,000 weekly trips by year-end remains an ambitious target (though obtainable).

Waymo's integration with Google Maps continues to expand across its service areas, enhancing user accessibility. Furthermore, the company has initiated testing operations in Tokyo, Japan, marking international expansion efforts.

The upcoming launch of Tesla’s unsupervised FSD in June will be intriguing

Tesla's progress with Full Self-Driving (FSD), particularly the release of version 13.2 in the March/April timeframe, garnered substantial attention. Alphabet CEO Sundar Pichai acknowledged Tesla as a leader in AV development alongside Waymo. The anticipation surrounding Tesla's planned launch of FSD Unsupervised in June 2025 introduces a potentially significant competitive challenge to Waymo's established model.

Wayve, a UK-based startup, secured its first Original Equipment Manufacturer (OEM) deal with Nissan and expanded its on-road testing operations into Germany. Wayve's emphasis on an AI-centric, mapless approach, designed to be "vehicle-agnostic," is gaining considerable traction and attracting investment.

Nuro raised USD 106 million in capital and expanded its testing operations into Japan, its first international data collection initiative. This move, combined with Nuro's increasing focus on an autonomous driving licensing model, strongly suggests an OEM partnership announcement may be forthcoming, potentially with a Japanese manufacturer.

Zoox is now out in the wild, testing trips in designated areas

Conversely, Amazon-backed Zoox faces persistent concerns regarding its timeline and ability to launch and scale a commercial robotaxi service effectively, leading to a neutral outlook from some analysts. Despite plans for employee-only services in San Francisco and Las Vegas, with public access targeted for 2026, operational clarity remains limited.

However, in April 2025, Zoox commenced autonomous testing operations (with safety drivers) in a designated area of Los Angeles, initiating mapping missions before starting full autonomous driving trials later in the summer. Also, Zoox maintains its goal of welcoming public riders in Las Vegas and San Francisco later in 2025.

#2)

One-Person AI Companies

Update: The Big Guns are Arriving

The first third of 2025 as seen intense activity in the AI coding assistant space. Multiple competing platforms (Cursor, Replit, Windsurf, Lovable, Firebase Studio) launched, received updates, or gained significant user traction. Major technology companies like Google made significant entries (Firebase Studio), and OpenAI pursued a major acquisition (Windsurf), indicating the high strategic value placed on this space.

Tools explicitly designed to empower non-technical users are flourishing, directly enabling the core concept of the prediction. Although USD 100m solo ventures are not confirmed, the enabling technology is proliferating rapidly and demonstrating value creation.

News flow so far in 2025:

Windsurf (formerly Codeium) differentiates itself with a cleaner, more beginner-friendly interface and its "Cascade" AI agent, designed for more collaborative and autonomous multi-file editing and project generation. The strategic significance of this space was underscored by reports in April 2025 that OpenAI was in talks to acquire Windsurf for approximately USD 3 billion, which would be OpenAI's largest acquisition to date.

This potential acquisition, despite Windsurf reportedly having lower Annual Recurring Revenue (ARR) than competitor Cursor, suggests OpenAI places high value on Windsurf's agentic technology, enterprise traction, user data, or talent pool for its long-term strategy of embedding capable AI agents into developer workflows.

Lovable is developing in warp speed

Lovable specifically targets non-technical entrepreneurs and beginners, enabling the creation of full-stack web applications directly from natural language prompts. Its success in empowering individuals is highlighted by a user report of raising USD 450,000 in a single day from a website built using the platform.

Replit provides a comprehensive, browser-based Integrated Development Environment (IDE) that handles both development and deployment, attracting users seeking an end-to-end solution. Its Replit Agent feature further aims to democratize full-stack development.

Google is now throwing its body into the field

Google entered the fray decisively in 2025 with Firebase Studio, an AI-powered, browser-based IDE deeply integrated with the Firebase ecosystem and leveraging its Gemini models. It offers templates and an "App Prototyping agent," positioning itself as a powerful, all-in-one competitor.

While the specific benchmark of USD 100 million valuations for one-person AI companies is not directly confirmed in the March/April snippets, the potential for high value in AI-driven creation tools is evident. For instance, Synthesia, an AI video avatar startup (though not a solo venture), achieved USD 100 million in ARR and secured strategic investment from Adobe.

The rapid proliferation and competition among these diverse AI coding tools signal that the market is actively seeking the most effective paradigm for AI-assisted software development. This fragmentation, catering to different user segments (non-technical users, professional developers, 'vibe coders'), suggests that specialized tools may coexist before potential consolidation around dominant approaches occurs.

The explosion of powerful, user-friendly AI coding tools is significantly lowering the barrier to software creation, making the "idea to app in seconds" concept increasingly tangible.

This democratization strongly supports the prediction of growth in the gig economy, empowering individuals to become software creators and entrepreneurs.

#3)

Rise of AI Music Artists

Update: AI Generated Music is a Novel Melody

The technology enabling high-quality AI music generation (Suno V4, Google Vertex AI) advanced significantly, and mainstream integration took a major step forward with the Amazon/Suno partnership. This supports the "rise" aspect. However, the specific prediction of a Billboard 100 hit was not confirmed in thus far in the 2025 news flow.

Furthermore, the intensifying legal battles and ethical debates introduce significant friction and uncertainty regarding the smooth integration of AI-generated music into established industry structures like official charts.

News flow so far in 2025:

The Recording Industry Association of America (RIAA), representing major labels like Universal Music Group (UMG), Sony Music, and Warner Music, along with the German collection society GEMA, filed lawsuits against Suno (and competitor Udio) alleging widespread copyright infringement for training AI models on protected music without authorization. Suno acknowledged its training data "presumably" included copyrighted works but maintained a "fair use" defense, arguing that learning is not infringement.

After streaming, AI-generated music is the next battle for copyright claims

Amazon's decision to partner with Suno amidst these high-profile lawsuits signals a potentially complex dynamic. It occurred just months after Amazon Music expanded its relationship with UMG, explicitly pledging to collaborate on combating "unlawful AI-generated content".

This suggests a possible strategic divergence, where tech platforms prioritize rapid AI feature deployment, potentially forcing the legal and industry landscape to adapt, rather than waiting for clear legal precedents or fully licensed models. This could accelerate pressure on traditional music labels to adapt their licensing models or settle disputes more quickly.

Google is likely looking to scale its’ immense music database: Youtube

Other major platforms also advanced their AI music features. YouTube introduced new tools for creators, including AI-generated instrumentals and beat-matching capabilities. Google's Vertex AI platform also gained the ability to create music from text prompts. Consequently, the ethical and legal debate surrounding AI music has intensified as of late.

While the technological capability and mainstream accessibility for AI music creation surged, there were no confirmed reports in 2025 snippets of an AI-generated song reaching the Billboard 100 chart.

Though achieving a Billboard 100 hit seems increasingly plausible given the technology's potential for creating viral content, although copyright disputes could complicate official recognition and charting. The potential disruption to traditional music industry revenue streams, copyright frameworks, and the definition of creativity is profound.

#4)

General-Purpose Humanoid Robots

Update: The Machines are Coming (to Work)

The prediction of humanoid robots entering factories in 2025 is strongly materializing. Concrete deployments and contracts within the automotive (BMW, Mercedes-Benz, XPeng, Schaeffler) and logistics (GXO) sectors are already occurring. The establishment of dedicated high-volume manufacturing facilities (Figure, Agility, Tesla plans) confirms the industry's shift towards scalable production. The integration of sophisticated AI, driven by partnerships with NVIDIA and OpenAI, is rapidly enhancing robot capabilities.

This convergence suggests that humanoid robots could soon begin to significantly impact manufacturing and logistics labor dynamics.

The early months of 2025 have shown strong validation for this prediction, with substantial activity across the humanoid robotics landscape, particularly concerning factory and logistics applications.

News flow so far in 2025:

Figure made significant strides, unveiling "BotQ," its dedicated high-volume manufacturing facility designed to produce up to 12,000 humanoid robots annually. This move signals a critical transition from research and development towards scalable industrial production, essential for achieving the cost reductions needed for widespread adoption. Following a successful trial period, Figure 02 robots reportedly began full-time work at BMW's Spartanburg, South Carolina automotive plant in January.

Figure seems to be figuring out how to roll out their product strategy

Figure also plans to scale production further with its next iteration, Figure 3, later in 2025. The company's partnership with OpenAI has also yielded impressive demonstrations of conversational interaction and task execution, highlighting the rapid convergence of advanced AI and robotics.

Boston Dynamics, a long-standing leader in advanced robotics, unveiled a new, fully electric version of its Atlas humanoid, retiring the previous hydraulic model and explicitly designing the new platform for real-world applications. The company deepened its collaboration with NVIDIA, becoming an early adopter of the Isaac GR00T framework and utilizing the Jetson Thor computing platform to enhance Atlas's AI capabilities. Furthermore, a partnership with Toyota Research Institute (TRI) was announced.

Agility Robotics is seeing its Digit humanoid robot deployed in logistics settings. GXO Logistics, a major global supply chain provider, is utilizing Digit robots in its operations. Automotive supplier Schaeffler Technologies has not only invested in Agility Robotics but also stated its intention to purchase Digit robots for deployment across its global network of manufacturing plants.Agility is preparing to manufacture Digit at scale in its dedicated RoboFab facility in Oregon.

Tesla continues to advance its Optimus humanoid robot program. Following a showcase alongside its Robotaxi unveiling, the company reiterated plans for limited production in 2025, scaling up in 2026, with ambitious long-term goals of producing millions of units annually. CEO Elon Musk has suggested Optimus could eventually represent greater value than Tesla's automotive business.

Apptronik, developer of the Apollo humanoid, secured an additional USD 53 million in funding in March 2025 (bringing its Series A total to USD 403 million) to fuel production. Demonstrating automotive sector interest, Mercedes-Benz is actively trialling Apollo robots for lineside logistics tasks, such as parts delivery and component inspection, at its plant in Kecskemét, Hungary.

XPeng's PX5 humanoid robot is reportedly undertaking pilot work within its Guangzhou automotive factory, with mass production slated for 2026. Dobot, known for collaborative robot arms, entered the humanoid race with its Atom robot.

Fourier unveiled its N1 humanoid, notably designed as open-source and compatible with NVIDIA's GR00T foundation model.

Unitree Robotics also showcased its H1 and G1 models. This intense activity and state backing position China as a major competitor, potentially accelerating global development through a robotics "Sputnik moment".

#5)

AI Agents go Enterprise Grade

Update: The Agentic Force is Strengthening

The evidence from surveys, major vendor product launches and partnership announcements (Accenture, Salesforce, Google, Microsoft ), and analyst commentary  overwhelmingly points to significant enterprise investment, development, and initial deployment activity for AI agents so far in 2025.

Specific industry solutions and cited customer examples underscore the move towards practical, enterprise-grade applications. While the layoff aspect is less concretely documented, the core prediction regarding enterprise implementation is strongly supported.

News flow so far in 2025:

The first third of 2025 strongly indicates that enterprise adoption of AI agents is indeed a major strategic focus. Survey data reveals a significant commitment from IT leaders: 87% view investment in AI agents as crucial for maintaining competitiveness, and an overwhelming 96% plan to increase their organization's use of agents within the next year. Notably, half of these organizations aim for widespread, enterprise-level implementation. This rapid embrace is relatively recent for many, with 57% having initiated agent implementation within the last two years.

Accenture is introducing an AI Agent building platform

Major consulting and technology firms are heavily investing in and rolling out enterprise-grade agent platforms. Accenture announced an expansion of its AI Refinery™ platform, introducing an AI agent builder designed for business users (requiring little to no coding) and developing a portfolio of over 50 industry-specific AI agent solutions built on NVIDIA AI Enterprise, with a target of 100 by year-end.

Accenture also expanded its partnership with Google Cloud to scale agentic AI capabilities using Google Agentspace and Gemini models, highlighting client work with Mondelēz International, where content agents are scaling personalized content creation. This focus on industry-specific solutions by major consultancies suggests a move beyond general AI tools towards tailored applications designed to deliver tangible ROI and integrate with complex enterprise systems.

Salesforce is clearly doing all they can not to be left behind, with Agentforce

Salesforce made significant moves with the launch of Agentforce 2dx, enhancing its platform to enable proactive, autonomous AI agents that can operate behind the scenes within workflows, extending beyond reactive chat interfaces. Complementing this, Salesforce introduced AgentExchange, a trusted marketplace where partners (including Google Cloud, Docusign, Box) can offer pre-built agent actions, templates, and full agents, accelerating deployment for customers.

Microsoft is embedding specialized agents, such as "Researcher" and "Analyst," directly into its Microsoft 365 Copilot suite, allowing autonomous operation within familiar Office applications. Anthropic's Claude model offers robust tool-use capabilities, enabling developers to build agents that interact with external APIs and data sources.

The startup explosion seen for AI Agents is massive

Startups like Manus are pushing the boundaries towards fully autonomous task completion. The development is further supported by open-source frameworks like AutoGen, crewAI, and LangChain, which facilitate the creation and orchestration of multi-agent systems. The increasing focus on agent interoperability, exemplified by Google's Agent2Agent (A2A) protocol and frameworks for multi-agent coordination , signals a future where complex business processes may be handled by orchestrated teams of specialized agents rather than single, monolithic systems.

#6)

Personal AI Supercomputers

Update: AI PCs - Soon on Every Desk

The concept of powerful, AI-focused personal computing hardware has rapidly materialized in 2025, strongly supporting this prediction. NVIDIA has explicitly launched products fitting the "personal AI supercomputer" description, while AMD and Intel are driving a broader market shift towards "AI PCs" equipped with dedicated neural processing units (NPUs).

The strong push towards powerful local AI processing capabilities can be seen as a strategic counterpoint to the dominance of cloud-based AI infrastructure. Offering benefits like improved latency, enhanced data privacy and security, offline functionality, and potentially reduced long-term operational costs compared to continuous cloud usage.

News flow so far in 2025:

NVIDIA's GTC conference in March 2025 served as a major validation point for this prediction. The company officially unveiled a new line of NVIDIA DGX™ personal AI supercomputers, powered by its latest NVIDIA Grace Blackwell platform architecture. This launch effectively creates the distinct market segment anticipated by the prediction, positioned between consumer-level GPUs and full-scale data center DGX systems.

This "prosumer/developer" tier specifically targets the needs of individual AI researchers, data scientists, and small development teams.

Two key products were announced:

1. DGX Spark: Dubbed the "world's smallest AI supercomputer," the DGX Spark utilizes the NVIDIA GB10 Grace Blackwell Superchip. This chip integrates a powerful NVIDIA Blackwell GPU (with fifth-generation Tensor Cores and FP4 precision support) with a high-performance NVIDIA Grace CPU via the NVLink™-C2C interconnect.

   
   

2. DGX Station: Positioned as a high-performance desktop supercomputer, the DGX Station is built around the NVIDIA GB300 Grace Blackwell Ultra Desktop Superchip (featuring a Blackwell Ultra GPU and Grace CPU). A standout feature is its massive 784GB of coherent memory space, designed to accelerate large-scale AI training and inference workloads directly on the desktop.

   
   

A crucial architectural element emphasized for both systems is the coherent memory model enabled by the NVLink-C2C connection between the CPU and GPU. This allows the CPU and GPU to access a shared memory pool with significantly higher bandwidth (claimed 5x faster than PCIe Gen 5).

Computer makers are lining up to cater this novel product category, together with NVIDIA

Leading computer manufacturers, including ASUS, BOXX, Dell Technologies, HP, Lambda, Lenovo, and Supermicro, have partnered with NVIDIA to manufacture and bring systems based on the DGX Spark and DGX Station platforms to market later in 2025.

While NVIDIA's announcements dominate this specific niche, the broader trend of needing powerful local hardware for AI tasks is also reflected in the growing ecosystem of tools designed to run large language models (LLMs) locally on existing hardware, such as Llama.cpp, LM Studio, and Ollama, though these typically target high-end consumer or prosumer GPUs rather than dedicated desktop supercomputers.  

#7)

Large Quantitative Models (LQMs) Arrive

Update: LQMs; The Next Big Thing in AI?

The substantial funding, high valuation, and strategic partnerships secured by SandboxAQ strongly suggest significant industry and investor belief in the potential of LQMs. By focusing on scientific principles and deterministic outcomes, LQMs appear poised to carve out a distinct and potentially high-value niche within the broader AI landscape, particularly in R&D-heavy sectors like pharmaceuticals, materials science, and finance.

The demonstrated success in complex simulations validates their capability. If this momentum continues, LQMs could indeed drive transformative changes in how research, discovery, and complex quantitative modeling are conducted, fulfilling the prediction's core premise.

News flow so far in 2025:

The early part of 2025 saw significant developments solidifying the emergence of LQMs, largely driven by the activities of SandboxAQ. This company, which spun out of Alphabet Inc., is prominently positioning itself as a leader in the development and application of LQMs.

SandboxAQ is the clear pioneer in the LQM space

SandboxAQ established strategic partnerships: selecting Google Cloud as its preferred cloud provider for LQM development and utilizing the Google Cloud Marketplace for deployment (announced January 2025), and collaborating closely with NVIDIA for high-performance computing (DGX Cloud) and joint research.

In April, SandboxAQ announced a substantial USD 150 million extension to its Series E funding round, bringing the total Series E funding to over USD 450 million and the company's cumulative funding to more than USD 950 million. This round, which included participation from strategic investors like Google and NVIDIA, as well as financial institutions like BNP Paribas and investors like Ray Dalio, valued SandboxAQ at USD 5.75 billion.

Using Google’s Cloud Marketplace is one of SandboxAQ’s distribution channels

Also in April, SandboxAQ unveiled a strategic collaboration with NVIDIA. This partnership involves SandboxAQ leveraging the NVIDIA DGX Cloud AI supercomputing platform, hosted on Google Cloud, to build and scale its LQM platform. The immense computational power offered by DGX Cloud is highlighted as crucial for training and operating these complex, equation-based models, suggesting that, like LLMs, the development of sophisticated LQMs relies heavily on high-performance cloud infrastructure.

SandboxAQ consistently differentiates LQMs from generalized LLMs (like ChatGPT). They emphasize that LQMs are specifically engineered to model the underlying laws governing complex systems, aiming for outputs that are not just predictive but scientifically reliable and deterministic. This focus is deemed critical for high-stakes applications where accuracy and explainability are essential.

High-profile names, but few, seem to be driving the emergence of LQMs

There are strong signals centered around a key proponent, SandboxAQ, including significant funding, high-profile collaborations (NVIDIA, Google Cloud), clear articulation of the LQM concept, and demonstrated applications in complex scientific domains. The technology is being actively developed and promoted. However, the evidence presented in 2025 is heavily concentrated on this single company.

#8)

Protectionism on a New Path

Update: Escalation is an Understatement

The prediction of a significant move towards protectionism has unequivocally materialized. The scale and speed of tariff escalations, the elimination of de minimis benefits for China, and targeted technology restrictions represent a fundamental shift in US trade policy. This is already causing tangible economic disruption and cost increases.

In due time, these actions are likely to strengthen trends of reshoring and nearshoring  as businesses seek to mitigate supply chain risks and tariff impacts. The increased costs associated with imported goods and components could incentivize manufacturers to invest more in domestic automation.

News flow so far in 2025:

The events of March and April 2025 provided stark confirmation of this prediction, particularly concerning US-China trade relations. The US administration implemented a rapid series of escalating tariffs targeting goods imported from China, Hong Kong, and Macau.  

Tariff Escalation: Following a 10% tariff imposed in February and a subsequent increase to 20% effective March 4, the US announced further "reciprocal" tariffs in April. Initially set at 34% (effective April 9), this was swiftly raised to 84% on April 8 following Chinese retaliation announcements, and then dramatically increased again to 125% effective April 10.

Cumulative Impact: These tariffs are additive to pre-existing Section 301 duties, resulting in total tariff rates exceeding 145% for many Chinese products. Certain strategic goods, such as EVs and syringes, face tariffs reaching up to 245%.

Chinese Retaliation: China responded in kind, eventually matching the 125% US tariff rate on American goods by April 12. This tit-for-tat escalation within a matter of weeks signals a highly volatile trade environment, moving beyond calculated policy adjustments towards more reactive measures.

Retaliation has only begun, and will be much broader than “only” tariffs

De Minimis Elimination: A significant policy shift involved the elimination of the de minimis threshold (which allowed duty-free import of shipments valued under USD 800) for goods originating from China and Hong Kong, effective May 2, 2025. Low-value postal items from these regions will instead face a 120% ad valorem duty plus escalating per-item postal fees (USD 100 from May 2, rising to USD 200 from June 1). This change disproportionately affects e-commerce businesses reliant on direct, low-value shipments from China.

Targeted Restrictions: Beyond broad tariffs, specific sectors faced targeted restrictions. A Bureau of Industry and Security (BIS) rule, effective March 17, 2025, prohibited the import and sale of connected vehicle hardware and software (including Vehicle Connectivity Systems and Automated Driving Systems) made in, or incorporating technology from, China or Russia.

Economic Consequences: These protectionist measures are already generating economic repercussions. Retailers were reported to be halting or canceling orders from China due to the uncertainty and cost increases. The automotive industry, heavily reliant on global supply chains, faces particularly severe cost pressures.

Estimates suggest the new tariffs could add nearly USD 5,000 in cost per US-assembled vehicle for the Detroit Big Three (GM, Ford, Stellantis) due to imported parts, and nearly USD 9,000 for each finished vehicle they import, totaling an estimated USD 41.7 billion impact for these companies alone.  

The economic consequences of protectionism are not clear long-term, but clearly negative in the short-term

#9)

Legacy Automotives Hold Major Risks

Update: Pressure on Many Fronts

The prediction of major risks confronting the legacy automotive value chain appears well-founded based on recent developments. Traditional players are caught in a multi-front squeeze. Tariffs inflate costs, EV competition intensifies (particularly from agile Chinese players), and the expensive transition to electric and autonomous technologies must be navigated amidst uncertain consumer demand and economic headwinds.

Suppliers heavily reliant on ICE components and dealers locked into traditional sales models face significant downstream vulnerability. Industry consolidation may become increasingly common as companies seek scale and cost synergies to survive this transformative period.

News flow so far in 2025:

Developments in the first quarter of 2025 lend significant weight to this prediction, illustrating the mounting pressures on legacy automotive companies.

Cost Pressures (Tariffs): The implementation of new US tariffs in March and April 2025 directly impacts the cost structure of legacy OEMs like General Motors, Ford, and Stellantis (the "Big Three"). Due to their reliance on global supply chains, these companies face estimated additional costs of approximately USD 4,911 for each vehicle assembled in the US (due to tariffs on imported parts) and around USD 8,641 for each finished vehicle they import. The total estimated cost increase for the Big Three alone is projected at USD 41.7 billion.

EV Competition and Demand Uncertainty: The competitive landscape in the electric vehicle market is becoming increasingly fierce. While legacy OEMs are introducing new EV models and gaining some market share in the US (reaching a combined 9.3% EV share in March 2025, up from 8.4% year-over-year), the overall growth rate of the US EV market appears more incremental than explosive.

Industry Consolidation: As a potential response to these mounting pressures, signs of industry consolidation are emerging. Talks of a potential merger between Honda and Nissan were noted as reflecting a broader trend where traditional players seek economies of scale and shared R&D costs to better compete with industry leaders and disruptive newcomers.

Consolidation is usually the case for legacy companies, in their stagnating phase

Technological Shifts (Autonomy & Sales Models): The ongoing development of autonomous driving technology (detailed in Prediction #1) represents another significant R&D cost center and potential disruption to traditional business models, although legacy OEMs are actively involved (e.g., Nissan's partnership with Wayve ). The persistent challenge from direct-to-consumer sales models pioneered by companies like Tesla continues to pressure the established dealership network.

Macroeconomic Factors: Underlying economic concerns, exacerbated by the new tariffs potentially dampening consumer spending  and impacting vehicle affordability , add further pressure to the entire automotive value chain.

Everything is converging in a whirlwind of challenges

The confluence of these factors – rising costs from tariffs, uncertain EV demand necessitating difficult investment choices, fierce competition eroding market share (especially in China), and the ongoing expenses of technological transformation – creates a precarious environment for legacy automakers and their associated suppliers and dealers.

#10)

Verification of the Truth

Update: The Truth Needs a new Solution

The need for reliable content verification in the age of AI is undeniable and driving innovation across multiple fronts. While a single dominant solution has yet to emerge, the development of standards like C2PA provides a foundational layer. Cryptographic watermarking and ZKP-based approaches offer theoretical advantages in robustness but face significant technical and practical hurdles before widespread adoption.

The debate spurred by AI-generated content is active, and while cryptographic solutions are part of that discussion, their emergence as a mainstream, deployed countermeasure on large platforms appears to be lagging behind the prediction's timeline.

News flow so far in 2025:

The threat landscape has intensified. Deepfake technology, encompassing both video and audio, achieved remarkable realism by 2025, making manual detection increasingly difficult. Studies confirm that humans struggle to reliably distinguish authentic media from synthetic content. In response, the focus has shifted towards multi-layered defense strategies that combine technical analysis with human oversight. AI-powered detection tools are improving, analyzing subtle artifacts, micro-expressions, vocal patterns, and employing techniques like liveness detection to identify synthetic media.

The response to AI being able to copy anything is intensifying

A major development in verification is the rise of the Content Credentials standard, developed by the Coalition for Content Provenance and Authenticity (C2PA). This standard utilizes cryptographic methods to embed tamper-evident metadata directly into digital media files. This metadata provides verifiable information about the content's origin, creation process, and any modifications, including whether AI tools were used. Unlike easily alterable traditional metadata (like EXIF), C2PA credentials offer a secure way to establish provenance.

Adoption of C2PA has gained significant momentum in 2024 and 2025. Key technology platforms and organizations have either joined the C2PA steering committee or announced plans to integrate Content Credentials into their products and services. This includes major players like LinkedIn (planning to display the 'Cr' icon), Google (exploring integration), and OpenAI (already integrated into DALL-E 3).

Simultaneously, social media platforms are implementing simpler labeling policies. Meta (Facebook, Instagram) announced in March 2025 that it will begin labeling advertisements created or significantly edited using its own generative AI tools, particularly those featuring photorealistic humans. This labeling system is expected to be fully rolled out by the end of 2025, with policies for third-party AI tools to follow.

Google's approach, as per its search guidelines, focuses more on the overall quality and helpfulness of content rather than penalizing AI generation itself, provided the content adds value. These platform labels serve as a basic indicator but lack the cryptographic security and detailed provenance information offered by C2PA.

Blockchain is a technique to reach authenticity – it’s not the only technique

The role of blockchain specifically, as highlighted in the original prediction, appears less direct in the context of mainstream media verification in 2025 compared to the C2PA standard. While C2PA employs cryptography, its implementation doesn't necessarily rely on a public blockchain ledger for media provenance itself.

#11)

AI Research Agents Starts Disrupting

Update: Research is Going Streaming

AI research agents are undeniably emerging and being integrated into both general and specialized platforms. Their ability to rapidly synthesize information, conduct literature searches, and potentially identify connections or gaps holds significant potential to accelerate the research process, aligning with the prediction of disruption.

The "streaming service" thesis could manifest as platforms offer subscription-based access to various specialized research agents or curated research outputs. The initial impact seems geared towards augmenting human researchers, automating time-consuming tasks, rather than fully autonomous research generation.

News flow so far in 2025:

The first third of 2025 has seen a clear and rapid emergence of AI agents specifically designed for research-related tasks, validating the core of this prediction:

OpenAI enhanced its offerings with features like ChatGPT's "Deep Research" mode, capable of in-depth analysis and report generation, and "Operator," an agent designed to interact with live websites for information gathering and task completion.

Google integrated research-focused agents into its platforms. Google Agentspace features pre-built agents like "Deep Research" and "NotebookLM Plus" to automate reporting and data synthesis tasks. Google also introduced an Agent-to-Agent framework.

Microsoft embedded a "Researcher" agent within its Microsoft 365 Copilot suite. This agent leverages OpenAI models alongside Copilot's orchestration capabilities and access to work data and external sources (including third-party databases) to perform complex research tasks directly within Office applications.

AI Agents will collaborate, ensure quality, and find any type of insight – Clarivate is progressing well with big plans for the future

Clarivate made a significant move in April 2025 by expanding its Academic AI Platform to include purpose-built AI Agents for research workflows. Key launches include a "Literature review agent" (available from April 10) designed to optimize queries, analyze data from sources like the Web of Science, summarize insights, and identify research themes, gaps, and hot topics.

Additionally, "Research Intelligence agents" (rolling out by August 2025) aim to connect researchers with collaborators. Clarivate is also developing an "Agent Builder," allowing customers to create their own customized research agents.

Academic publishers are also increasingly adopting AI. Elsevier launched ScienceDirect AI, a tool that extracts key findings from millions of articles and generates summaries to help researchers manage information overload. Wiley published guidelines for responsible AI use in research and writing, acknowledging AI's role while emphasizing integrity. Springer Nature implemented an in-house AI program to assist editors with quality checks.

Furthermore, tools like Perplexity AI are gaining popularity as research assistants. They function as AI-powered search engines that provide summarized answers to queries, complete with citations from web sources, offering a faster alternative to traditional search methods for initial exploration.

Research will “go streaming” and the signals are already here, but way-of-working will change first

While the prediction's specific analogy of research transforming into a "streaming service" is not explicitly confirmed, the observed trends point towards a related evolution. The disruption in 2025 is primarily centered on how research is conducted (process efficiency) rather than fundamentally altering what constitutes original research discovery.

#12)

General-Purpose Household Robots

Update: Scaling up Production (in 2026)

There is strong, direct evidence of progress from robot manufacturers beginning to test general-purpose humanoid robots for household tasks. Supporting trends, such as increasingly sophisticated single-task robots (e.g., advanced vacuums) and rapid advancements in underlying AI and humanoid technology in industrial settings, are also clearly visible.

However, the prediction of an "explosion" in availability and adoption across multiple players, signifying a true "ChatGPT moment" for this category in 2025, is not yet supported by the recent news flow. The signals indicate significant development and early deployment steps, but not yet widespread market impact.

News flow so far in 2025:

While a full-scale "explosion" of widely available, truly general-purpose humanoid robots in homes has not yet materialized in the first quarter of 2025, significant progress is being made, particularly by specialized companies, and the underlying technologies are advancing rapidly.

1X Technologies: This company is explicitly focused on the home market. They unveiled NEO Gamma, the second major iteration of their humanoid robot designed for household tasks, in March 2025. Their ambition is to scale production significantly in 2026 and beyond, aiming for 100,000 units by 2027. The acquisition of robotics startup Kind Humanoid in January 2025 aims to accelerate this development.

Figure to begin testing their robots in home environments in 2025

Figure: Although their initial high-profile deployment is with BMW in an industrial setting , Figure AI has stated plans to begin testing its humanoid robots in home environments starting in 2025. They intend to repurpose their core technology for domestic use and are targeting a price point of $20,000-$30,000 at scale.

Tesla: The Optimus robot is conceived as a general-purpose humanoid, implying potential future household applications. However, the primary focus for 2025 deployment appears to be within Tesla's own manufacturing operations. The target price is also in the $20k-$30k range.

Advancements in Consumer Robotics (CES 2025 Context): The broader consumer electronics space, particularly showcased at CES in January 2025, demonstrated increasing AI integration in home devices. This includes smarter appliances capable of interaction (LG's "Affectionate Intelligence") and significantly more capable single-task robots.

Notably, new robot vacuums like Roborock's Saros series feature AI-powered vision systems and robotic arms, enabling them to not only vacuum but also identify and pick up objects like socks or slippers, moving closer to multi-functionality.

There is continued intense activity for general-purpose robots in China

Activity in China: While specific household robot deployments were less detailed in the snippets, the intense activity in China's humanoid robotics sector (see Prediction #4), including smaller form-factor robots like Unitree's G1, suggests potential for future entries into the consumer market. Yango Group and ROOTS also unveiled autonomous robots in Dubai, though details on their intended application were scarce.

#13)

Simulation of the Future Emerges

Update: Simulation will be Massively Disruptive

The necessary technological components for simulating complex societal futures are rapidly developing. Early applications in marketing simulation (predicting consumer choices) and crisis gaming (modelling geopolitical responses) represent nascent forms of this capability. As these technologies mature and converge, the feasibility of creating more comprehensive simulations of social, economic, or political futures will undoubtedly increase.

Startups dedicated to this specific application will do doubt emerge and gain prominence, but recent data suggests this might occur later than by year-end 2025. The discussion on simulation seems (currently) to be confined to more specialized circles rather than being a "hot topic" in the broader tech discourse.

News flow so far in 2025:

Large Quantitative Models (LQMs): As detailed under Prediction #7, SandboxAQ is actively developing and deploying LQMs. These models, grounded in physics and chemistry, excel at simulating complex real-world systems with high fidelity, particularly in areas like molecular behavior for drug discovery and materials science.

LQMs could be crucial for enabling simulation of the future

AI Agents: The proliferation of enterprise-grade AI agents (Prediction #5) capable of autonomous action and complex task execution provides the necessary "actors" or components for sophisticated simulations. Frameworks for coordinating multiple agents (multi-agent systems), such as crewAI and Autogen mentioned in relation to 2025 developments, are crucial for modelling interactions within a simulated environment.

Salesforce leaders envision using multi-agent teams to simulate business scenarios like product launches or marketing campaigns.

Plenty of domain specific simulation platforms emerging

Simulation Platforms & Digital Twins: Established simulation software providers (e.g., SimScale for cloud-based CAE ), specialized platforms (e.g., Applied Intuition for AV simulation, NVIDIA Isaac Lab for robotics simulation ), and high-performance computing platforms (e.g., Rescale ) provide the necessary infrastructure.

The concept of Digital Twins – virtual replicas of physical systems or processes that run in parallel – is also gaining traction, with applications in manufacturing, smart cities, healthcare, and dedicated academic conferences emerging (like the IEEE Digital Twin 2025 conference ).

Science and R&D: Simulating molecular interactions for drug discovery and predicting material properties (SandboxAQ's LQMs).

Engineering and Product Development: Simulating CAD models for design validation (SimScale), testing and training autonomous vehicles (Applied Intuition), and optimizing robot behavior in virtual environments (NVIDIA Isaac Lab). Digital twins are used for optimizing manufacturing processes.

Financial simulation requires (probably) the potential of LQMs..?

Business and Finance: Simulating the potential impact of marketing campaigns or product launches (Salesforce vision), potentially modeling financial market behavior (SigTech's multi-agent system MAGIC), and optimizing supply chains.

Crisis Simulation: AI is being employed in strategic exercises to simulate responses to large-scale crises, such as hypothetical bioattacks or cyberattacks involving multiple actors (governments, corporations). Georgia Tech conducted such a simulation in April 2025.

#14)

Personal Avatars Flood Social Media

Update: Cloning Ourselves to Scale Ourselves

The technological foundation for the predicted flood of personal avatars on social media is firmly in place and rapidly improving. Platforms like HeyGen make creating realistic personal avatars attainable. The explicit targeting of social media content creation as a primary use case, combined with the potential for monetization (either directly by individuals or through AI UGC for brands), strongly suggests that the volume of avatar-based video content is likely to increase significantly.

As the technology becomes more widespread and the quality continues to improve, personal avatars could become a mainstream feature of online interaction and content creation, potentially altering the nature of social media feeds and online identity expression.

News flow so far in 2025:

AI avatar generation platforms have matured significantly, offering users the ability to create realistic digital personas with relative ease:

HeyGen: Positioned as an advanced AI avatar generator, HeyGen allows users to create custom avatars based on uploaded photos/videos or text prompts, or choose from stock AI actors. It emphasizes customization, offering unlimited looks, responsive conversations, emotional expression, and various stances. HeyGen supports voice cloning and multiple languages.

Synthesia: Another leading platform, Synthesia enables users to create professional videos using AI avatars and voiceovers simply by typing text. It offers over 230 stock AI avatars and the option to create custom ones, supporting over 140 languages. Synthesia is widely used for corporate training, HR communications, and marketing.

Other Platforms: Numerous other tools like Elai (focused on corporate learning ), D-ID (focused on outreach with performance tracking ), VEED (general personal/professional use with editing tools ), Colossyan (compliance training focus), Gan.AI, Affogato (creation of image and video clones), and others cater to different niches within the AI avatar market.  

Currently, these avatars are beginning to be seen not just as tools but as extensions of personal identity in the digital realm.

There is an explosion of startups aiming to solve the most realistic clones

Social Media & Personal Branding: AI avatars offer a customizable way to represent oneself online, from professional networks like LinkedIn to creative platforms like Instagram and TikTok. Influencers and creators use them to maintain a consistent presence without always being on camera. The trend includes hyper-realistic avatars, stylized digital art avatars, and even avatars featuring pets.

Marketing & Sales: Businesses use AI avatars as digital brand ambassadors for personalized marketing videos, advertisements, social media content, and sales pitches. This allows for consistent messaging and reduces production costs compared to traditional video shoots.

Corporate Communications & Training: Companies utilize avatars for internal communications, onboarding, HR announcements, and creating scalable, multilingual training materials.

Clones will decimate human customer service

Customer Service: While perhaps less "personal" in the sense of representing a specific individual, AI avatars are used to create more engaging virtual assistants and chatbots compared to text-only interfaces.

#15)

First AI Movie Released

Update: “Love You“ – The First Ever AI Movie

The prediction of an explosion in short-form AI movies seems to be materializing, fuelled by increasingly accessible and powerful tools like Veo 2, Sora, and others demonstrated in online creations. This space is exploding with more startups and better quality.

A product of this is the Indian feature film “Love You”, which is said to be the world’s first fully AI-made film (95 minutes long) – crafted without a single human actor (the trailer looks like a traditional “firsts” for a new technology, but just wait..).

AI is no doubt becoming a powerful tool within traditional filmmaking. The technology for short clips is advancing rapidly, but scaling to feature films appears to still be a significant challenge.

News flow so far in 2025:

OpenAI's Sora model, made available via ChatGPT Plus in 2025, allows users to generate video clips (initially up to 20 seconds at 1080p for Pro users) from text prompts or by remixing existing generations. It features capabilities like Storyboard mode for sequential scene generation with visual consistency.

Google launched Veo 2 within Gemini Advanced (globally rolled out in 2025), capable of generating 8-second, 720p videos from text prompts, emphasizing cinematic realism and understanding of physics/motion. Other platforms like Runway offer sophisticated AI video tools (text-to-video, image-to-video, background removal, slow-motion).

Tools like HeyGen, Synthesia, Pictory, Lumen5, InVideo, Veed, CapCut, and FlexClip offer varying levels of AI video generation and editing capabilities, often integrated with AI avatars and voiceovers, making video creation more accessible to non-professionals.   This technological progress is fueling creative experimentation:

We will soon have thousands of short-films that are fully AI-generated, uploaded on Youtube every day

Independent creators are using tools like Sora to produce short films. For example, the sci-fi short "EDEN" was created with visuals generated entirely by Sora and voices by OpenAI models, based on a human-written script. The annual Runway AI Film Festival (with screenings in NYC and LA in June 2025) specifically celebrates filmmakers using AI techniques.

Other AI film festivals are emerging globally, including events in Amsterdam, Dubai, Austin, and the Reply AI Film Festival in Venice (September 2025), indicating a growing community and interest in AI filmmaking. These festivals showcase short films (typically 1-10 minutes) created or enhanced using AI.

“Love You” – the "world’s first" fully AI-generated long-form movie

Feature Films: Reports surfaced in April 2025 about Kannada writer-director Narasimha Murthy completing a 95-minute feature film titled "Love You," which claims to be the world's first fully AI-generated feature. The project reportedly used around 30 licensed AI tools over six months, with software costs around 10 lakh (appr. USD 12,000).

However, limitations were acknowledged, including inconsistencies in character features noted by censors, and the AI technician involved stated the result would be vastly better if recreated with newer tools due to rapid AI evolution. This is a testament to the progression of the underlying tech.

Separately, the trailer for another Indian AI-generated feature film, "Naisha," dropped in March 2025. Directed by Vivek Anchalia, it features fully AI-generated visuals and lead characters and was slated for a theatrical release in May 2025. Hollywood production company Staircase Studios AI also announced plans to produce multiple AI movies, with "The Woman With Red Hair" under production.  

#16)

Service-as-a-Software: Billion Dollar Market

Update: Flipping the Script on SaaS

The "Service-as-a-Software" model, powered by increasingly autonomous AI agents, appears to be a significant emerging trend. The rapid market growth projections  and the development of dedicated SaaS 2.0 platforms support the prediction's core idea. Enterprises are actively deploying agents to automate workflows and perform tasks previously done by humans or software.

This trend pose a threat to legacy software companies whose value proposition is based on providing tools for humans to perform tasks, vs. agents performing tasks directly. As agents become more capable and integrated (via platforms like n8n, Zapier, or enterprise suites), they could indeed start "eating" traditional software functions, leading to significant market disruption.

News flow so far in 2025:

The distinction between traditional software/GenAI assistants and autonomous AI agents is becoming clearer and more central to industry discussions. Agentic AI is defined by its ability to perform complex tasks, make decisions, and adapt in real-time without constant human intervention.

This autonomy is the core of the "Service-as-a-Software" idea – employing an AI entity to do a service, rather than just providing a tool (software) for a human to use. Evidence of this shift is abundant:

Salesforce “pivoting” its gravitational center toward AI Agents is very telling

Enterprise Agent Adoption: As detailed in Prediction #5, enterprises are rapidly adopting AI agents for tasks like customer service resolution, security monitoring, development assistance, and process automation. Salesforce's pivot to "Agentforce"  and OpenAI's development of agents like A-SWE (software engineer replacement)  exemplify this move towards autonomous task execution. Gartner predicts 15% of daily work decisions will be made by agentic AI by 2028.

Workflow Automation Platforms: Platforms like n8n are specifically designed to enable technical teams (and increasingly, others) to build complex AI-powered workflows and agents. n8n reported raising EUR 55 million in Series B funding (March 2025) after experiencing 5x ARR growth and tripling its user base, demonstrating strong market demand. Other platforms like AutoGen and LangChain provide frameworks for building multi-agent systems.

Specialized AI creating software on command will make “AI Eating Software” a reality

"AI Eating Software": The prediction's notion of AI "eating" software is reflected in discussions about AI agents disrupting the traditional SaaS model. Instead of licensing complex software suites, businesses might increasingly employ specialized AI agents (potentially from different providers) orchestrated to perform specific business functions. This could lead to more modular, flexible, and potentially cost-effective solutions compared to monolithic legacy systems.

Real-World Agent Services: Examples of AI agents performing services are emerging: customer support bots resolving issues autonomously , AI agents performing research, AI agents managing projects or taking meeting notes , AI agents optimizing logistics, and AI agents handling financial tasks like fraud detection or investment analysis. Tools like Atera's Action AI offer IT service capabilities like remote session summarization and troubleshooting.

#17)

AI Virus Starts Digital Pandemic

Update: The Next Pandemic Awaits?

While a full-blown AI virus pandemic has not been reported, the underlying concerns are valid and growing. The development of autonomous malware and AI-enhanced attack tools signifies a clear escalation in the cyber threat landscape. If such malware were to achieve rapid self-propagation capabilities, the potential for a digital pandemic scenario exists.

The current focus on AI-powered threats is already driving increased investment and research into AI-based cybersecurity defenses (e.g., AI threat detection, automated incident response ). A major incident, even if contained, would undoubtedly accelerate these efforts and likely trigger significant regulatory and policy responses globally.

News flow so far in 2025:

AI-Generated Malware: Security researchers and analysts confirm the emergence and observation of AI-written and AI-generated malware "in the wild". Hackers are actively using AI tools to create, modify, and obfuscate malicious code, making it harder to detect. This represents a significant leap in malware development.

AI malware with the ability to evolve is really scary stuff

Adaptive and Autonomous Malware: AI-powered malware exhibits characteristics that traditional malware lacks. It can potentially "learn" to bypass antivirus tools, adapt its behavior based on the environment it encounters, change its code (polymorphic malware) to evade signature-based detection, and even make decisions autonomously. This creates a more persistent threat landscape.

Examples like the Morris II worm, designed to target GenAI infrastructure with self-replicating prompts, demonstrate the potential for AI-specific propagation mechanisms. Egan Ransomware uses GANs to mutate while preserving functionality, specifically targeting AI-powered defences.

AI Enhancing Attack Sophistication: AI is being used across the attack chain. This includes generating highly realistic and personalized phishing emails or messages by analyzing targets' digital footprints, creating convincing deepfake audio or video for social engineering scams, accelerating vulnerability discovery, and automating credential stuffing or brute force attacks. Multimodal AI could automate attacks.

AI tools like WormGPT and FraudGPT have emerged in chat forums, designed specifically for malicious purposes like generating phishing content or malware code, lowering the barrier to entry for less skilled cybercriminals. Discussions around "jailbreaking" public LLMs to bypass safety protocols are also increasing.

There is currently a surge in AI-powered cybersecurity threats

Cybersecurity reports highlight a surge in AI-powered threats. Darktrace's 2025 report found 78% of CISOs see AI threats significantly impacting their organizations (up from 2024). The Imperva 2025 Bad Bot Report noted automated traffic (including malicious bots, often AI-enhanced) surpassed human traffic online for the first time, with bad bots accounting for 37% of all web traffic.

Account Takeover (ATO) attacks, often automated using AI, surged 40% year-over-year. Global tensions and potential trade wars could increase the likelihood of AI-leveraged cyberattacks.

Focus on AI Security & Defence: In response, there's a growing focus on AI-driven cybersecurity defences. Organizations are implementing AI for threat detection, anomaly analysis, and automated response. However, skills gaps in understanding and deploying AI security tools remain a challenge. Formal policies for safe AI use are being discussed or implemented in many organizations.

#18)

AI Doctors aid Human Doctors

Update: Your Next Doctor could be an AI Doctor

The prediction of AI significantly aiding medical professionals in 2025 is strongly supported by the ongoing trends. AI is increasingly integrated into clinical workflows for documentation, diagnostics support, and treatment planning. The heavy VC investment ensures continued innovation and deployment.

While full "AI doctors" might be further off, AI co-pilots and specialized diagnostic/research tools are becoming reality. This could lead to improved diagnostic accuracy, more personalized treatments, accelerated drug development, and potentially alleviate some of the administrative burden on clinicians.

News flow so far in 2025:

AI in Diagnostics and Clinical Support:

• AI applications are expanding in healthcare, ranging from drug discovery to diagnostics and clinical support. LLMs are noted for enhancing clinical support, diagnosis, and treatment through their natural language processing capabilities.

  
  

• AI is being used for predictive analytics, personalized medicine, and clinical decision support, aiming to improve patient outcomes and operational efficiency. Specific examples include AI-powered diagnostics analyzing biomarkers  and AI assisting in genetic testing analysis for precision medicine.

  
  

• AI copilots are being developed to assist clinicians at the point of care, presenting actionable insights from patient health histories by aggregating structured and unstructured data. Abridge raised $250M for its clinical AI scribe and copilot technology.

  
  

• Ambient listening technology, using AI to analyze patient-provider conversations and automate clinical note generation, is seeing increased adoption, reducing documentation burden and allowing clinicians to focus more on patients.

  
  

  
  

AI in Drug Discovery and Development:

• AI (including ML and DL) is revolutionizing drug discovery, target identification, lead optimization, and predicting clinical trial outcomes. AI models analyze vast datasets to identify targets, design molecules, and predict drug behavior and protein structures (e.g., AlphaFold)

  
  

• Generative AI is accelerating the development and re-engineering of molecules. While no fully AI-designed drug was FDA-approved as of Feb 2025, several candidates developed using AI have entered clinical trials with reportedly higher success rates than traditional methods.

  
  

• Healthcare and AI remain leading sectors for venture capital investment in early 2025. Digital health startups raised $3 billion in Q1 2025, with significant funding rounds for AI-focused companies like Innovaccer ($275M), Abridge ($250M), Hippocratic AI ($141M), and Qventus ($105M). This robust funding environment fuels the development and deployment of AI healthcare solutions.

Hospitals and healthcare systems are moving beyond viewing AI as just a cost-cutting tool towards embracing it as a strategic imperative for improving care and outcomes. While caution remains, particularly for patient-facing roles, 2025 is seen as a year of "measured adoption," focusing initially on operational support (data processing, workflow automation) before broader clinical application.  

#19)

Autonomous Software Companies

Update: The Pieces are Coming Together

There is strong evidence for the rapid development of both highly autonomous AI agents and the integration of AI with blockchain protocols, DAOs, and tokenization for financial operations and governance. Projects like AI16z (AI-governed DAO) and protocols enabling AI agent tokenization (Virtuals, ModelFi) are direct steps towards this prediction.

As AI agents gain more sophisticated planning and execution capabilities, and as governance mechanisms become more autonomous, the line between an advanced AI agent system and an "autonomous software company" could blur. This could lead to entirely new organizational structures, and investment models (tokenized ownership of AI agents). Not yet though.

News flow so far in 2025:

Advanced AI Agents: As detailed in Predictions #5 and #16, AI agents are rapidly becoming more autonomous and capable. They can plan, reason, use tools (APIs, databases), and execute complex, multi-step tasks with minimal human oversight.

Platforms like OpenAI's AgentOS, Microsoft's AutoGen, and frameworks like LangChain, crewAI, and LangGraph Pro are enabling the development of sophisticated, often collaborative, agent systems. Startups like Manus are explicitly aiming for agents that complete full tasks end-to-end without user intervention.

Agentic systems are all the rage, developing in warp speed

AI Agents in Crypto/Blockchain: The intersection of AI agents and blockchain/smart protocols is an active area of development, directly relevant to the prediction's mention of "smart protocols" and "autonomous financial entities.“

Projects are emerging where AI agents operate on blockchains, manage wallets, interact with decentralized protocols (DeFi), and execute tasks like trading or vulnerability scanning. Examples mentioned include H4CK (ethical hacking agent), AIXBT (social media/market analysis agent), AI16z (VC investment agent), and Vader.ai (portfolio management agent).

Decentralized Autonomous Organizations (DAOs) are being combined with AI. AI16z is structured as a DAO governed by an AI agent ("Marc AIndreessen"), using community proposals and token holder input for AI-driven investment decisions.

Protocols like Virtuals Protocol allow users to create, tokenize (as ERC-20 assets), and monetize AI agents, enabling co-ownership and potentially revenue sharing. MagnetAI's ModelFi protocol similarly focuses on tokenizing AI models/agents (ERC-7641 standard) for funding, revenue sharing, and community governance. Frameworks like Bittensor create decentralized marketplaces for AI models, using tokens ($TAO) for incentives and governance.

Hive Intelligence launched its mainnet, providing streamlined multi-chain data access via API, specifically designed for AI agents and LLMs to query and interact with blockchain data, overcoming previous fragmentation challenges.

Self-sustaining financial entities are yet to be explicitly demonstrated

Gaps to Full Autonomy: While these developments show AI agents operating financially (trading, investing) and being governed decentrally (DAOs, tokenization), the concept of a fully autonomous software company – an agent that defines its own strategy, manages its own operations, hires resources (human or AI), and exists as a self-sustaining financial entity without human owners in the traditional sense – is not yet explicitly demonstrated.

#20)

First Streaming AI TV & Radio

Update: The “Next Netflix” could be a DYI AI-Platform

The technological building blocks required to create fully automated AI TV or radio channels are rapidly falling into place. Realistic voice and video generation, combined with AI's ability to research and script content, make the concept technically feasible. The emergence of dedicated AI news anchor tools is a direct step in this direction. Challenges remain in ensuring the quality, coherence, and engaging nature of fully automated live content streams over extended periods.

While the prediction of dedicated platforms and widespread channel creation might be slightly ahead of the curve for 2025, the potential for AI to automate significant portions of TV production is undeniable. There might come a “next” Netflix soon…

News flow so far in 2025:

AI Content Generation: Tools for automated research and writing are available (Prediction #11). Voice synthesis and cloning are highly advanced (e.g., HeyGen, ElevenLabs ). AI video generation models like Sora and Veo 2 can create realistic short-form video content , and platforms like Synthesia and HeyGen allow for the creation of videos featuring AI avatars delivering scripts. AI can also generate background music (Prediction #3). Platforms exist for automating social media posting and content repurposing.

AI in Media Production: AI is being integrated into traditional media workflows. At NAB Show 2025, AWS showcased innovations including generative AI for voice applications (Amazon Nova Sonic) and video generation (Amazon Nova Reel 1.1), alongside AI tools for enhancing live sports production (e.g., Warner Bros. Discovery Sports Europe's CCI platform for cycling commentary enhancement).

AI is used for quality assurance, compression, classification, and understanding consumer behavior in streaming. AI-Media partnered with AudioShake (April 2025) to use AI for separating live sports audio (commentary vs. stadium noise) and enabling real-time AI voice translation for global broadcasts, with a first release scheduled for July 2025.

Despite these capabilities, evidence of fully automated, live AI TV or radio channels launching and gaining traction in 2025 is scarce in the news flow

Channel 1 News: While not explicitly mentioned in the 2025 snippets provided here, Channel 1 News, which launched previously promising AI-generated news anchors and personalized news broadcasts, represents an early example of this concept. Its progress or lack thereof in 2025 would be a key indicator, but isn't covered here. General discussions mention the potential for AI newsgathering and virtual anchors.

Platform Emergence: The prediction mentioned new platforms emerging for hosting these channels. While platforms for creating AI content are abundant , platforms specifically designed for streaming numerous live, AI-generated channels as a new form of social media are not highlighted in the 2025 news. Existing platforms (YouTube, Twitch etc.) are integrating AI tools , but not necessarily pivoting to host fully automated AI channels en masse.

New platforms often emerge with novel technologies; this could be the case this time around as well

Focus on Enhancement: The dominant trend appears to be using AI to augment human creators and existing media operations – generating background music , enhancing live commentary with data or translation , creating marketing videos with avatars, or repurposing content. Fully autonomous content generation and live broadcasting seem less developed or adopted.  

#21)

EV Batteries Continue Cost Declines

Update: Cost-Parity with ICE Cars is Nearing

The continued decline in battery costs, particularly the push towards and potentially below the $100/kWh mark, is fundamentally altering the economics of electric vehicles. It makes upfront price parity with ICE vehicles increasingly achievable, likely accelerating mainstream consumer adoption. The dominance of LFP (Lithium Iron Phosphate) chemistry highlights a shift towards more cost-effective and sustainable battery solutions.

Furthermore, the prediction regarding China's EV market share is strongly supported by both forecasts  and actual March 2025 sales data showing high penetration rates.

News flow so far in 2025:

S&P Global Mobility analysis (Jan 2025) also suggested potential upward price adjustments in the coming years as some suppliers are currently operating with minimal profit margins, which is unsustainable, and as regional supply chains outside China struggle to meet demand. Goldman Sachs forecasts for lithium prices suggest a gradual recovery starting in 2025 but remaining well below 2023 peaks, with supply surplus continuing.

Cobalt-free LFP batteries comprise 75% of installations in China

LFP Adoption: The shift towards lower-cost, cobalt-free LFP batteries continues, particularly in China where they account for 74.6% of installations. LFP cell prices fell below $100/kWh on average for the first time in 2023/2024, averaging $95/kWh at cell level and $130/kWh at pack level globally. LFP packs were significantly cheaper than Nickel Cobalt Manganese (NCM) packs. LFP production is expanding outside China, notably in Europe.

Sodium-Ion: Sodium-ion batteries are emerging as a potential low-cost alternative, particularly for grid storage and budget EVs, with companies like CATL investing and the US DOE supporting commercialization.

Solid-State: Considered a next-generation technology promising higher energy density and safety, solid-state batteries are still under development by companies like Toyota, QuantumScape, and Samsung SDI, with commercialization likely still a few years away.

Super intriguing cell-to-pack tech and ultra-fast charging from CATL

Other Advancements: CATL's Qilin battery (up to 255 Wh/kg NMC, 160 Wh/kg LFP) with advanced cell-to-pack tech and ultra-fast charging (6C) is in production. BYD plans to launch Blade Battery 2.0 by 2025 with improved density (up to 210 Wh/kg LFP) and 8C charging. These advancements improve performance and potentially lower pack costs through better integration.

Price Parity: The price gap between EVs and ICE vehicles has been shrinking globally, falling significantly in the US from ~50% in 2021 to ~15% in 2023/2024. The benchmark for EV-ICE price parity is often considered $100/kWh at the pack level. Analysis indicates this threshold has already been crossed for LFP battery packs manufactured in China (average price $94/kWh in 2024.

EVs will likely become a majority of Chinese car sales during 2025

China EV Sales: The prediction of EV sales overtaking ICE sales in China during 2025 appears plausible. In February 2025, plugin vehicles (BEVs + PHEVs/EREVs) achieved a remarkable 50% market share. For the first two months of 2025 (Jan-Feb), NEV (New Energy Vehicle, primarily plugins) shipments surged 52% year-over-year, while ICE vehicle sales declined 3.6%. This strongly suggests plugins are on track to outsell traditional ICE vehicles in China over the full year 2025.  

#22)

New Record in Global Emissions

Update: Politics is not the Solution

The data confirms that atmospheric CO2 concentrations continue their relentless climb to unprecedented levels, keeping the world on a trajectory far exceeding the goals of the Paris Agreement. Direct measurements show CO2 levels continuing to rise through March and April 2025, reaching levels consistent with the prediction. Also, reputable forecasts (Met Office) predict the May 2025 peak will hit the 429 PPM (Parts Per Million) prediction.

The confirmed US withdrawal from the Paris Agreement casts a significant shadow over international climate diplomacy, making the task of achieving global consensus at the COP30 even more challenging.

News flow so far in 2025:

Assessing the trajectory of global emissions and climate indicators for the full year 2025 based on data from the first half is inherently challenging, as comprehensive annual data requires time for collection and analysis. However, available reports and forecasts provide some context. Below are some of the key news items so far in 2025:

Recent Emissions Trends: Global CO₂ emissions from power generation (the largest emitting sector) rose by a modest 1% in 2024, following a 1.4% rise in 2023, reaching approximately 13.8 billion tons. This increase occurred despite growth in clean energy, driven by a 1.3% rise in fossil fuel generation amid strong global electricity demand growth (4.3%). While this indicates emissions were still rising entering 2025, the rate of increase in the power sector was slowing.

IEA forecasts flattish trend for CO2 from the power sector going forward

2025-2027 Power Sector Forecast (IEA): The International Energy Agency (IEA), in its "Electricity 2025" report, forecasts that global CO₂ emissions from the power sector are expected to remain relatively flat (average -0.1% change) over the 2025-2027 period.

This stabilization, despite projected annual electricity demand growth of 3.9%, is attributed to substantial anticipated growth in clean energy sources (renewables and nuclear) offsetting the need for increased fossil fuel generation.

This specific sector forecast suggests a potential peaking or plateauing of power sector emissions, which would be a positive development counter to the prediction of ever-new overall records. However, this only covers one sector, and total global emissions depend on trends in industry, transport, buildings, and land use as well.

Atmospheric CO2 Concentration: The UK Met Office forecast for the annual average CO2 concentration at Mauna Loa, Hawaii, projects it to be 426.6 ± 0.6 ppm for 2025. This represents an increase of 2.26 ± 0.56 ppm compared to 2024. While this confirms the ongoing rise in atmospheric CO2, the forecast value (around 426.6 ppm) is lower than the prediction's specific figure of 429 ppm for year-end 2025.

The warming of the planet has accelerated in recent years, and continues to do so despite cooling La Niña impacts

Global Temperature & La Niña: January 2025 was reported by Copernicus Climate Change Service (C3S) as the warmest January on record globally, with an average surface air temperature 0.79°C above the 1991-2020 average and 1.75°C above the pre-industrial level.

This marked the 18th month out of the last 19 where the global average temperature exceeded 1.5°C above pre-industrial levels in the ERA5 dataset. This occurred despite the emergence of La Niña conditions in the tropical Pacific, which typically have a temporary cooling effect on global temperatures.

#23)

New Manufacturing Tech Stack Forming

Update: The Story of the Century

There’s a convergence taking place. Advanced automation (robots), smarter logistics (AVs), flexible production (AM), and sustainable inputs (renewable materials) are collectively reshaping manufacturing. This tech stack is increasingly viewed not just as efficiency boosters but as strategic enablers for achieving greater supply chain resilience through localization (reshoring/nearshoring) and improved environmental performance through circularity.

Each component technology (robotics, AVs in logistics, AM, sustainable materials) shows significant progress and adoption trends in early 2025. Crucially, multiple sources explicitly identify the trend towards localized supply chains/reshoring as a major theme for 2025.

News flow so far in 2025:

Autonomous Vehicles in Logistics: While consumer AVs grab headlines, autonomous trucks and delivery vehicles are advancing in logistics. Companies like Kodiak, Aurora, and Gatik are testing and deploying autonomous trucks for freight, particularly in the US. Autonomous delivery robots (Nuro, Starship, etc.) are scaling operations for last-mile delivery (See Prediction #1 and 25).

Additive Manufacturing (AM) & Circularity: AM is increasingly recognized for its role in sustainable manufacturing and circularity. Its ability to enable on-demand production reduces waste from overstocking, while facilitating the direct use of recycled materials as feedstock closes material loops. AM also supports repair and remanufacturing by allowing the creation of custom replacement parts. AI is being integrated with AM to optimize processes and ensure the quality of parts made from recycled or renewable feedstocks.

Renewable/Sustainable Materials: There's growing adoption of biodegradable plastics and plant-based packaging materials (derived from sugarcane, cornstarch, etc.) as alternatives to petroleum-based plastics. Advanced chemical and enzymatic recycling technologies are improving the quality and feasibility of reusing plastics.

Reshoring is accentuated by protectionism, but will become a sc aled reality due to a new manufacturing tech stack

Reshoring and Localized Supply Chains: A major trend highlighted for 2025 is the strategic shift towards reshoring and nearshoring manufacturing. Driven by the need for supply chain resilience (exposed by the pandemic and geopolitical tensions) and supportive trade policies, companies are investing in domestic or regional production. Automation and robotics are often seen as key enablers for making localized production economically viable.

Connecting the Dots: Crucially, these trends are increasingly discussed in tandem. Reports explicitly link advancements in AI, automation (including robotics), and digital technologies to enabling more resilient, localized, and sustainable manufacturing paradigms.

The World Economic Forum notes the alignment between sustainability initiatives (reducing energy/material use) and business performance, driven by digital transformation. AI is seen as optimizing not just production but also supply chains and energy use.

#24)

AI Merges with Blockchain Tokens

Update: This is not a Crazy Idea anymore

The prediction of AI merging with blockchain tokens is strongly materializing, particularly through the tokenization of AI agents and models. Platforms like ModelFi and Virtuals are creating direct mechanisms for shared ownership and revenue distribution, mirroring the prediction's core idea. AI agents are actively participating in on-chain financial activities and even governing DAOs.

This convergence is fostering new economic models where value generated by AI can be captured and distributed decentralized via tokens, potentially disrupting traditional software monetization and investment paradigms.

News flow so far in 2025:

MagnetAI's ModelFi Protocol: This protocol allows AI creators to tokenize their models, agents, or bots using the ERC-7641 standard. These tokens represent shared ownership and enable revenue sharing between creators and token holders, who can benefit from the AI's success through subscriptions or other monetization methods facilitated by the decentralized "Model Store" marketplace. This directly addresses the prediction's core concept of token holders sharing in AI-generated revenue.

Virtuals Protocol: This platform enables users to create, tokenize (as ERC-20 assets), and monetize AI agents on the blockchain (specifically Ethereum L2 Base). Token holders gain co-ownership and governance rights, participating in the agent's potential earnings.AI Agents Operating Financially on Blockchain: AI agents are being deployed on-chain to perform financial tasks, often linked to specific tokens:

The $WAI token from Web3 AI is explicitly linked to a platform offering AI tools for crypto trading and security, featuring a revenue-sharing model for token holders.

AI16z is a DAO for VC investments

AI16z: This project operates as a DAO where an AI agent ("Marc AIndreessen") makes venture capital investment decisions, advised by proposals from $AI16Z token holders. This demonstrates AI acting as an autonomous financial entity governed via tokens.

Bittensor: This network functions as a decentralized marketplace for AI models, using its $TAO token to incentivize developers (miners) to contribute and share models. $TAO also serves governance functions.

Hive Intelligence seems to be an interlayer between AI models and blockchain

Hive Intelligence: Launched its mainnet providing an API designed for AI agents and LLMs to easily query and interact with multi-chain blockchain data, facilitating AI operations within the crypto space. Oraichain: Aims to integrate AI models directly into smart contracts.

Artificial Superintelligence (ASI) Alliance: The merger of Fetch.ai (FET), SingularityNET (AGIX), and Ocean Protocol (OCEAN) tokens into the single $ASI token represents a major consolidation effort to build a large-scale decentralized AI ecosystem. This signifies significant investment and belief in the convergence of AI and blockchain.

#25)

First Autonomous Stores Arrive

Update: 7-Elevens Will Soon Come to Us

Autonomous mobile stores, characterized by a "store-hailing" model where customers summon and shop directly from a self-driving vehicle, is still nascent retail concept. The infrastructure and tech for autonomous movement of goods are rapidly developing, driven by companies like Nuro, Robomart, NEOLIX, and Starship.

However, the primary focus in the autonomous delivery space currently seems centered on optimizing the logistics of delivering pre-ordered items (last-mile efficiency) rather than deploying vehicles that function as retail stores. Nevertheless, we should expect autonomous retail stores to emerge going forward.

News flow so far in 2025:

Nuro: Consistently ranked as a leader (amongst other of course), Nuro utilizes purpose-built autonomous vehicles (not sidewalk bots) for deliveries and has established partnerships with major retailers like Kroger, Walmart, FedEx, and Domino's. Nuro continues testing and operating its R2 pods in several US cities.

Robomart: founded in 2017, Robomart is a prominent company identified as explicitly pursuing the store-hailing model. Currently, its operations involve human-driven, retrofitted vans (the "Oasis" model) conducting beta tests and initial commercial activities, primarily in Southern California (Los Angeles) with some testing in Boston. Partnerships with consumer brands like Unilever, Mars, and Ben & Jerry's confirm its focus on convenience goods, snacks, and potentially prepared foods.

Estonian Starship surpassed 8 million autonomous deliveries

Starship Technologies: A pioneer in sidewalk delivery robots, Starship surpassed 8 million autonomous deliveries globally by early 2025. Operating over 2,000 Level 4 autonomous robots in 150+ locations (primarily university campuses and neighborhoods), they have significantly normalized the concept of robotic delivery.

Leveraging recent funding, PIX Moving began pilot production of the 'Haven' skateboard chassis for Robomart while also advancing deployments of its own branded 'Robo-Shop' autonomous retail vehicles in select international markets.

The landscape of autonomous mobile stores is now forming

Other Players: The field includes drone delivery companies (Zipline, Manna), other sidewalk robot providers (Serve Robotics, Kiwibot, Cartken), autonomous trucking firms involved in logistics (Kodiak, Aurora), and platform integrators like Uber Eats, which partners with multiple robot companies (including Nuro and Serve) for autonomous food delivery in various cities. Gatik operates autonomous box trucks for Walmart deliveries. Chinese companies like JD Logistics and Alibaba/Cainiao are also active, primarily in controlled domestic environments.

This is a such a nascent space, that it’s difficult to find research and data for it

Market reports in early 2025 forecast strong compound annual growth (often cited above 25-35%) for the autonomous delivery robot sector through 2030. Key drivers identified include booming e-commerce, rising labor costs, and increasing demand for contactless, efficient last-mile solutions.

---

Until next time,

Christopher Lyrhem

Chief Future Officer at Sircular

christopher@sircular.io