Article image

The Great Acceleration

Saturday, 06 December 2025 20:55

Summary

The global artificial intelligence sector has entered a phase of unprecedented capital investment and technological maturation, yet this acceleration is running headlong into profound challenges of safety, governance, and intellectual property. Private investment in AI reached record highs in 2025, fuelling a massive infrastructure build-out and a fierce competitive battle among large language model developers, evidenced by the market share shifts away from the early leader, ChatGPT. Simultaneously, the technology's real-world deployment is facing intense regulatory scrutiny, particularly in the autonomous vehicle sector, where federal probes into systems like Tesla's Full Self-Driving and Waymo's robotaxis highlight critical safety failures. The creative industries are grappling with a legal and philosophical crisis over copyright, with landmark court rulings attempting to define the boundaries of fair use for training data. As the European Union implements its comprehensive AI Act, the United States remains divided, creating a fragmented global regulatory landscape that struggles to keep pace with the technology's exponential growth.

The New Industrial Revolution's Capital Surge

The year 2025 marked a decisive inflection point where artificial intelligence transitioned from a cutting-edge research field into a foundational, ubiquitous technology across the global economy29,39. This shift was underpinned by an unprecedented surge in capital, with global private AI funding reaching record highs1,35. Total funding for AI companies globally reached $47.3 billion in the second quarter of 2025, making it the second-highest quarter on record1. The first half of 2025 saw total funding outpace the entirety of the previous year, climbing to $116.1 billion compared to 2024’s $104.7 billion1,12. This financial momentum was concentrated in the United States, which captured approximately 84 per cent of global funding in the second quarter, totalling $39.7 billion1. The US dominance in private AI investment is stark, with its $109.1 billion in 2024 being nearly twelve times higher than China’s and twenty-four times that of the United Kingdom35. The investment is not merely speculative; it is driving a massive infrastructure build-out necessary to support the next generation of large language models (LLMs)36. In September 2025, OpenAI signed a contract with Oracle to purchase $300 billion worth of computing power over five years36. Just days earlier, Nebius Group announced a $17.4 billion deal to provide Microsoft with graphic processing units (GPU) capacity over a five-year period36. These deals, which followed earlier pledges from OpenAI and Meta worth over $560 billion, underscore the industry’s focus on the ‘picks and shovels’ of AI—the semiconductors, cloud computing, and big data technologies that form the bedrock of the revolution36. The financial sector has been quick to adopt LLM-driven solutions, identifying numerous potential use cases for automating workflows and enhancing decision-making39. Mergers and acquisitions (M&A) activity for AI startups surged to a record 177 deals in the first half of 2025, doubling the five-year average1. This consolidation reflects a maturing market where established technology giants are aggressively integrating AI capabilities38. A notable example of this strategic integration occurred in December 2025 when Meta Platforms acquired the AI wearable startup Limitless2,3. Limitless, known for its AI-powered ‘Pendant’ device that records and summarises real-world conversations, was folded into Meta’s Reality Labs wearables organisation7,9. The acquisition aligns with Meta’s stated vision to deliver ‘personal superintelligence to everyone’ through AI-enabled consumer hardware3,11.

The Battle for the Digital Mind

The initial phase of explosive, uncontested growth for the first wave of generative AI tools has concluded, giving way to a fierce competitive landscape17. The market leader, OpenAI’s ChatGPT, showed clear signs of maturation in late 2025 as its user growth slowed dramatically14,17. According to data from Sensor Tower, ChatGPT’s global monthly active users (MAUs) grew by only 6 per cent between August and November 2025, reaching approximately 810 million users14,17. This deceleration, following a year-over-year growth rate of 180 per cent, suggests the platform may be approaching market saturation, particularly in early-adopter regions like the US and parts of Europe17,18. The slowdown was compounded by the rapid acceleration of key rivals14. Google’s Gemini, for instance, saw its user base surge by 30 per cent during the same three-month period14,18. This growth was significantly boosted by the success of its Nano Banana image generation model, which helped Gemini users double their average daily time spent in the application14,18. While ChatGPT maintained a leading 55 per cent share of global MAUs, Gemini gained three percentage points of market share between May and November 202514,18. The competitive pressure is not limited to the major players; smaller, specialised models are also gaining traction18. Perplexity saw explosive 370 per cent year-over-year growth, and Claude jumped 190 per cent, demonstrating that the market is fragmenting beyond the initial duopoly14,18. Internally, the competitive threat was acknowledged by OpenAI, with reports of a ‘code red’ memo from CEO Sam Altman urging enhancements to the technology14,18. The shift in user behaviour indicates that the novelty phase of AI adoption is over, and the market is transitioning to one focused on utility, integration, and feature differentiation17,23. The future of LLMs is increasingly multimodal, with systems capable of generating and processing text, images, audio, and video simultaneously, a capability that will further blur the lines between different forms of creative and professional expression29,31.

The Creative Class and the Copyright Crucible

The rapid advancement of generative AI has precipitated a legal and philosophical crisis within the creative industries, centring on the use of copyrighted material for training LLMs28. The core legal question revolves around whether the ingestion of vast datasets, including copyrighted books, music, and images, constitutes ‘fair use’ under intellectual property law28,40. The first major fair use rulings in generative AI training cases arrived in June 2025, offering early, albeit nuanced, victories for AI developers32. In the *Bartz v. Anthropic* class action, a federal judge ruled that Anthropic’s use of copyrighted works to train its Claude LLMs was fair use, analogising the process to human learning28,32. However, the same judge distinguished this from the act of copying and storing over seven million pirated books in a central library, which was deemed copyright infringement28,32. Anthropic subsequently agreed to a $1.5 billion settlement28. Similarly, in *Kadrey v. Meta*, the court found that training Meta’s Llama LLMs on copyrighted material was fair use, citing a lack of meaningful evidence that the AI’s output caused market dilution for the original authors28. These rulings have highlighted the difficulty faced by plaintiffs in demonstrating direct market harm, even as the potential for AI to flood the market with competitive content is acknowledged by the courts28. The US Copyright Office has maintained a clear stance that human authorship is a prerequisite for copyright protection, a position upheld by a federal court in the *Thaler v. Perlmutter* case40. Beyond the courtroom, the philosophical debate over AI’s role in art has intensified25. Acclaimed filmmaker James Cameron, speaking ahead of the release of his film *Avatar: Fire and Ash*, expressed his strong disdain for generative AI actors15,24. Cameron, a director at the UK-based AI company Stability AI, called the idea of generative AI creating a performance from a text prompt “horrifying”15,22. He argued that because AI models are trained on everything that has been done before, their output is inherently an “average” of human art and experience15. Cameron stressed that his own work, which relies heavily on performance capture, is a “celebration of the actor-director moment” and that he would not replace actors with generative AI24,26. He conceded that the technology could potentially be useful in making visual effects (VFX) cheaper, thereby enabling more expensive, imaginative films to be greenlit22,24.

The Autonomy Paradox: Safety and Scrutiny

The deployment of AI in safety-critical applications, particularly autonomous vehicles, has brought the technology’s limitations and the need for stringent oversight into sharp focus42. The National Highway Traffic Safety Administration (NHTSA) in the US has escalated its regulatory scrutiny of the two leading autonomous driving systems13,4. In December 2025, Waymo, the self-driving unit of Alphabet, announced a voluntary software recall after an NHTSA investigation into its robotaxi fleet4,5. The probe was initiated after Waymo vehicles were observed failing to stop for school buses that had their stop signs extended and red lights flashing4,5. Incidents were reported in Atlanta and Austin, Texas, with Austin school officials documenting 19 illegal passes this year, including at least five after a November 17 software update intended to improve performance5,8. Waymo’s chief safety officer, Mauricio Peña, stated that the company was filing the recall to address the issue of appropriately slowing and stopping in these scenarios4. This was the second software recall for Waymo in 2025, following an earlier one for 1,200 robotaxis involved in minor collisions with gates, chains, and similar roadway objects4,10. Concurrently, the NHTSA expanded its investigation into Tesla’s Full Self-Driving (FSD) software13,16. The regulator documented at least 80 instances where the AI-powered system allegedly violated basic traffic laws, a significant jump from the approximately 50 violations cited when the investigation began in October13,20. The violations included running red lights, improper lane use, and drifting into wrong lanes, with the evidence drawn from 62 customer complaints, 14 Tesla-submitted reports, and four media accounts13,20. The expanded probe covers nearly 2.9 million Teslas and seeks to determine if the FSD system can accurately detect and respond to traffic signals and provide adequate driver warnings16,20. Tesla was formally required to respond to the regulator’s queries by January 19, 202613,20. These incidents underscore the ‘autonomy paradox,’ where the promise of superior safety records compared to human drivers is undermined by systemic, unpredictable failures in complex, real-world scenarios5,16.

The Regulatory Patchwork and the Race for Governance

The rapid technological progress has forced governments and international bodies to scramble to establish regulatory frameworks, resulting in a fragmented global governance landscape34,44. The European Union has continued to lead the world in comprehensive AI regulation with the implementation of the EU AI Act30. Key milestones for the Act are set for 2025, including the ban on AI systems deemed inherently harmful, such as those deploying subliminal or manipulative techniques, which is set to come into force in February30. The Act also introduces rules for General Purpose AI (GPAI) models, establishing a risk-based approach that is expected to have an extraterritorial impact beyond EU borders30. In contrast, the United States has adopted a more permissive, innovation-first approach at the federal level34. In January 2025, the Trump administration issued an Executive Order for Removing Barriers to American Leadership in AI, which rescinded the previous administration’s more risk-focused order34. This was followed in July 2025 by the America’s AI Action Plan, which identified over 90 federal policy actions aimed at securing US AI leadership34. This federal stance, however, exists in tension with a growing patchwork of state-level legislation43. States like Texas and Illinois have enacted their own AI in hiring and governance laws, with the Texas Responsible Artificial Intelligence Governance Act (TRAIGA) and the Illinois AI in hiring law both set to take effect on January 1, 202643. Colorado’s AI in hiring law is scheduled to follow in June 202643. This divergence creates a complex compliance environment for businesses operating across state lines43. The global regulatory race is further complicated by the philosophical challenge of ‘alignment,’ the effort to ensure AI systems work toward human good26. As James Cameron noted, the lack of human consensus on ethics and morality makes it nearly impossible to define whose sense of what is best should prevail in an AI system26. The current moment in AI governance is often compared to the early days of the commercial internet, but the cycle of innovation, harm, and regulation is accelerating, with the entire process expected to unfold in a fraction of the time it took for the internet to mature44.

Conclusion

The current era of artificial intelligence is defined by a powerful duality: a technological capability that is advancing at an exponential rate, matched by a regulatory and ethical framework that is struggling to keep pace. The record-breaking capital flowing into the sector is not merely funding incremental improvements; it is financing a fundamental restructuring of global industry, from the automation of finance to the creation of new forms of digital content. Yet, the real-world deployment of these systems, particularly in high-stakes domains like autonomous driving, has exposed critical vulnerabilities that cannot be solved by simply adding more data or computing power. The safety probes into Waymo and Tesla FSD, alongside the complex copyright battles in the creative sphere, demonstrate that the transition from laboratory breakthrough to reliable, ethical public utility is fraught with risk. As the market for LLMs matures and competition intensifies, the focus will inevitably shift from raw capability to trustworthiness and responsible integration. The fragmented global response to governance, with the EU pursuing a risk-averse, comprehensive framework and the US prioritising innovation and leadership, suggests that the future of AI will be shaped by a geopolitical and regulatory tug-of-war. Ultimately, the success of this new industrial revolution will hinge not just on the speed of its technological acceleration, but on the collective ability of policymakers, developers, and the public to establish a consensus on the ethical guardrails necessary to manage a technology that is rapidly becoming the most powerful tool in human history.

References

  1. AI Investment reaches all-time highs: The State of AI Fundraising: (2025-09-04)

    Supports the financial data on global AI investment, Q2 and H1 2025 funding totals, US dominance, and the surge in M&A activity.

  2. Meta's latest acquisition suggests hardware plans beyond glasses and headsets - Engadget: (2025-12-05)

    Confirms the acquisition of Limitless by Meta in December 2025 and details the nature of the startup's product (Pendant, formerly Rewind).

  3. Meta To Acquire AI Wearable Startup Limitless - Nasdaq: (2025-12-06)

    Verifies the Meta/Limitless acquisition and quotes the CEO's statement about Meta's vision for 'personal superintelligence'.

  4. Waymo's robotaxi fleet is being recalled again, this time for failing to stop for school buses: (2025-12-06)

    Details the December 2025 Waymo software recall, the reason (failing to stop for school buses), the NHTSA investigation, and mentions the earlier 2025 recall for collisions with gates/chains.

  5. Waymo to recall robotaxi software after school bus incidents spark scrutiny from feds: (2025-12-05)

    Provides specific details on the Waymo incidents, including the locations (Atlanta and Austin) and the nature of the software issue (initially slowing, then proceeding).

  6. Meta acquisition of AI wearable startup Limitless deal: (2025-12-05)

    Confirms Limitless was formerly known as Rewind and that the team will be integrated into Reality Labs' wearables organisation.

  7. Waymo recalls software after robotaxis don't stop for school buses - NewsBytes: (2025-12-06)

    Cites the Austin School District's report of 19 illegal passes this year, including five after the November 17 software update.

  8. Meta Acquires AI Wearables Startup Limitless - Alpha Spread: (2025-12-05)

    Confirms the Pendant device's function of recording and transcribing real-world conversations.

  9. Waymo recalls 1200 robotaxis following low-speed collisions with gates and chains - Reddit: (2025-05-15)

    Provides details on the earlier 2025 Waymo recall involving 1,200 robotaxis and collisions with gates/chains.

  10. Meta acquires AI wearable startup Limitless - Investing.com: (2025-12-05)

    Reinforces the alignment of the Limitless acquisition with Meta's vision to bring 'personal superintelligence to everyone'.

  11. The Historic AI Investment Surge of July 2025: A Watershed Moment for Artificial Intelligence: (2025-08-01)

    Supports the H1 2025 funding total of $116.1 billion.

  12. NHTSA Finds 80 Tesla FSD Violations, Expands Safety Investigation | The Tech Buzz: (2025-12-05)

    Provides the specific number of documented Tesla FSD violations (80), the nature of the violations (running red lights, wrong lanes), and the deadline for Tesla's response (January 19, 2026).

  13. ChatGPT Growth Stalls as Gemini Surges 30% in User Battle - The Tech Buzz: (2025-12-05)

    Cites Sensor Tower data on ChatGPT's 6% growth (August-November 2025), Gemini's 30% surge, the Nano Banana model, and the 'code red' memo.

  14. James Cameron says AI actors are 'horrifying to me' - The Guardian: (2025-12-01)

    Provides James Cameron's quote calling generative AI actors 'horrifying' and his philosophical argument that generative AI creates an 'average' because it is trained on existing art.

  15. Tesla FSD Is Under Federal Investigation. Again. Should You Still Turn It On?: (2025-12-06)

    States the scope of the NHTSA investigation (nearly 2.9 million Teslas) and the nature of the violations (running red lights, steering into oncoming lanes).

  16. ChatGPT's user growth is slowing down - Lapaas Voice: (2025-12-06)

    Supports the 6% MAU growth figure (August-November 2025) and the conclusion that the platform may be nearing market saturation.

  17. ChatGPT's rapid surge slows down as Gemini outshines its growth - Daily Jang: (2025-12-06)

    Confirms the 55% market share for ChatGPT, the 30% surge for Gemini, and the strong growth of rivals like Perplexity and Claude.

  18. Tesla FSD Safety Complaints Detail Traffic Light Violations - Bangla news - ZoomBangla: (2025-12-06)

    Details the breakdown of the 80 cited cases (62 complaints, 14 Tesla reports, 4 media accounts) and the January 19, 2026 response deadline.

  19. James Cameron on having AI generative actors: That's horrifying - The Economic Times: (2025-12-01)

    Quotes Cameron on the potential for AI to make VFX cheaper, which could enable more expensive films.

  20. ChatGPT Mobile Growth Slows as Gemini, Claude, and Copilot Gain Ground: (2025-10-17)

    Supports the idea that the experimentation phase for ChatGPT is over and the platform is transitioning from viral novelty to a utility tool.

  21. Avatar Creator James Cameron Says Using AI to 'Make Up a Performance From Scratch With a Text Prompt' Is 'Horrifying to Me' - IGN: (2025-12-01)

    Confirms Cameron's comments were made ahead of the release of *Avatar: Fire and Ash* and his praise for performance capture as a 'celebration of the actor-director moment'.

  22. James Cameron Warns Against Generative AI in Filmmaking: Preserving Human Creativity: (2025-12-01)

    Supports the idea that Cameron's comments highlight a growing divide in the entertainment industry.

  23. Terminator Creator James Cameron Reveals the Real Problem With AI, And Why It Scares Him - ComicBook.com: (2025-12-02)

    Provides Cameron's quote on the 'alignment' problem and the lack of human consensus on ethics and morality.

  24. Artificial intelligence: courts weigh in on clash with creatives over copyright: (2025-11-27)

    Details the *Bartz v. Anthropic* and *Kadrey v. Meta* rulings, including the $1.5 billion settlement and the legal distinction between training and storing pirated works.

  25. The Rise of Large Language Models: What's New in 2025 – IT Exams Training - Pass4sure:

    Supports the claim that AI is a foundational technology in 2025 and mentions the trend toward multi-modal models.

  26. 2025: Global AI governance takes shape - what to expect from the EU and US: (2024-12-16)

    Details the EU AI Act's key milestones for 2025, including the ban on prohibited AI systems in February and the introduction of rules for General Purpose AI (GPAI) models.

  27. AI's next leap: key generative AI trends in 2025 - Toloka AI: (2025-01-03)

    Supports the trend of multimodality in 2025, where AI processes and combines information from multiple formats.

  28. AI Intellectual Property Disputes: The Year in Review | 12 - Debevoise & Plimpton LLP: (2025-12-02)

    Confirms the June 2025 date for the first major fair use rulings and the distinction made in *Bartz v. Anthropic* between training (fair use) and storing pirated works (infringement).

  29. AI Watch: Global regulatory tracker - United States | White & Case LLP: (2025-09-24)

    Details the US federal regulatory approach, including the January 2025 Executive Order and the July 2025 America's AI Action Plan, contrasting it with the EU's risk-focused approach.

  30. Economy | The 2025 AI Index Report | Stanford HAI:

    Provides the specific figures for US, China, and UK private AI investment in 2024, supporting the claim of US dominance.

  31. AI investments surge in 2025, driving market gains, fund flows | Blog posts - STOXX: (2025-09-24)

    Cites the massive infrastructure deals in September 2025, specifically the $300 billion OpenAI/Oracle and $17.4 billion Nebius/Microsoft contracts.

  32. Large Language Models: Evolution, State of the Art in 2025, and Business Impact | Proffiz: (2025-04-11)

    Supports the claim that LLMs are at the core of new enterprise solutions in 2025, including in finance and law.

  33. The Legal Consequences of the AI Copyright Wars of 2025: (2025-11-17)

    Confirms the US Copyright Office's stance that human authorship is required for copyright protection and the *Thaler v. Perlmutter* decision.

  34. McKinsey technology trends outlook 2025: (2025-07-22)

    Supports the general context of AI's evolution and the need for thoughtful approaches to safety and governance.

  35. State laws regulating AI take effect in the new year. Here's what HR needs to know.: (2025-12-05)

    Provides details on the US state-level regulatory patchwork, citing the Texas and Illinois AI laws taking effect in January 2026 and Colorado's in June 2026.

  36. AI Governance Is Where the Internet Was in 1998 | by Rosemi Mederos - Medium: (2025-12-03)

    Offers the comparison between the current state of AI governance and the early days of the commercial internet, noting the accelerating cycle of regulation.