Global AI Briefing: Publicis Acquires LiveRamp, Anthropic's $900B Valuation, and New Interaction Models Reshape the Intelligence Age

The artificial intelligence ecosystem on 18 May 2026 is characterized by a high-velocity convergence of massive capital realignments, radical architectural departures from the transformer paradigm, and a burgeoning energy crisis that threatens the physical scalability of the "Intelligence Age." This Global Artificial Intelligence Intelligence Briefing, Sponsored by Best-AI.org, highlights a critical week for the industry, marked by significant market events, groundbreaking product releases, and evolving regulatory discussions. From Publicis Groupe's strategic acquisition of LiveRamp to Anthropic's record-breaking fundraising and the debut of innovative AI interaction models, these developments underscore AI's rapid integration into global commerce, governance, and human interaction.

Major Market Shifts and Strategic Acquisitions

Today's most substantial market-moving event is the announcement of Publicis Groupe’s definitive agreement to acquire LiveRamp for a total enterprise value of $2.167 billion in an all-cash transaction. This acquisition, priced at $38.50 per share — a 29.8% premium to its last trading close, represents a strategic inflection point for the global advertising and marketing industry. By integrating LiveRamp’s data collaboration platform, which connects over 25,000 publisher domains and 500+ technology partners, Publicis aims to accelerate "data co-creation for smarter agents." The strategic rationale centers on transitioning from traditional programmatic advertising to autonomous sales and marketing agents that require high-fidelity, interoperable data to function deterministically. Publicis has consequently raised its 2027 and 2028 net revenue growth objectives to 7%–8%, signaling confidence that AI-driven data unification will be the primary driver of agency-side value in the latter half of the decade.

Simultaneously, the foundational lab landscape is being reshaped by Anthropic’s imminent closure of a record-breaking fundraising round. Reports circulating today suggest the round is expected to secure between $30 billion and $50 billion at a valuation exceeding $900 billion, potentially surpassing OpenAI’s last private valuation of $852 billion. This surge in capital is driven by Anthropic’s infrastructure-scale requirements, with the firm projecting revenue of $18 billion in 2026, scaling toward $148 billion by 2029. However, this growth comes with significant financial complexity, as Anthropic anticipates losing $11 billion annually through 2027, primarily due to "compute commitments" with AWS and Google Cloud. This "circular financing" model, where hyperscalers invest in the very labs that pay them for cloud capacity, has become a systemic characteristic of the 2026 AI market.

In the semiconductor and hardware sector, Tokyo-based Kioxia Holdings has reported a massive profit surge, with shares being "untraded" in a glut of buy orders on the morning of 18 May 2026. The storage provider for AI data centers has seen its valuation triple this year as the demand for high-speed SSDs and memory-centric computing architectures outstrips current supply. This contrasts sharply with the broader emerging market narrative, where India’s stock market is on the verge of dropping out of the world’s top five as global investors rotate capital toward AI-rich benchmarks in Taiwan and South Korea. While Korea and Taiwan’s benchmarks have surged by 78% and 42% respectively, India’s Nifty IT Index has dropped more than 26%, reflecting a structural revaluation of economies that lack sovereign AI infrastructure and domestic chip manufacturing.

Innovations in AI Models and Operating Systems

Today’s most compelling technical updates center on the official release of Thinking Machines Lab’s (TML) research preview for its first "interaction model," TML-Interaction-Small. Founded by former OpenAI CTO Mira Murati, TML is challenging the industry’s reliance on "turn-based" interfaces. The model is a 276-billion parameter mixture-of-experts (MoE) system with 12 billion active parameters per token, trained from scratch to process audio, video, and text in 200-millisecond "micro-turns." Unlike the prevailing approach — where a standard LLM is wrapped in an external voice-activity-detection (VAD) harness, TML-Interaction-Small utilizes "encoder-free early fusion," allowing raw signals to be processed directly through lightweight embedding layers within the transformer. This architecture enables the AI to listen and speak simultaneously, interject when it perceives a visual or auditory error, and maintain presence without the "awkward pause" characteristic of 2025-era assistants, paving the way for more natural human-AI collaboration.

In a parallel development, Google is providing pre-keynote leaks regarding its "Aluminum OS" (or "Aluminium OS"), a radical replacement for ChromeOS that is expected to debut at I/O tomorrow. The operating system is built on Android 17 but features a genuine desktop window manager and a signature feature called "Magic Pointer." Developed in collaboration with the DeepMind team, Magic Pointer turns the system cursor into a context-aware AI agent that can summarize text, schedule meetings from email dates, or composite images simply by hovering and "wiggling" over elements. This represents a shift from "AI as a feature" to "AI as the kernel," where the OS exists to facilitate Gemini’s interactions with the user’s local and cloud data.

Furthermore, the "Architecture War" has escalated with the commercial launch of SubQ 1M-Preview. As the first commercially available non-transformer LLM, SubQ utilizes a sparse subquadratic attention mechanism that allows for an unprecedented 12-million-token native context window. The developers claim this model achieves a 52x speedup in attention at scale compared to standard transformers, effectively reducing the cost of long-context workloads by 80%. This release, alongside Zyphra’s AMD-trained ZAYA1-8B, indicates that the competitive focus is shifting from raw parameter scale to "intelligence density" and hardware-agnostic training paths. For those interested in optimizing their AI budgeting, these new models offer compelling alternatives.

The Energy-Intelligence Paradox and Infrastructure Challenges

A major investigative report from The Guardian on 18 May 2026 highlights a growing "energy-intelligence paradox" in the United Kingdom. More than 100 new UK data centers are currently planning to burn gas for onsite electricity generation because the wait for a connection to the National Grid has reached 15 years. These operators have requested gas connections amounting to over 15 terawatt hours per year—equivalent to the energy needs of London for four and a half months. This reliance on fossil fuels to power the AI revolution has raised alarms regarding the UK’s climate targets, particularly the Clean Power 2030 goal. Stuart Okin, director of cyber regulation and AI at Ofgem, noted that there are 100GW of data center projects in the queue, indicating that the physical capacity of the grid is becoming a hard cap on AI expansion.

Evolving Regulatory Landscape and Academic Integrity

On the legislative front, the US Senate is evaluating the TRUMP AMERICA AI Act, a sweeping 291-page discussion draft released by Senator Marsha Blackburn. The bill seeks to preempt the "patchwork" of state-level AI regulations by establishing a single federal framework focused on four "Cs": Children, Creators, Conservatives, and Communities. Notably, Title III of the bill would repeal Section 230 of the Communications Act after two years, fundamentally altering the liability landscape for platforms hosting AI-generated content. The bill also takes a hardline stance on intellectual property, declaring that the unauthorized use of copyrighted material for AI training does not constitute fair use, a move that contradicts the White House’s more flexible National AI Policy Framework released on March 20.

The academic community is also reacting to the proliferation of low-quality AI content. The preprint repository arXiv has officially instituted a one-year ban for authors who submit papers containing "incontrovertible evidence" of unchecked LLM generation. This includes meta-comments left by bots (e.g., "This table is illustrative; please enter actual values") and hallucinated citations. Thomas G. Dietterich, head of the computer science section at arXiv, stated that authors assume full responsibility for content regardless of how it was generated, marking a new era of accountability in research.

Why These Developments Matter for the Future of AI

The rapid pace of innovation and market consolidation highlighted in this AI news briefing signals a maturing yet still volatile industry. The Publicis acquisition of LiveRamp underscores the critical role of data in powering the next generation of AI agents for marketing and sales. Anthropic's massive fundraising, despite projected losses, illustrates the immense capital required for foundational model development and the unique "circular financing" dynamics with hyperscalers. The emergence of TML-Interaction-Small and Google's "Aluminum OS" points to a future where AI is not just a feature but the core operating system, enabling more intuitive and real-time human-computer interaction. Furthermore, the "Architecture War" with non-transformer models like SubQ 1M-Preview indicates a diversification in AI model design, potentially leading to more efficient and specialized AI tools.

However, these advancements are not without challenges. The UK's "energy-intelligence paradox" reveals a looming physical constraint on AI's growth, demanding sustainable infrastructure solutions. Simultaneously, legislative efforts like the TRUMP AMERICA AI Act and arXiv's new policies reflect a growing need for robust regulatory frameworks and accountability standards to manage the ethical and practical implications of widespread AI adoption. These intertwined developments shape the trajectory of the "Intelligence Age," demanding careful consideration from developers, policymakers, and consumers alike.

What to Watch Next in the Intelligence Age

As the industry enters a critical week, all eyes will be on the impending Google I/O 2026 keynote for further details on "Aluminum OS" and other AI innovations. Additionally, pivotal earnings reports from the semiconductor sector, particularly NVIDIA, will provide crucial insights into the hardware backbone supporting this rapid AI expansion. The ongoing discussions around AI regulation and the practical implications of new model architectures will continue to define the landscape, making the coming months pivotal for the future of artificial intelligence.

Sources

• https://www.publicisgroupe.com/en/news/press-releases/publicis-to-acquire-liveramp-to-accelerate-data-co-creation-for-smarter-agents
• https://www.buildfastwithai.com/blogs/ai-news-today-may-18-2026
• https://m.economictimes.com/markets/stocks/news/india-missed-out-on-ai-and-now-its-run-as-market-darling-may-be-over/articleshow/131163876.cms
• https://simplywall.st/stocks/us/software/nasdaq-gen/gen-digital/news/the-bull-case-for-gen-digital-gen-could-change-following-str
• https://www.theinformation.com/