What We Mean by a Slowdown

A slowdown doesn’t mean halting. Adoption remains widespread, and investments are still climbing. However, various signals indicate that the boom phase is transitioning into a more measured growth:

• According to McKinsey’s 2024 State of AI report, enterprises are moving from pilot projects to full production while scrutinizing return on investment (ROI) more closely (McKinsey).
• With the skyrocketing demand for computing power, infrastructure costs and energy use are coming under pressure, with data centers’ electricity needs expected to surge significantly this decade (IEA).
• Licensing data has become essential for model quality and legal assurance, leading companies to sign costly, long-term content agreements.

Why Distribution Matters Just as Much as Model Quality

In the initial wave of AI development, the biggest progress came from training larger models. Nowadays, the advantage often lies in integrating trustworthy AI into everyday workflows. This is where distribution partnerships become crucial. Platforms with vast audiences—such as social networks, search engines, productivity tools, and smartphones—are prime avenues for AI assistants and agents.

TikTok has already dipped its toes into AI, experimenting with AI helpers like the “Tako” chatbot in select markets and launching creative tools for advertisers (The Verge; TikTok for Business). For AI model developers, being incorporated within such a platform could provide extensive reach, engagement data, and new monetization opportunities. For platforms, the attraction lies in enhanced creator tools and richer content experiences.

However, distribution has political implications. In the United States, lawmakers have passed legislation that could compel TikTok to divest from its Chinese parent company, ByteDance, or face a ban, raising regulatory uncertainty around deep integrations (Reuters).

The Cost Gravity: Compute, Energy, and Data

Compute Bottlenecks and Capital Expenditures

Training and serving models still depend on scarce, high-cost hardware. Cloud service providers and AI labs have poured billions into GPUs and custom silicon. Microsoft, for example, reported a significant increase in capital expenditure related to AI infrastructure in 2024 (CNBC). Even as next-gen chips promise better performance per dollar, accessibility and power costs remain limiting factors.

Energy and Water Footprints

The environmental impacts of AI are becoming clearer. The International Energy Agency anticipates that electricity demand from data centers will keep climbing, with AI being a major contributor (IEA). Google has also reported increased water consumption associated with cooling systems in AI-powered data centers (Google Environmental Report 2024).

Data Licensing Becomes the Standard

As lawsuits and quality demands escalate, AI developers are entering licensing agreements to secure trusted training and product data. OpenAI has formed partnerships with publishers including News Corp and the Financial Times, as well as platforms like Reddit for accessing real-time content (News Corp; FT; OpenAI). Meanwhile, litigation—such as The New York Times’ lawsuit against OpenAI—highlights the stakes of unlicensed content usage (NYT).

Regulation is Coming, and It’s Changing the Game

Policymakers are moving from discussions to concrete regulations. The EU’s AI Act lays down a tiered risk framework that will gradually introduce requirements for transparency, safety, and oversight (European Parliament). In the U.S., a comprehensive executive order outlines directives on safety testing and security, while agencies examine competition and consumer protection issues (White House).

These frameworks won’t stifle innovation, but they will introduce additional compliance costs and coordination efforts, especially for general-purpose models and high-risk applications. Expect compliance, auditability, and data provenance to be key competitive advantages.

The Next Chapter: From Chatbots to Agents, and from Giant to Efficient Models

Even in this period of cooling, product innovation remains active. Three trends stand out:

• Agents and Automation: Transitioning beyond simple Q&A to executing multi-step tasks such as research, booking, and workflow management. This evolution will challenge reliability and safety at scale.
• Multimodal by Default: The integration of voice, image, and video understanding is becoming standard. OpenAI’s 2024 updates highlighted cost-effective multimodal options for developers (OpenAI).
• Smaller, Efficient Models: Compact, finely-tuned models are being developed for on-device applications to gain advantages in privacy, latency, and cost, as seen with Apple’s recent announcements and Google’s Gemini Nano (Apple; Google).

Open-source developments like Llama 3 and emerging models from independent labs are also speeding ahead, offering builders more choices and negotiation power (Meta; Mistral).

So, Is the AI Boom Slowing?

Yes and no. The rapid growth fueled by novelty and abundant capital is beginning to ease. What lies ahead is a more stable phase where success will favor those who can deliver solutions safely, manage costs, and demonstrate real value in workflows. In this light, speculated partnerships between model developers and platforms like TikTok make strategic sense: they exchange raw computational power for reach, data feedback loops, and lasting user engagement.

The next stage of AI growth will focus less on who possesses the largest model and more on who can provide the most trusted and useful experience at the lowest marginal cost.

What to Watch in the Coming Year

• Distribution Deals: Which platforms will integrate assistants natively, and under what conditions?
• Unit Economics: Trends in inference costs, energy efficiency, and on-device processing.
• Regulatory Milestones: Timelines for the EU AI Act, guidance from U.S. agencies, and any rulings affecting platform collaborations.
• Data Strategy: More licenses from publishers and platforms, along with tools for content provenance and attribution.

FAQs

Is the AI Market a Bubble?

Parts of the market may have been overhyped, but the underlying adoption is genuine. A more selective phase is likely, where value will become concentrated in products with clear economics and distribution paths.

Does AI Use Too Much Energy?

AI contributes to increased data center demand; however, efficiency improvements, better scheduling, and the use of clean energy can help mitigate the impact. Policymakers and providers are focusing on all three aspects (see the IEA report in Sources).

Are Smaller or Open Models Overtaking Closed Giants?

It varies by task. For many enterprise applications, compact or open models with retrieval capabilities can perform well at a lower cost, while large models still excel in frontier research.

How Do Regulations Affect AI Rollouts?

Regulation introduces additional compliance steps and transparency requirements. Teams that prioritize safety, data provenance, and auditability early will have a smoother rollout later on.

What Does TikTok Have to Do with AI’s Trajectory?

Platforms like TikTok shape audience engagement and content distribution. Deep integrations or partnerships can accelerate AI adoption while also raising policy and trust concerns, especially amid shifting platform regulations.