Artificial Influence: The Rise of Advertising in AI
Research Rundown #169, plus new memos on Nominal, Varda, and more.
Research Rundown
On January 16, 2026, OpenAI released a widely-anticipated update on its plans for including ads in AI outputs for lower-paying tier users. This news comes as the company, as well as other major AI frontier labs, has discussed the potential of ads in AI outputs to increase revenue. The primary revenue channels for existing AI companies are enterprise and consumer subscriptions to AI tools, which include chatbots, coding assistants, and potential new products like personal health platforms or wearables. This is true of both hyperscaler AI teams and frontier AI labs. For each major AI provider, enterprise offerings are almost always monetized, though the share of total revenue attributed to enterprise sales varies by company. Consumer usage, on the other hand, is 97% unpaid.
AI companies operating under this freemium structure for consumer usage are not yet profitable, and some analysts don’t believe they will become profitable without major changes to their business models or significant reductions in compute costs. While the exact cost for individual consumer users varies by user and tool, both free and paid users can cost AI companies money. ChatGPT users, for example, pay $0 (Free), $8 (Go), $20 (Plus), or $200 (Pro) per month. The cost of using ChatGPT varies by user, but it can run OpenAI over $1K per query for large, context-rich requests. Some individual users of coding tools have run bills of over $35K in one month, while paying only $200 for the service.
It isn’t likely that these companies would consider abandoning free usage altogether, given the strength of the freemium model in increasing active users and the potential to convert this base to paid users over time. This reality is continually forcing AI providers to consider additional sources of revenue, including integrating ads into outputs. Read more about how AI platforms are thinking about integrating ads and what it will take to break even in our deep dive here.
Nominal gives modern aerospace and defense teams real-time visibility into machines that can’t fail. Check out the full memo here to learn more.
Varda is building the missing supply chain between space and terrestrial industry. Find out more in our memo here.
By turning clinical trials into a software problem, Formation Bio is compressing the path to new medicines. See the full breakdown in our memo here.
xAI is positioning itself as the AI lab that learns directly from the world as it unfolds. Find out more in our memo here.
Oracle has finalized a localized data-residency agreement with ByteDance to manage TikTok’s US user metadata through the “Project Texas” evolution involving rigorous source-code auditing and isolated server instances. This agreement utilizes Oracle’s Sovereign Cloud architecture to ensure that zero-trust protocols prevent unauthorized cross-border data exfiltration to ByteDance’s domestic clusters.
Anthropic made headlines multiple times this week, including with news that the company’s annualized revenue has surpassed $9 billion, driven by high-volume API consumption from enterprise-scale Retrieval-Augmented Generation (RAG) implementations and fine-tuning services. Anthropic’s current funding round, valuing the company at $350 billion, is oversubscribed.
Anthropic updated the Constitutional AI training objectives for Claude this week, incorporating new heuristic guidelines designed to govern emergent behaviors that mimic high-order self-awareness during long-context reasoning. The update raised eyebrows for addressing the possibility of model consciousness, formally acknowledging that Claude’s moral status is “deeply uncertain.”
CEO Dario Amodei advocated for stricter computational export restrictions at Davos this week, arguing that the transfer of advanced H100-class silicon and beyond to China is tantamount to “selling nuclear weapons to North Korea”. He contended that compute-asymmetry is the primary determinant in maintaining the safety and security of global frontier AI development.
Sam Altman is negotiating a $50 billion funding round with Middle Eastern sovereign wealth funds to finance the massive CAPEX required for global GPU clusters and dedicated energy infrastructure. This capital injection is structured to support OpenAI’s transition from general-purpose LLMs to vertically integrated, hardware-heavy compute environments.
Medical AI platform OpenEvidence has secured a $12 billion valuation in a round led by Thrive and DST, highlighting the premium placed on domain-specific, high-fidelity reasoning engines even as general AI companies pursue healthcare offerings. The platform utilizes a proprietary medical-knowledge graph to minimize hallucination rates within clinical decision-support systems.
SambaNova Systems is seeking $500 million in private capital after acquisition discussions with Intel reached an impasse over valuation and IP integration complexities. The company’s Reconfigurable Dataflow Architecture (RDA) is a key alternative to GPU-centric hardware, offering efficiency for large-scale inference workloads.
Reporters shared on Wednesday that Apple is reportedly developing a wearable AI pin that leverages an on-device “Neural Engine” to facilitate real-time, multimodal interaction without relying on a tethered iPhone. This followed news on Monday that OpenAI is accelerating the development of a dedicated consumer device in collaboration with Jony Ive’s LoveFrom, targeting a 2026 release of a “voice-first” agentic interface.
Blue Origin unveiled a satellite internet initiative, TeraWave, designed to compete with SpaceX’s Starlink by leveraging Amazon’s Project Kuiper infrastructure for low-latency global connectivity. The technical framework utilizes a high-altitude orbital mesh network and phased-array antenna technology to provide high-bandwidth backhaul for enterprise and governmental clients.
Tesla officially discontinued its legacy “Autopilot” suite in a bid to force fleet-wide adoption of its end-to-end neural network-based Full Self-Driving (FSD) software. This news comes as the company launched the first fully autonomous Robotaxi service in Austin, now operating with zero human safety-driver oversight within a geofenced urban environment.
At Contrary Research, we’ve built the best starting place to understand private tech companies. If you're interested in researching and writing about tech, apply here for our Research Fellowship!