OpenAI’s Ad Pivot: When Assistants Become Ad Surfaces

Welcome to Memorandum Deep Dives. In this series, we go beyond the headlines to examine the decisions shaping our digital future. 🗞️

Today, we’re diving into OpenAI’s move to test ads inside ChatGPT—and why it may mark the moment “subsidized AI” quietly died.

For the past couple of years, the world has watched AI labs unveil ever more powerful models—reshaping investment markets, geopolitics, and the conversation around work, infrastructure, and regulation. But behind the breakthroughs sits a harder problem: how OpenAI, Anthropic, Google, and Microsoft pay for intelligence at scale, where costs grow with every new user and every extra capability. In 2026, that tension is pushing the industry back toward an old answer with new consequences: advertising as the revenue engine for the AI era.

From breakthrough to balance sheets

As of January 2026, the industry is shifting from growth-at-all-costs to sustainable revenue generation.

Starting with OpenAI, the company has announced it will begin testing advertisements within ChatGPT. Users will find themselves paying for a service that still shows them commercials. This shift isn't just about greed; it’s a cold calculation of the cost of compute.

Google, meanwhile, is playing a high-stakes game of self-cannibalization. The company is working to replace the ten blue links with AI overviews. This essentially allows the company to display direct offers to users within the chat interface.

And, while Google VP Dan Taylor insists the standalone Gemini app will remain ad-free to preserve its role as a utility, the reality is that Google’s real AI, the one used by 2 billion users worldwide, is being turned into a hyper-personalized billboard.

When compute costs catch up

The introduction of advertising into the ChatGPT ecosystem marks a watershed moment, shifting the focus from user acquisition to unit economics.

The reason behind the shift is clear.

OpenAI Chief Financial Officer Sarah Friar said in a blog post that the company's annualized revenue has surpassed $20 billion in 2025, up from $6 billion in 2024, with growth closely tracking an expansion in computing capacity.

However, despite the growth, the company reportedly expects to rack up massive annual losses, including roughly $74 billion in operating losses in 2028 alone.

These projections are primarily driven by ballooning spending on computing costs.

Reaching profitability will require tough changes, as revenue from ads and subscriptions becomes necessary for survival. The sector's key challenge in the coming years will be finding a model that balances innovation, costs, and user trust.

The company is also facing increasing competition from Google, which recently released Gemini 3 with strong benchmark results in multimodal reasoning, math, and code, giving it credibility and momentum, as did new data showing Gemini grew to 650 million monthly users in October 2025.

OpenAI responded with an alarm, and CEO Sam Altman reportedly declared a “Code Red” to marshal more resources towards improving ChatGPT.

These, when combined with the increasing costs of securing infrastructure, including chips, data centers, and devices, explain why the company is looking for more profitable sources of revenue.

Search without blue links

For Google, the blue links are becoming a friction point. The company, in Google’s AI Overviews, now synthesizes the "Best Buy" directly into the response.

When users search for "best mountain bikes," the AI doesn't return a list of blogs; it provides a synthesized review of three bikes, with "Direct Buy" buttons powered by Google Merchant Center.

While OpenAI and Google choose an ad-supported path, this is not universal. The differing ad strategies across labs underscore the wider debate: future AI models may be shaped as much by economic need as by technical progress.

Not every lab chooses ads

Microsoft has taken a fundamentally different path by treating AI not as an advertising surface but as infrastructure. It remains one of the quietest players in AI advertising because it does not need that revenue yet.

Its growth is tied to Azure-based inference services and enterprise Copilot licenses. That said, Microsoft has begun quietly injecting AI-powered interactions into the Bing ad stack within Windows, effectively turning the operating system itself into a conversational surface rather than monetizing a standalone assistant.

Meanwhile, Anthropic has positioned itself as the clean alternative. Claude remains deliberately ad-free, reinforcing its structure as a public benefit company.

Backed by Amazon and Google, Anthropic is betting on a trust premium rather than scale. Its target audience is a narrow but lucrative segment of users in legal, medical, and executive roles who are willing to pay higher monthly fees for an AI system that can credibly claim to be insulated from commercial influence.

Meta, however, represents the opposite extreme. Mark Zuckerberg has repeatedly described Meta as AI-first, but the company does not sell its models.

Instead, it uses Llama to strengthen its existing advertising empire. AI functions as a data recycler, improving prediction, reducing targeting costs, and making ads feel indistinguishable from recommendations. For Meta, AI is less a new product than a multiplier for its core business.

Mistral and Perplexity illustrate two divergent approaches on a smaller scale.

Mistral has remained focused on enterprise APIs, prioritizing predictable infrastructure revenue over consumer adoption.

Perplexity, by contrast, struggled to make advertising work. Its early CPM based model in 2025 was criticized for poor attribution, prompting a pivot toward a paid Pro subscription in early 2026.

OpenAI’s pivot to advertising, then, presents itself as a corporate intelligence risk. While it may lose some user trust, advertising revenue will give it the much-needed breathing space to gain a competitive edge.

For Google, the transition to a more personalized experience for its users feels like a natural shift.

When assistance becomes influence

Advertising is not a detour for AI companies. It is the destination their economics have always pointed toward. Once the cost of intelligence becomes continuous and compounding, subsidies stop making sense. Someone has to pay, and history suggests it will not be the user alone.

For Google, this shift is less a reinvention than a return to form. The company is not abandoning search; it is absorbing it. By embedding advertising into AI-mediated actions rather than search results, Google preserves its core advantage while reshaping how influence is exercised. The blue links disappear, but the business model survives in a more intimate form.

For users, the tradeoff is subtle but profound. AI assistants will feel more helpful, more timely, and more personal. Yet the line between assistance and persuasion will grow harder to see. In a world where intent is inferred before it is expressed, the real question is no longer whether AI will show ads, but who decides when help turns into influence.