OpenAI ChatGPT to test ads as cash burn tops billions

OpenAI will begin testing advertisements inside ChatGPT as the company looks to add new revenue amid multibillion-dollar cash burn. The move, disclosed on a recent Last Week in AI podcast, signals a shift in how the company might monetize the consumer-facing chat product while it continues to invest heavily in model development.

The pilot will place promoted content inside ChatGPT's free user interface, the podcast said. Details on the scope, targeting, and partners remain limited, but the trial is framed as a way to reduce operating losses without immediately gating core capabilities behind paywalls.

OpenAI and ChatGPT ads

The advertising pilot follows months of investor and market pressure as OpenAI balances rapid infrastructure spending with plans for long-term profitability. Sources discussing the test suggested the experiment will be limited in scale at first, focusing on nonintrusive placements and likely excluding contextual prompts tied to health or other sensitive categories.

Ad testing joins other monetization experiments under consideration, including expanded subscription tiers, enterprise contracts, and new API pricing. Industry observers say an ad layer could preserve free consumer access while extracting revenue from advertising budgets, but the approach raises content-safety and user-experience tradeoffs that OpenAI will need to manage.

Regulatory scrutiny is another risk. Displaying ads inside a conversational AI product creates questions about disclosure, personalization, and data use that differ from both search and social platforms. Any targeting or ranking of promoted content inside ChatGPT will draw attention from privacy regulators and consumer advocates.

Thinking Machines drama and STEM research

The Last Week in AI episode also covered governance turmoil at Thinking Machines, describing a boardroom dispute that triggered leadership changes and a pause in some hiring and fundraising activities. Public filings and company statements remain limited, but the podcast outlined a sequence of personnel shifts and investor tensions that underscore how fragile startups can be when rapid growth meets governance gaps.

On the research front, the podcast highlighted a STEM paper that proposes scaling transformers using modular embedding components. The technique attaches small learned embedding modules that can be swapped or added to extend capacity and specialize parts of a model without retraining a monolithic network. Early results described in the discussion claim efficiency gains on certain downstream tasks by localizing updates to compact modules rather than the entire parameter set.

Researchers interviewed in the segment argue that embedding modules could reduce compute and memory costs for continual learning and domain adaptation. The approach remains early stage and will need independent replication and benchmarks on standard workloads before claims of practical benefit can be validated.

Why it matters

OpenAI testing ads in ChatGPT marks a pragmatic turn toward more aggressive monetization as cash burn pressures mount, which could reshape user experience and regulatory scrutiny for consumer AI. The Thinking Machines dispute and the embedding-modules research both highlight growing pains in the industry: governance risks at fast-moving startups and technical bets that aim to make large models more modular and cost effective.