OpenAI spending hit $34 bln in 2025 ahead of planned IPO- FT

OpenAI reportedly spent $34 billion in 2025, including about $19 billion on R&D and nearly $6 billion on sales and marketing, while revenue was only about $13 billion and net loss reached roughly $39 billion. The figures underscore persistent cash burn and margin pressure as the company prepares for a planned IPO this year. OpenAI is also said to be cutting pricing and shelving side projects, including Sora, to rein in costs.

Analysis

The key read-through is not just that OpenAI is spending aggressively, but that its current unit economics still require either a dramatic demand acceleration or a pricing reset to avoid permanent margin compression. If a leader with unusually strong distribution is already forced to cut prices and defer optional projects, the implication is that the AI layer is entering a classic scale-vs-profitability phase: compute suppliers keep the near-term revenue, while model/API vendors face faster commoditization than consensus assumes. That tends to favor picks-and-shovels businesses with pricing power, not the application layer where customer acquisition costs and inference bills both move against them.

Second-order effects matter here: a more aggressive price war would pressure smaller frontier-model competitors and private AI startups that rely on venture funding to subsidize growth. That likely pulls forward consolidation across model providers and raises the probability that enterprise buyers delay commitments, waiting for a lower-cost regime. Over the next 3-6 months, the bigger risk is not a single company’s losses, but a broad reassessment of AI revenue durability across late-stage private markets and IPO candidates.

The contrarian view is that this spending is not necessarily a warning sign if it is still expanding market share into a category with winner-take-most characteristics. The market may be overestimating near-term dilution and underestimating the strategic value of locking in user behavior before competitors catch up. Still, if pricing is cut faster than usage grows, the trade shifts from growth-at-any-price to a margin reset, which is typically when multiples compress first in the private market and only later in public comps.

Key Decisions for Investors

• Go long AI infrastructure beneficiaries vs. model/application names over 1-3 months: favor semis/networking/colo exposure (e.g., NVDA, AVGO, ANET) versus private-market AI beta where funding is more dependent on future profitability; thesis is that capex continues even if software pricing weakens.
• Short a basket of high-multiple AI software/adjacent names or use put spreads over the next 4-8 weeks; the risk/reward improves if the market starts marking down implied long-run AI take rates after another round of price cuts.
• For private-market exposure, reduce or hedge late-stage venture-style AI positions with the weakest gross margin visibility; use a barbell of cash-generative infrastructure and limited exposure to frontier-model economics.
• Consider a pair trade long compute suppliers / short AI-enabled SaaS names that are most exposed to inference-cost pass-through pressure; target 10-15% relative outperformance if pricing competition persists into the next earnings cycle.