AI Infrastructure Costs: The New Gravity in Software Valuations
AI infrastructure costs are the ongoing compute, storage, and model-servicing expenses required to run AI features at scale inside software products, and they are now significant enough to reshape startup valuations, investor expectations, and the difference between sustainable unit economics and cash-burning growth stories across the software industry. High-growth AI software companies are learning that the real test is not how fast they can add AI features, but whether they can afford to keep them running once users show up in force. This is the key takeaway behind today’s valuation corrections: AI is not a free growth multiplier, it is an expensive operational commitment that markets are beginning to price with far less patience.
Lovable: AI-Native Growth That Investors Still Reward
Lovable’s trajectory shows what the market now rewards: AI-native software that can turn heavy infrastructure into convincing revenue momentum. The company has reached a reported annualized run rate of $500 million, and investors responded by backing a new Series C that set its valuation at $13.3 billion. That confidence is not blind. Lovable says it now hosts 60 million projects that attract 900 million monthly visitors, a level of demand that forces serious back-end sophistication. The startup offers both its own in-house trained AI model and access to frontier models, and recently signed a multiyear deal with a major cloud provider that increases its usage fivefold. In other words, Lovable is embracing high AI infrastructure costs but pairing them with growth investors can underwrite, rather than hoping those costs will stay invisible.
Canva: When AI Operational Expenses Trigger a Valuation Correction
Canva’s recent valuation markdown is a warning shot for every software company racing to bolt AI onto an existing product. Its largest institutional backers have cut the design platform’s valuation by around $10 billion from a previous high near $42 billion toward the low-$30 billion range. The writedowns followed Canva’s decision to trim its 2026 revenue growth forecast from 30% to 20%, after discovering that the AI features it aggressively rolled out were far more expensive to operate than management had projected. CEO Melanie Perkins told investors that the company had been “relying too heavily on frontier models” from third-party providers. Some AI features were reportedly six times more expensive to serve than what users pay for them. This is a classic software valuation correction driven not by weak demand, but by AI operational expenses that outpaced pricing power and forced a reset in profitability expectations.
Rebuilding Economics: Slowing AI Rollouts to Protect Profitability
Canva’s response shows how quickly strategy must change once AI inference costs threaten the business model. Rather than continuing to burn cash on unsustainable unit economics, the company chose to slow the rollout of AI features while rebuilding its underlying architecture. Perkins wrote that “rather than broadly rolling out a product before the underlying economics were ready, we decided to slow the rollout while we rebuilt the architecture, reduced unit costs and strengthened the business model”. Behind that choice is a hard lesson: AI operational expenses are forcing software companies to recalibrate both profitability expectations and investor confidence. Canva now reports a roughly 90% reduction in AI servicing costs by developing proprietary in-house models and acquiring AI startups such as Leonardo.AI. Its video model runs 17 times cheaper than comparable frontier models, and its image model costs 30 times less. The message is blunt—own your AI stack, or your margins will own you.
What This Means for Startup Funding Trends and Valuations
These stories signal a shift in startup funding trends: investors now separate AI-native companies with credible unit economics from those still drowning in compute costs. The broader lesson reaches every SaaS business integrating AI features today. If a company with billions in revenue, long-term profitability, and large cash reserves discovers that AI inference costs can outrun its pricing power, then every smaller software company running the same playbook with far fewer reserves should take notice immediately. Markets are no longer impressed by AI feature checklists alone; they want proof that AI infrastructure costs can be contained, priced, and supported at scale. Lovable’s rising valuation shows enthusiasm for AI-native models that align infrastructure commitments with growth. Canva’s valuation correction shows how quickly confidence can be cut when AI operational expenses force growth forecasts down and IPO timelines to slip into later years.
