The AI Invoice Is Coming. Nobody Knows How Big It Will Be.

The AI Invoice Is Coming. Nobody Knows How Big It Will Be.

We're building AI-powered tools that genuinely work. They save time, add value for our staff, our customers, and their end users. That's the good news.

Here's the uncomfortable part. The companies we depend on to power those tools are spending and losing money. A lot of it.

Anthropic lost $5.3 billion in 2024, continued to make a loss in 2025 but are reporting significant revenue growth. That growth, predominantly enterprise-led, poses challenges for the company which recently faced backlash for changes to model performance and changes to inclusions in their Pro and Max tiers.

OpenAI, in 2024, lost $5 billion on $3.7 billion in income. The company is reportedly missing internal user and revenue targets, and are concerned about affordability of their future compute contracts. They’re projecting total losses of $143 billion before profitability arrives around 2030.

Elon Musk's xAI recorded a net loss of $1.46 billion in a single quarter in Q3 2025.

In 2026, Meta plan capital expenditure of $145 billion, Google will spend up to $185 billion, Microsoft projecting $145 billion and Amazon $200 billion. These four hyperscalers have plans to add as much as $2 trillion of AI-related assets to their balance sheets by 2030, facing annual depreciation expenses exceeding their combined profits.

These are not sustainable numbers. Investors know it. At some point, the bill lands on us.

There is currently no realistic path to self-hosting for a company like Vibe. Not with the hardware and compute costs involved, and not with a lack of open-weight models capable of replacing what we use today. Training our own models is not a conversation worth having.

So we're dependent. Structurally, deeply dependent. And the repricing has already started.

GitHub Copilot, which our entire development team uses daily, just announced a move to usage-based billing from June 1, 2026. Every interaction will now consume AI Credits charged on token consumption. The flat-rate simplicity is gone. This is the second pricing recalibration for Copilot in under a year.

GitHub absorbed escalating inference costs for as long as it could. Their position is now explicit: the current model is no longer sustainable.

That's not a criticism. It's a warning. Every provider will reach the same conclusion.

None of this makes AI the wrong bet. It makes it a bet that needs managing.

We build with eyes open. We track usage rigorously. We architect for provider portability where we can. We pressure-test the commercial case for every AI-dependent tool before the costs normalise upward. We treat vendor lock-in as a risk, not just an inconvenience.

Nobody has a clean answer yet. Anyone claiming otherwise is selling something.

But that's also what makes this exciting. We're all navigating the same uncharted territory in real time. At Vibe, our instinct has always been to move fast, stay adaptable, and find the opportunity in the uncertainty. That's not a slogan. It's how we've approached every wave of change before this one.

The AI journey is genuinely brilliant, even when it's complicated. We'll keep building things that matter, watching the economics carefully, and pivoting when we need to. That's the job. And right now, it's an exciting one.