The AI infrastructure bubble question — and why your software should stay portable
Together AI raised $800M and Blackstone pledged $30B for AI data centers, reigniting the bubble debate. Here's the practical move for a small business: build systems you can move.
There's a loud debate running through AI right now: is the money pouring into infrastructure meeting real demand, or inflating a bubble? You don't have to pick a side to protect yourself from the answer.
The numbers driving the debate
The buildout is staggering. On July 1, Together AI raised $800M at an $8.3B valuation — more than double its value 16 months earlier — led by Aramco Ventures, with annual bookings crossing $1.15B. Days earlier, Blackstone pledged $30B toward AI data centers in Japan over three to five years, eyeing sites above a gigawatt. Blackstone's COO put the bull case bluntly: the risk of too little compute outweighs bubble fears.
The bear case is quieter but sharper. Models keep getting cheaper to run as efficiency improves — and if inference costs fall faster than demand rises, a chunk of today's spending math gets rewritten. So does the pricing you're quoted.
Why it matters for your business
Here's the translation for a company under 500 people: you feel this through your vendor's invoice. If inference costs collapse, prices should fall — but only if you're free to move. If your operation is welded to one provider's proprietary API, you inherit their pricing, their outages, and their roadmap, whether or not it still serves you.
The defense isn't guessing the market. It's architecture. When we build automations and internal tools, we keep the expensive, swappable parts — the AI model, the compute — behind a clean boundary. Your business logic lives in your infrastructure, not rented inside someone else's platform. If a cheaper or better model shows up next quarter, you switch a config, not your whole stack. That's the difference between owning the shape of your software and paying rent on it forever.
Moves that hold up regardless of which side of the debate wins:
- Negotiate 12-month terms with renewal caps — don't lock in multi-year at today's prices.
- Keep workloads portable so you can switch providers if inference gets cheaper.
- Treat current vendor pricing as temporary, in both directions.
Key takeaways
- Massive AI infrastructure spending (Together AI $800M, Blackstone $30B) has reignited the bubble debate
- If efficiency keeps driving inference costs down, today's pricing could reset fast
- Small businesses feel all of it through vendor invoices — portability is the hedge
- Build with the model behind a swappable boundary so you own the system, not rent it
Want automations you can actually move when the market shifts? We build vendor-agnostic systems you own. See how we build or send us the problem.
Sources: TechCrunch, Nikkei Asia.
- #ai-infrastructure
- #vendor-lock-in
- #automation
- #small-business
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Anthropic passed OpenAI on revenue — and enterprise coding agents are why
Anthropic hit a reported ~$47B revenue run-rate on the back of enterprise Claude Code adoption. Here's what the shift to AI coding agents actually means for a small business that needs custom software built.
Read it7 automations that pay for themselves in a month (with the actual math)
No moonshots — seven boring, high-ROI automations we build for small businesses, with honest numbers on what each one saves.
Read it