AI model switching hits record 8%, inference market to reach $350B
A few threads worth tracking this week if you’re building in or around AI: provider loyalty is eroding, AI startups are converging on identical infrastructure, and the inference market is on track to dwarf everything that came before it.
AI Provider Switching Is at a Record High
Ramp’s spending data puts the three-month average of firms switching AI model providers at 8% in September, a new record. Anthropic picked up significant share in May last year after launching Opus 4 and Sonnet 4. That momentum faded over the following year until OpenAI introduced price cuts and new model launches in June.
The back-and-forth between the two companies is starting to look like early commodity behavior. When customers move this easily, pricing and availability matter more than brand loyalty.
Every AI Startup Is Building the Same Stack
Angular Ventures draws a direct comparison to web companies in 2005, when every startup needed a database and a login page. Today’s AI startups are converging on the same agent loops, memory systems, connectors, and sandboxes. Sharing that infrastructure didn’t make 2005-era web companies interchangeable, and it may not here either.
The concern the piece raises: founders are pitching basic infrastructure as their competitive advantage while saying little about the specific customer problem they’ve solved. According to Angular Ventures, customer references and the team’s depth of industry knowledge may be more telling than which model or memory system a startup has chosen.
Inference Is the Next $350B Market
AI inference is projected to reach $350 billion by 2027, according to Tom Tunguz, which would put it at double the size of the current database market. That scale brings a structural cost problem: inference sits in cost of goods sold, not operating expenses. Gross margins compress below typical SaaS levels when your product runs on GPU cycles.
Businesses building on top of inference will need to optimize unit economics more aggressively than SaaS founders have historically had to. The upside is that new infrastructure efficiencies tend to create new product categories.
️ Can Agents Build Network Effects?
Andrew Chen asks the obvious question: if your agent can email mine, why do we need to use the same platform? That’s a different dynamic from social networks, where everyone benefits from joining a single shared service.
The more interesting answer in the piece: network effects for agents might come from what they create and share rather than from direct connections. A shared reputation system, for example, could help agents identify trustworthy counterparties. That system could be owned by one assistant or become a standalone service that every agent accesses. Neither outcome is obvious yet.
Other Signals Worth Noting
- Claude for Google Workspace is now in public beta on all paid Claude plans, adding Claude directly inside Google Docs, Sheets, and Slides. Anthropic also released connectors so Claude users can create and edit Google files without leaving Claude.
- Higgsfield Ads Studio generates static ads for every product on a website. Paste a URL, and it builds ads based on what converts in that niche. Users can adjust logos, colors, tone, and selling points before generation. Output works for organic posts and paid campaigns.
- Apple cancellation-screen data from 309 apps and 22 million messages shows subscribers accepted 0.68% of discounts and 0.52% of plan switches, with no measurable effect on cancellation rates in early observational data (via RevenueCat).
- Zappos had roughly five years with little competition because the market doubted customers would buy shoes without trying them on first. The patience required to hold a position others dismiss is the point of the piece.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
