When Your AI Provider Becomes Your Competitor: A Platform Risk Playbook
TL;DR
OpenAI, Anthropic, and Google have moved from selling model access to shipping products. In 2026, Claude Code crossed $2.5 billion in annualized revenue, and OpenAI acquired Astral to fold Python tooling directly into Codex. Every AI startup building on top of these providers faces a version of the same risk: the infrastructure layer you depend on can absorb your value proposition. This article explains the pattern, gives you a framework for assessing your own platform risk exposure, and lays out six strategies for building AI products that survive and even benefit from the competition.
The AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.
The Shift: From API Providers to Product Companies
For most of 2023 and 2024, the working assumption in AI startup land was that model providers would stay at the infrastructure layer. OpenAI would provide the API. Anthropic would provide the API. Google would provide the API. Startups would build the products on top, and everyone would benefit from the arrangement.
That assumption is no longer operational. By mid-2026, the three major frontier providers have each made deliberate moves up the product stack. Anthropic's Claude Code crossed $2.5 billion in annualized revenue, accounting for more than half of all enterprise spending on Anthropic products, directly competing with coding assistant startups that were built on Claude APIs. OpenAI acquired Astral and folded uv, Ruff, and the Astral engineering team directly into Codex, absorbing developer tooling that hundreds of startups were using or competing against. Google's Gemini advanced features are deeply integrated into Workspace, creating a competitive surface for document AI and productivity tools that Google can price at zero for existing enterprise customers.
This is not a bug in the AI ecosystem. It is the same pattern that played out with app stores, cloud computing, and social media platforms. The platform provider improves their position by moving up the stack. The question for AI PMs is not whether this will happen but how to build products that are not destroyed by it when it does.
The Uncomfortable Math
If your product's core value proposition can be described in one sentence, and that sentence sounds like a feature OpenAI or Anthropic would add, your product has platform risk. The more closely your feature set overlaps with what model providers want to offer natively, the shorter your window to build durable differentiation.
How Platform Absorption Happens: The Four Patterns
Provider competition does not always look like a head-on attack. Understanding the four patterns helps you anticipate it before it lands on your roadmap.
Pattern 1: Feature absorption
The provider adds your core feature directly to their model or platform. Your product was a wrapper; theirs is now the same wrapper built in. Example: memory across conversations, web search, code execution. These were all third-party products before becoming standard model capabilities.
Watch for: beta features in the provider's developer preview that overlap with your core use case.
Pattern 2: Acquisition and integration
The provider acquires a company that competes with or complements your product, and folds the technology in. Example: OpenAI acquiring Astral to strengthen Codex's developer tooling story. You were not the target, but you became the collateral damage.
Watch for: providers acquiring teams in your space even if the specific company was not your direct competitor.
Pattern 3: Enterprise bundle pricing
The provider includes AI functionality similar to yours in their enterprise tier at no additional cost. Your product now has to compete against 'free' for the customer who is already paying for the underlying platform. Example: Google Workspace's AI features bundled into Business and Enterprise plans.
Watch for: provider pricing announcements that mention features in the same breath as features you charge for.
Pattern 4: API deprecation with native replacement
The provider deprecates the API capability your product depended on and replaces it with a first-party product that has the same interface. Your integration still works, but the provider now owns the customer relationship for that feature.
Watch for: API deprecation notices combined with announcements of new first-party products in the same domain.
Six Strategies for Building Provider-Resistant Products
There is no strategy that fully eliminates platform risk. The goal is to build products where the provider's entry into your space is not fatal, and in the best case is actually a tailwind that validates the category and accelerates growth for defensible players.
Strategy 1: Own the proprietary data layer
The model providers can copy your feature. They cannot copy your customers' data. Build products that ingest, learn from, and improve on proprietary customer data. The value then lives in the trained artifacts and feedback loops, not the model capability itself. Providers entering the space start from zero on customer data; you have a head start they cannot buy.
Strategy 2: Own the workflow, not just the AI feature
If your product is an AI feature, you are exposed. If your product is a workflow that happens to use AI, you are much more resilient. The workflow includes the UI, the integrations, the reporting, the approval flows, and the institutional memory of how teams have adapted it over time. A provider can ship a similar AI feature in a model update; they cannot easily replicate three years of workflow customization.
Strategy 3: Go deep vertical
Horizontal AI tools (general writing, general coding assistance, general summarization) are where providers compete most aggressively, because the returns are highest. Vertical AI (legal contract analysis for a specific firm structure, clinical note generation for a specific EHR, underwriting assistance for a specific insurance product type) is where your domain expertise creates a moat providers cannot replicate quickly. The more niche your vertical, the less likely a frontier provider sees it as worth owning.
Strategy 4: Build multi-provider resilience
Do not make your product technically or commercially dependent on a single provider. Abstracting your model calls through a routing layer means any provider entering your space cannot use API deprecation as a lever against you. Multi-provider architecture also gives you negotiating leverage on pricing and lets you adopt new models quickly as the market evolves.
Strategy 5: Build switching costs through integration depth
The harder your product is to rip out, the safer you are. Deep integrations with adjacent systems (CRM, ERP, ticketing, HR), custom trained models on customer data, and complex permission and approval workflows all create switching costs that a provider's native feature does not inherit. Build integrations early and broadly.
Strategy 6: Out-execute in the niche
Providers optimize for the broad case. They cannot be excellent at your specific use case and also serve every other use case well. Your advantage is focus: you can build a dramatically better experience for your specific customer segment than any general-purpose platform. Ship at a cadence that keeps you ahead of what a provider can copy, and build user trust that is hard to transfer to a substitute product.
Build a Defensible AI Product Strategy
The AI PM Masterclass covers competitive moats, platform risk, and the strategic frameworks that help AI product teams build for long-term survival, not just quick adoption. Taught live by a Salesforce Sr. Director PM.
The Multi-Provider Playbook in Practice
Multi-provider architecture is the single most controllable action a PM can take to reduce platform risk today. It does not eliminate the risk, but it removes the single-provider leverage that makes platform absorption most damaging. Here is what it looks like in practice.
Abstract your model calls
Route requests through a provider-agnostic layer (your own or a third-party router like LiteLLM, PortKey, or NeMo Switchyard) rather than calling provider APIs directly. When a provider changes their pricing or enters your space, you can reroute traffic without rewriting your product code.
Maintain eval coverage for multiple models
Keep evaluations running against at least two providers at all times. When a provider enters your space or raises prices, you should be able to switch within a sprint, not a quarter.
Negotiate based on alternatives
Having a working alternative forces every provider to compete for your volume. 'We are currently running 40% of our inference with your competitor' is a credible negotiating position. Single-provider dependence eliminates that leverage entirely.
Use open weights as an escape valve
MIT-licensed models like DeepSeek V4-Pro and Qwen 3.8 Max give you a self-hosting option if a provider raises prices beyond your unit economics or deprecates a feature. The option does not need to be exercised to be valuable in negotiations.
Reading the Signals: What to Watch Before It Happens to You
Platform risk events rarely come without advance signals. The challenge is that providers give you 6 to 18 months of warning if you know what to look for, but most PMs are not actively tracking the right signals.
Provider developer blog posts about use cases
When a provider publishes a detailed guide on how to use their model for a use case that happens to be your product's core function, that is not just marketing. It is often a preview of a native feature coming in 6 to 12 months. Read provider technical blogs with that lens.
Beta features and playground experiments
Providers frequently test future product directions through playground features and beta programs. Early access to a playground feature in your domain is a signal that the feature is being productized. Join every relevant beta program your providers offer.
Acquisition patterns in adjacent categories
When a provider acquires two or three small companies in a domain, a native product in that domain usually follows. Track provider acquisitions even when the specific target does not compete with you directly.
Pricing changes that make specific use cases more attractive
When a provider cuts prices significantly for a use case that is close to your product, they are often pricing to grow usage in advance of a product announcement in that space. Price cuts are not always competitive benevolence.
Conference keynote themes
The demos at OpenAI Dev Day, Anthropic's research showcases, and Google I/O are not random. They show where each provider is directing product energy. Watch for use cases being demoed that overlap with your product.
Platform Risk Assessment: How to Score Your Own Exposure
Run this assessment for your product quarterly. Score 1 to 5 on each dimension, where 5 is high risk. A total score above 20 means platform risk should be on your strategic roadmap now, not when it becomes urgent.
Provider overlap
How closely does your core feature match what your primary provider has already announced or shipped in their product? 1 = no overlap. 5 = your core feature is already a beta in their platform.
Provider dependency
What percentage of your technical stack depends on a single provider's API? 1 = fully multi-provider. 5 = all production traffic goes through one provider with no fallback.
Data defensibility
How much proprietary customer data does your product accumulate that a provider could not replicate? 1 = extensive proprietary data creates strong moat. 5 = pure model API wrapper with no proprietary data layer.
Workflow depth
How deeply embedded is your product in the customer's operational workflow? 1 = core workflow dependency with high switching cost. 5 = standalone feature that customers could replace with a provider's native tool in a day.
Vertical specificity
How niche is the vertical your product serves? 1 = extremely specific niche with deep domain expertise required. 5 = horizontal use case that applies to any industry and any company size.
Build AI Products That Outlast the Platform Wars
The AI PM Masterclass covers competitive strategy, platform risk, and the product decisions that determine which AI companies survive consolidation. Live cohorts starting September 2026.
Related Articles
Before you go: get the AI PM Minute
One tactic to make you a sharper AI PM, twice a week. 60 seconds to read. Free.
No fluff. Unsubscribe anytime.