OpenAI just made a bold move against their own API business. Last week, they dropped GPT-5 at $10 per million tokensnearly 10x cheaper than Anthropic’s flagship model while matching performance benchmarks. This isn’t just about winning back market share. It’s about reshaping the inference market and forcing the industry toward a completely different battlefield: the $334 billion advertising empire.
By mid-2025, Anthropic had pulled off something remarkable. They captured 32% of the enterprise LLM API market from OpenAI, up from just 12% eighteen months earlier. More importantly, they made inference actually profitableut “path to profitability” or “unit economics positive,” but real profit according to their CEO. While OpenAI was dealing with consumer drama and inconsistent model releases, Anthropic focused on the boring stuff: enterprise reliability, superior coding capabilities, and consistent pricing.
OpenAI’s response was aggressive. They didn’t just compete on priceut they significantly undercut the market pricing.
The Commoditization Playbook: Why OpenAI Attacked Inference Margins
To understand OpenAI’s move, you need to understand a fundamental tech strategy that Joel Spolsky wrote about in 2002: commoditize your complements. Pick one thing that gives your business fat profit margins, then make everything consumed alongside that main thing as cheap as possible erly free.
Disney makes margins on in-park purchases, so they historically made hotels, parking, and flights cheaper. Casinos profit from gambling, so they subsidize hotel rooms, food, and drinks. Google makes money on ads served during searches, so they built Chrome and gave it away for free. Every free Chrome download became another funnel into Google’s $207 billion advertising machine.
Facebook took this further. They realized expensive phones and WiFi meant fewer Facebook users, so they literally handed Samsung bags of cash to preinstall their app and subsidized phone distribution across Africa with WiFi. Microsoft did it first with PCs and MS-DOS they wanted to sell software, so they made hardware a commodity. They didn’t build PCs; they made PCs so cheap that Compaq and Dell fought each other to the death while Microsoft collected Windows licenses from every corpse.
OpenAI appears to be applying elements of this playbook to AI inference. They’ve chosen their profit center onsumer subscriptions and potentially advertisingnd they’re aggressively pricing inference to gain market position.
The Four Kingdoms of Tech Profit
There are only four places where tech companies make real money:
The four major profit pools in tech, measured in billions
- Advertising: $334 billion. Google Ads alone does $207 billion. Meta adds another $127 billion. Together they make more profit than most countries’ GDP.
- Cloud: $193 billion. AWS ($67B), Azure ($97B), GCP ($29B). They sell boring infrastructure at criminal margins.
- Systems of Record: $132 billion. Salesforce, Oracle, SAP. The databases that run the world and nobody understands.
- Productivity: $132 billion. Microsoft Office, Adobe Creative Suite. The stuff you hate using but can’t live without.
You pick one kingdom. You defend it with your life. You price aggressively in other areas to defend it.
How Anthropic Built a Strong Inference Business
AI inference was positioned to fit into the cloud computing profit bucket. It’s a value-added servicend you take commodity GPUs, add your proprietary models, wrap it in enterprise features, and charge a markup for peace of mind. Same playbook as Snowflake adding SQL on top of EC2, or Dropbox adding sync on top of S3.
Think about what AWS actually is. Amazon takes commodity hardware you could buy yourself, wraps it in APIs, and charges a 70% markup for not having to think about it. EC2 instances are just servers. S3 is just hard drives. But Bezos figured out enterprises would pay substantial margins for someone else to manage the complexity.
Anthropic was making this work. They dominated coding and had a consistently better model for almost two years. Claude 3.5 was the breakthrough that spawned an entire ecosystem ursor and Windsurf for code editors, Lovable and Bolt for app builders. Superior function calling that actually worked in production. Five-nines uptime through Bedrock and Vertex partnerships. No consumer drama.
The coding dominance was significant. Every major coding tool ursor, Windsurf, Lovable, Bolt, Replitll runs on Claude. Anthropic owns 42% of the coding assistant market, twice their nearest competitor. This gave them pricing power and forced their ecosystem to get used to their price point.
The customers followed: Bridgewater analysts, GitLab engineers, financial firms deploying through Deloitte and PwC. Anthropic kept prices flat at $75/million for Opus and $15/million for Sonnet across versions 3.0, 3.5, 3.7, and 4.0, while secretly making their models 50% more efficient behind the scenes. As Dario recently told Alex Kantrowitz: “For every dollar the model makes, it costs a certain amount. That is actually already fairly profitable.”
Translation: despite flat pricing and winning market share, they turned inference into a profit center. The coding capability edge meant the entire surge in AI coding happened on Anthropic’s infrastructure.
OpenAI Strikes Back: Targeting Different Markets
OpenAI’s response was to aggressively undercut the market and focus on different areas. They were never just an API company. $3.6 billion in consumer subscriptions versus $1.4 billion in API revenue. They make more money from consumers paying $20/month than from enterprises paying millions.
The $334 billion advertising market remains largely untapped for AI companies. Google and Meta need cheap AI inference toond they could subsidize inference extensively to protect their $207 billion and $127 billion advertising empires. Every dollar of potential ad profit creates pressure to keep AI costs low.
OpenAI’s hiring decisions reflect where they’re going. In December they hired Kate Rouch as their first CMOnd the same person who ran ads at Meta for 11 years and put ads in Instagram’s feed. In May they brought in Fidji Simo, Instacart’s CEO, as head of applications. These hires suggest a focus on consumer engagement and potential advertising opportunities.
OpenAI doesn’t rely solely on API revenue. They need to become a default place people go to think, to search, to create. So they priced aggressively in the API market.
The $10 Token Market Impact
GPT-5 at $10 per million tokensnd represents aggressive competitive pricing. Several application layer companies that were previously working with Anthropic’s pricing switched to test OpenAI’s offering. Cursor, Lovable, Bolt, Devin, Windsurf were all launch partners with OpenAI’s announcement.
OpenAI could have maintained higher margins in the API space while pursuing other opportunities. But they chose to bring costs significantly below the market rate Anthropic had established, signaling their focus on volume and market share over API margins.
The move mirrors historical tech competition. When Microsoft wanted to sell software, they made hardware competitive. When Google wanted ad revenue, they made browsers free. When Facebook wanted users, they made phones cheaper. OpenAI wants consumer dominance and potential advertising opportunities, so they’re making AI inference more affordable.
Anthropic’s Strong Position Despite Challenges
Anthropic remains a significant player trying to capture margins on training and providing models, though they face new competitive pressure. Other companies went open source like Meta’s Llama or aligned with advertising like Google and OpenAI.
The neo-cloudsnd Together, Modal, Replicatere interested in different metrics. They run 10-30% markups on GPU resale. They benefit from higher volume regardless of which provider succeeds. Higher volume from lower prices can actually help their business model.
Even AWS, their partner, treats inference as part of their broader $67 billion infrastructure ecosystem strategy.
Anthropic has strong cards to play. Dario mentioned that reinforcement learning, interpretability, and fine-tuning are all areas where they can charge substantial margins. Their enterprise contracts are sticky. Enterprises often aren’t as cost-sensitive as consumers when it comes to critical infrastructure. They’ll pay for reliability, compliance, and features that matter to Fortune 500s.
They could double down on application layer margins too. Claude Code is developing features similar to Cursor or Devin. Claude Desktop is making artifacts hostable, potentially competing with Lovable and Bolt. They could enter the systems of record or productivity realms. Move up the stack to where the real $132 billion profit pools live.
API margins remain a real business, especially for specialized use cases and enterprise customers who value reliability and specific capabilities over raw cost savings.
The Market Dynamics Shift
This competitive dynamic shows different strategic approaches. OpenAI can afford to price aggressively in APIs because they have diversified revenue streams and different long-term goals. Anthropic has built a strong position in enterprise inference and continues to be profitable in this space.
We’re watching these AGI labs pursue different strategies. OpenAI is willing to sacrifice API margins for market share and consumer focus. Anthropic continues building a sustainable inference-focused business model while exploring expansion opportunities.
The market has shifted. OpenAI chose consumer subscriptions and potential advertising opportunitiesnd the $334 billion kingdom. Anthropic chose infrastructure specializationuilding premium services that command higher margins. OpenAI’s aggressive pricing was a competitive move targeting Anthropic’s market position, but also reflects their broader strategic focus on consumer markets rather than enterprise infrastructure margins.
OpenAI made their move. The API market is now more competitive, but there’s still room for different approaches. The question is how each company executes their distinct strategies in this new competitive environment.
The Stakes for Enterprise AI
This isn’t just a battle of pricing; it’s a competition for different approaches to enterprise AI. Anthropic’s success was built on the premise that enterprises value stability, performance, and specific capabilities like superior function calling for coding enough to pay a premium. Their focus on “boring stuff” like five-nines uptime through Bedrock and Vertex partnerships appeals to large organizations that prioritize reliability.
For companies like Bridgewater and GitLab, the cost per token is a factor, but not the only one. The ability to trust a model for critical operations, ensuring data privacy, and having consistent performance in production environments often outweighs cost considerations. This is where Anthropic maintains competitive advantages. Their “Constitutional AI” approach resonates with businesses concerned about AI safety and interpretability.
The question for enterprise customers now becomes: do you chase the lowest price, or do you stick with the provider that has demonstrated consistent reliability and deep understanding of enterprise needs? OpenAI’s aggressive pricing attracts some customers, especially those building less critical applications or operating on tighter margins. But for core business functions, switching costs and reliability concerns may keep many with Anthropic.
This dynamic parallels the broader cloud market. While commodity cloud services compete on price, specialized, managed services often command higher margins because they abstract away complexity and guarantee performance. Anthropic positioned itself as a specialized inference provider, wrapping commodity GPUs with proprietary models and enterprise-grade services. OpenAI is making the underlying service much cheaper, but the value of the specialized wrapper may persist for many enterprise use cases.
The Broader Impact on the AI Ecosystem
OpenAI’s move sends ripples throughout the ecosystem. The neo-clouds ompanies like Together, Modal, and Replicatere operate on thin margins, often reselling GPU access with value-adds. They thrive on volume. OpenAI making tokens cheaper drives more usage, which benefits their business model. They don’t need high API margins; they want throughput.
For the open-source community, if proprietary models become very affordable on the API side, it increases pressure on open-source projects to differentiate on factors like privacy, customization, and local deployment rather than cost. The performance gap between open-source and frontier models might remain, as I’ve noted before, but the economic dynamics become more complex.
For application developers building on top of these models, the immediate impact is positive: lower costs. The ability to build AI-powered applications at reduced prices opens up possibilities for innovation and market entry. Companies like Cursor, Lovable, Bolt, Devin, and Windsurf can offer their services more affordably or allocate more budget to other features. This benefits the developer ecosystem overall.
The shift also forces a re-evaluation of where value is captured in the AI stack. If basic model access becomes very affordable, value moves to the data, the fine-tuning, the application layer, or integration with proprietary systems. This represents a potential path forward for companies focused on AI infrastructure: specializing in areas beyond basic token provision.
Comparison of token pricing before and after GPT-5’s market entry.
Anthropic’s Path Forward: Specialization and Value-Add Services
Dario Amodei’s comments about interpretability, reinforcement learning, and fine-tuning represent areas where Anthropic can build defensible, high-margin businesses. Enterprises, particularly in regulated industries, will pay premiums for models they can understand, explain, and control. The black-box nature of many frontier models remains a significant hurdle for adoption in critical sectors.
Furthermore, Anthropic’s potential to move up the stack into application-layer products represents a strong strategic option. If Claude Code develops into a fully-featured coding assistant or if Claude Desktop makes artifacts hostable competing with existing tools, they can capture value at higher levels than basic token provision. This vertical integration allows them to own more of the customer experience and build stickier relationships.
Entering the systems of record or productivity realms, while ambitious, could also offer access to significant profit pools. An AI agent that deeply understands a company’s SAP or Salesforce data, providing insights and automation impossible with generic LLMs, requires more than just cheap inference. It requires deep domain expertise, robust integrations, and consistent reliability. This represents a long-term opportunity that could secure Anthropic’s position beyond current market dynamics.
The competitive dynamics have shifted. OpenAI chose consumer subscriptions and potential advertising opportunitiesnd the $334 billion kingdom. Anthropic chose infrastructure specializationuilding premium services that command higher margins. OpenAI’s aggressive pricing represents a competitive challenge to Anthropic’s market position, but also reflects their broader strategic focus on consumer markets rather than enterprise infrastructure profits.
OpenAI made their move. The API market is now more competitive, but remains a viable business with room for different approaches. The question is how each company executes their distinct strategies in this new competitive environment.