Chinese AI Models Are Increasingly Displacing OpenAI and Anthropic

July 7, 2026 | Authors: Alex Gove & Connie Loizos

---

1. Key Themes

Theme 1: Chinese AI Models Capturing Real Enterprise Token Share

U.S. companies are increasingly routing workloads to cheaper Chinese models, and the data is measurable and material.

"U.S. companies are increasingly routing AI workloads to cheaper Chinese models from DeepSeek, Z.ai, and Alibaba as OpenAI and Anthropic costs rise, with Chinese models now accounting for more than 30% of U.S. company token usage on OpenRouter in recent weeks."

"DeepSeek V4 Flash is the main winner on overall usage, processing 5.3 trillion tokens weekly. The most popular frontier model, Opus 4.8, handles just over 2 trillion."

---

Theme 2: Frontier Labs Are Defending Share With Credits, Not Just Product

The competitive response from OpenAI and Anthropic is promotional spend, not pure product differentiation — a signal of genuine pricing vulnerability.

"OpenAI and Anthropic are showering startups with free AI credits and discounts, in some cases worth millions of dollars, as they race to lock young companies into their models and keep cheaper open-weight rivals — including Chinese models — from gaining share."

---

Theme 3: Open Source and Frontier AI Are Complementary, Not Competing — For Now

Decagon CEO Jesse Zhang argues the AI market is not a zero-sum substitution story between open-source and frontier models.

"In Zhang's telling, they aren't competitors, and open source models' success isn't coming at the expense of frontier labs. Instead, they're two phases of the same life cycle, with expensive frontier models being used to prove out use cases that can be passed along to cheaper open source alternatives as they mature."

"As more mature use cases switch to lighter models, new use cases keep arising — and the overall spend on frontier models barely goes down."

---

Theme 4: Beijing May Weaponize Access to Chinese AI as Geopolitical Leverage

The same Chinese models gaining U.S. enterprise share could become restricted as a matter of national security policy.

"Chinese officials have held meetings with Alibaba, ByteDance, Z.ai, and other top tech companies about potentially restricting overseas access to China's most advanced AI models, as Beijing moves to treat homegrown AI as a national-security asset."

---

Theme 5: Deep-Tech Capital Formation Accelerating — Fusion, Quantum, Nuclear

Three high-conviction bets in hard science attracted enormous rounds, signaling that patient capital is flowing into historically long-horizon technologies.

Proxima Fusion raised $469.8 million at a $2.7 billion valuation for stellarator-based fusion power plants. Oratomic raised $300 million Series A for quantum computing targeting Shor's algorithm. Standard Nuclear is seeking up to $383 million in its IPO at a $3.55 billion valuation for advanced nuclear fuel for SMRs.

---

2. Contrarian Perspectives

Perspective 1: Open Source AI Isn't Killing Frontier Labs' Revenue — It's Actually Enabling Their Growth

The consensus fear is that cheap open-source and Chinese models will cannibalize frontier lab revenue. The data suggests otherwise: spend on expensive models holds even as volume migrates.

"If you scroll down to overall token spend, you'll see Anthropic still accounts for more than half of the overall AI spend on the platform. Given that much of the recent change comes from Anthropic's own rising prices, the share has dropped slightly over the past month, but not significantly."

The average token cost for Anthropic's Opus 4.8 is roughly 23x higher than DeepSeek V4 Flash ($1.37 per million tokens vs. $0.06), "which would mean Opus was still probably capturing the lion's share of spending."

The implication: volume share and revenue share are decoupling. Chinese models win on tokens; frontier labs still win on dollars.

---

Perspective 2: The Chinese AI Threat May Be Self-Limiting

The very models displacing OpenAI and Anthropic in U.S. enterprise could disappear from the market by government fiat — undermining the thesis that they represent a durable competitive threat.

"Chinese officials have held meetings with Alibaba, ByteDance, Z.ai, and other top tech companies about potentially restricting overseas access to China's most advanced AI models, as Beijing moves to treat homegrown AI as a national-security asset."

This creates a paradox: U.S. enterprises building cost structures around Chinese models may be building on an unstable foundation.

---

Perspective 3: Social Media Age Regulation Is Largely Theater

Australia's much-publicized under-16 social media ban is a case study in regulation that generates headlines but not outcomes.

"Australia's under-16 social media ban is failing at the first step, with researchers finding that Instagram, Snapchat, TikTok, YouTube, and other platforms did not seek age proof from any of 50 test accounts that claimed to be 16."

---

3. Companies Identified

---

4. People Identified

---

5. Operating Insights

Insight 1: Use Frontier Models to Prove Use Cases, Then Migrate to Cheaper Alternatives

Operators can sequence their AI spending strategically: validate with frontier models, then systematically move proven, mature workloads to cheaper open-source or Chinese alternatives.

"Expensive frontier models [are] being used to prove out use cases that can be passed along to cheaper open source alternatives as they mature."

This is already standard practice at AI-native companies: "More mature AI deployments are switching to lighter models, he says, even at his own company."

---

Insight 2: Negotiate Aggressively for AI Credits Before Committing to a Provider

The competitive pressure on frontier labs has created real leverage for startups at the vendor selection stage.

"OpenAI and Anthropic are showering startups with free AI credits and discounts, in some cases worth millions of dollars, as they race to lock young companies into their models."

Startups should treat AI provider selection as a negotiation, not a default — especially at early stages when labs are most motivated to lock in customers.

---

Insight 3: Don't Build Critical Infrastructure on Chinese AI Models Without a Contingency

The cost arbitrage is real, but the supply risk is material and government-driven.

"Chinese officials have held meetings with Alibaba, ByteDance, Z.ai, and other top tech companies about potentially restricting overseas access to China's most advanced AI models."

Operators routing significant workloads through Chinese models should maintain provider redundancy and avoid hard dependencies on models that could be restricted by Beijing.

---

6. Overlooked Insights

Insight 1: Quantum Computing Is Entering "Serious Capital" Territory — Right Now

Oratomic, a company founded this year, raised $300 million in a Series A to run Shor's algorithm at practical scale. Bezos Expeditions, Index, General Catalyst, and Khosla all participated. This is not exploratory funding — this is a bet that cryptographically relevant quantum computing has a near-term path. The implications for cybersecurity, encryption, and financial infrastructure are profound and underreported in mainstream AI coverage.

"Oratomic…develops quantum computing technology aimed at running Shor's algorithm for practical large-scale computation, raised a $300 million Series A round led by Khosla Ventures, with Bezos Expeditions, Index Ventures, General Catalyst…also stepping up."

---

Insight 2: Keyfactor's $1B+ Round Signals That Post-Quantum Cryptography Is an Investable Category Now

A 25-year-old, unglamorous certificate management company just raised over $1 billion. The timing — concurrent with quantum computing advances and AI-driven identity threats — suggests institutional capital is quietly positioning in cryptographic infrastructure as a defensive play, well ahead of public attention.

"Keyfactor…helps enterprises manage machine identities, digital certificates, and cryptographic assets across cloud, on-premises, and hybrid systems, raised a $1+ billion round led by Summit Partners."