In a sweeping analysis that has rippled through the tech investment community, Gavin Baker — CIO of Atreides Management and a deeply knowledgeable observer of AI markets — declared in a December 2025 podcast and lengthy X thread that the balance of power in frontier AI has fundamentally changed.
For the first time, OpenAI has slipped to third place, overtaken by Google (with Gemini 3) and xAI (Grok 4.1) in what Baker calls the "token war."
Baker's core thesis: raw benchmark dominance is giving way to token economics as the decisive battleground. While Apple and Nvidia command trillion-dollar valuations without being "cheap," AI is different — victory belongs to whoever produces intelligence at the lowest marginal cost.
OpenAI's much-anticipated GPT-5 (released in mid-2025) failed to reclaim the crown not because it was weak, but because it was deliberately constrained. Designed for efficiency rather than scale, GPT-5 is a smaller model routed behind load balancers to minimize inference costs.
Its goal: profitability in a world where token pricing has plummeted (Gemini 3 Flash tokens are now ~80% cheaper than comparable OpenAI offerings).
This pragmatic pivot reflects OpenAI's reliance on external cloud providers (primarily Azure and Oracle), forcing it to pay "rent" on compute — unlike vertically integrated rivals.
Google, Baker argues, is executing a classic "starve the oxygen" strategy. With full ownership of its stack — custom TPU v6 Trillium chips, vast data centers, and optimized software — Google can sustain negative margins (~-30% on cloud AI services) to undercut competitors dependent on external funding. This mirrors historical plays by Amazon and Uber: accept short-term losses to capture market share and raise barriers.
xAI emerges as the dark horse. Elon Musk's team is on track to be the first to train at scale on Nvidia's next-generation Blackwell architecture (B200/GB200 superchips) in early 2026, thanks to what Nvidia CEO Jensen Huang called "superhuman" buildout speed. The Colossus cluster — 100,000 H100 GPUs operational in just 19 days (versus years for peers) — exemplifies this velocity. Early Blackwell access could deliver a 3-5x training efficiency leap, widening xAI's lead.
The result? An insurmountable moat. Baker sees the industry consolidating into a closed oligopoly of four: Google, xAI, OpenAI, and Anthropic. Each already possesses non-public models "significantly superior" to released versions. New entrants face prohibitive capital requirements — training a frontier model now demands $10-20 billion in compute alone.
China's position looks increasingly precarious. By rejecting sanctioned Nvidia variants (B20/B30 series with crippled interconnects), Beijing bet on domestic alternatives that lag years behind.
When the West transitions to full Blackwell in 2026, the hardware gap could become "catastrophic," Baker warns.
Yet progress didn't stall during Blackwell delays, thanks to a paradigm shift: test-time compute and reasoning models (exemplified by OpenAI's o1 series). By allocating more tokens to internal chain-of-thought during inference, performance scales dramatically without new hardware — effectively discovering new scaling laws.
Baker's conclusion is sobering: the AI race is no longer about who builds the biggest model first, but who controls the cheapest, most abundant intelligence supply chain. Google and xAI currently hold the high ground. For investors and builders alike, the message is clear — the token war has entered its endgame.
Also read:
- Spotify's Aggressive Video Push: Leaked Deck Reveals Bid to Challenge YouTube with Non-Exclusive Creator Deals
- The AI 'Cheating' Assistant Drama: Cluely's Meteoric Rise, Pickle's Open-Source Clone, and a Code Theft Scandal
- Stanford's Free AI Agentic Reviewer: Accelerating Research with Instant Paper Feedback
Author: Slava Vasipenok
Founder and CEO of QUASA (quasa.io) - Daily insights on Web3, AI, Crypto, and Freelance. Stay updated on finance, technology trends, and creator tools - with sources and real value.
Innovative entrepreneur with over 20 years of experience in IT, fintech, and blockchain. Specializes in decentralized solutions for freelancing, helping to overcome the barriers of traditional finance, especially in developing regions.
