Anthropic's strategy to outmaneuver big-spending rival OpenAI
Speakers
Mackenzie Sigalos
— Crypto Reporter/Analyst, CNBC
John
— CFO
Summary
CNBC's MacKenzie Sigalos reports on Anthropic's strategy to outmaneuver bigger-spending AI rivals like OpenAI by doing more with less compute — relying on smarter algorithms, better training data, and reasoning techniques instead of pure scale. The segment highlights Anthropic's 10x annual revenue growth, its multi-cloud distribution and potential IPO plans, and details the custom-silicon push at Amazon (Trainium, Project Rainier), Alphabet (TPU and Gemini's comeback), and Broadcom (OpenAI chip partnership). It also raises the question of whether the efficiency shift reduces demand for Nvidia's latest chips, while noting Nvidia and AMD are competing through systems and software ecosystems.
- Anthropic is betting on efficiency over scale, aiming to do more with less compute than rivals.
- Anthropic revenue has grown 10x annually for three years, its models run on all three major clouds, and it is eyeing a potential IPO this year.
- OpenAI's $1.4 trillion in commitments exemplify the scale-first spending Anthropic is challenging.
- Amazon's Trainium is cited as 40% cheaper than Nvidia, with Project Rainier optimized around Anthropic workloads.
- Alphabet's decade-old in-house TPU is credited with supporting Gemini's comeback and price undercutting.
- OpenAI is partnering with Broadcom to design custom chips, echoing Apple's vertical-integration playbook.
- The efficiency shift raises open questions about demand for Nvidia's highest-end chips.
- Nvidia and AMD are focusing on systems and software ecosystems rather than raw horsepower alone.
Ideas
Anthropic does more with fewer resources.
Anthropic is making what might be the most contrarian bet in tech by showing you don't need the most chips to win: with a fraction of rivals' resources, it does more with less through smarter algorithms, better training data, and reasoning-focused techniques rather than just bigger pre-training runs. The market is validating it — revenue has grown 10x annually for three straight years, Anthropic is the first LLM offered across all three major clouds (Amazon, Microsoft, Google), and even rival Alphabet distributes its product — and it is a key AI name eyeing a potential IPO this year.
Amazon's Trainium chips win on cost.
Amazon is winning on cost with its homegrown AI silicon: Andy Jassy says Trainium is 40% less expensive than Nvidia, and Amazon has trained successive generations of its chips with Anthropic in mind at Project Rainier in Indiana, where Anthropic was an early adopter and the build-out targets fast, efficient inference for business customers.
Efficiency shift questions Nvidia's top-chip demand.
The post-DeepSeek shift toward efficiency raises a key open question for Nvidia: how much do customers really need its latest and greatest chips versus getting value from the most efficient models running on the most efficient chips — a premium-demand question worth monitoring as homegrown alternatives such as AWS's Project Rainier advance.
OpenAI taps Broadcom for custom chips.
OpenAI is teaming up with Broadcom to design its own AI chips, following Apple's playbook of vertical integration now spreading to the data center, positioning Broadcom as a custom-silicon design partner in that trend.
Alphabet's TPU undercuts rivals on price.
Alphabet's vertical stack is a key part of Gemini's comeback: its in-house TPU, developed over ten years, is competitive with an excellent reputation on the Street, allowing Alphabet to undercut rivals on price in AI models and cloud.
Nvidia, AMD compete via software ecosystems.
The AI chip race is not just about raw horsepower — whoever revs hardest is not guaranteed to win; Nvidia and AMD are focusing on full systems and software ecosystems that make their chips work better together across generations, which is the real competitive differentiator.
