Anthropic's $5B plan to take on OpenAI

When Dario Amodei and a band of top researchers walked out of OpenAI in 2020, they claimed they were disillusioned by the lab’s sudden pivot toward hyper-commercialization. Fast forward a couple of years, and they are pitching investors on an even more aggressive cash burn.

According to an internal Series C pitch deck obtained by TechCrunch, Anthropic is planning to raise as much as $5 billion over the next two years. The goal? Take on OpenAI head-to-head across more than a dozen major industries.

The centerpiece of the deck is an ambitious frontier model tentatively called Claude-Next. Anthropic claims this next-generation model will be 10 times more capable than today’s most powerful systems. But building it will come with a brutal price tag: the startup plans to burn $1 billion in the next 18 months alone.

10^25 FLOPs and tens of thousands of GPUs

To build a model an order of magnitude ahead of the competition, Anthropic is planning a staggering compute footprint.

The pitch deck estimates Claude-Next will require roughly 10^25 floating-point operations. That is multiple orders of magnitude larger than current flagship models, and Anthropic notes it will rely on server clusters running tens of thousands of GPUs.

Technically, the model will build on Constitutional AI, a training method Anthropic developed to let systems align themselves with human principles through automated self-critique rather than endless human feedback loops. Anthropic argues this approach makes its models significantly more steerable and far less prone to toxic hallucinations.

The company believes whoever wins this round wins the whole game. As the deck bluntly puts it: companies that train the best 2025/2026 models will be too far ahead for anyone to ever catch up.

From pure research to commercial overdrive

While Anthropic spent its first 18 months strictly as a research vehicle, the deck reveals leadership fully committed to commercialization in September 2022.

Following a closed beta late last year, Anthropic commercially launched its assistant, Claude, in March to around 15 early partners. Target industries range from legal document analysis and medical records to coding, customer support, and education. Notably, Anthropic marks high-risk areas like law, medicine, and executive assistants as requiring human supervision in the loop.

The fundraising machinery is already in motion. The deck confirms Anthropic sought $300 million at a $4.1 billion valuation, with half committed by a confidential investor at the time the deck was written. Google has also taken a 10% stake with a $300 million check, making Google Cloud its preferred compute partner.

Intriguingly, the deck also confirms that Alameda Research Ventures, the sister firm to collapsed crypto exchange FTX, led Anthropic's $580 million Series B as a non-voting silent investor. Those shares are expected to be auctioned off during ongoing bankruptcy proceedings.

Why it matters

We think this pitch deck puts an end to the idea of a gentle, research-only frontier AI lab. When even the safety-first dissidents from OpenAI feel compelled to raise $5 billion and burn $1 billion on a single training run, it proves that building frontier AI is exclusively an ultra-capitalist game of massive scale.

For enterprise buyers and developers, this aggressive expansion is a welcome development. Having a well-funded alternative to OpenAI that prioritizes steerability and constitutional guardrails gives legal, healthcare, and corporate teams leverage. They can adopt frontier intelligence without being completely locked into the Microsoft-OpenAI stack, especially given Claude's built-in human-in-the-loop workflows.

However, Anthropic's thesis carries massive financial risk. Spending billions under the assumption that 2025/2026 creates an unassailable moat leaves zero room for error. If scaling laws hit a plateau, or if open-source models make high-end training costs look like an obsolete tax, Anthropic’s burn rate could quickly become an existential trap.