deniz.in

Markets

Weather

Loading weather

· via TechCrunch

Anthropic's $81.3B record IPO meets open-source rivals pitching cheaper AI

Anthropic raised $81.3 billion in the largest IPO on record but disclosed over $1 trillion in infrastructure commitments, while Fireworks AI argues open models can handle most enterprise workloads.

Anthropic's $81.3B record IPO meets open-source rivals pitching cheaper AI

Anthropic's record-setting debut

Anthropic became a public company on Thursday in the largest IPO on record, according to TechCrunch, raising $81.3 billion in all — including $1.95 billion from shares sold by its seven founders. Each founder sold roughly 3.4 million shares. At the upgraded price of $83 per share, each took in about $280 million before underwriting fees and taxes, roughly $30 million more per person than the $250 million they appear to have targeted at the $74 midpoint of the original price range.

The company noted in its IPO filings that each founder has pledged 80% of their personal equity to philanthropy, while also stating that these pledges are voluntary, revocable and not legally binding.

Where the money went

Anthropic's own coffers took in $104.8 billion, a figure that could reach $117 billion if underwriters buy their full allotment, TechCrunch reports. Beyond Class A common stock, the company sold $26 billion of a special non-voting Class N share class to big names in chips and cloud: Nvidia and Google bought $10 billion each, and AMD, Broadcom, Micron, Microsoft and TSMC split the rest. For comparison, Facebook's then-record IPO raised $16 billion in 2012.

The offering also redrew the ownership map. Google finished with 13.4% of Anthropic, while Amazon converted part of the promissory notes from its earlier startup-stage investment into a 14.6% stake — it would have held nearly 24% had it converted all of them. Together the two cloud providers own 28% of the company.

A trillion-dollar bill behind it

Even that haul is largely spoken for. TechCrunch reports the IPO leaves Anthropic with roughly $230 billion of working capital once existing cash is counted, yet the company needs about $70.7 billion for compute and infrastructure commitments within 12 months, as of the quarter ended September 30, 2026, and $451.6 billion through the end of 2030. Including conditional and unpriced obligations, its disclosed infrastructure commitments run between $1.05 trillion and $1.08 trillion.

The open-source counterpoint from Helsinki

The same day, at the Slush startup conference in Helsinki, Fireworks AI co-founder Chenyu Zhao was making the opposite pitch: that most enterprises can skip frontier models altogether. Fireworks runs open-weight models and tunes them on customers' own data, producing systems Zhao says are smaller, faster, cheaper and harder for rivals to copy. He describes the company as a "specialized intelligence platform" and argues that simply calling Opus or Sonnet through an API is not a differentiator. Its customers have included Uber, Shopify and Notion.

In his view, open models now trail the frontier by at most six months and probably closer to three, and the economics tip once a product passes 10,000 to 100,000 users and token spending hits $10,000 to $100,000 a month. He expects a "two-model world," with frontier intelligence reserved for high-level planning and open models handling routine work, and says buyers increasingly think in cost per task rather than cost per token. The argument echoes Microsoft CEO Satya Nadella's July warning that proprietary-model customers pay twice, in money and in leaked know-how. Still, Zhao said he would probably buy Anthropic shares himself, calling the IPO a long-term bet on AI usage reaching a far larger scale within a decade.

Fireworks has its own credibility problem, having disclosed two security incidents this fall. One exposed API keys and tokens for a limited number of accounts; in the other, attackers reached a development system through a breached cloud provider and took customer data including billing records and datasets. Notion disconnected after chat transcripts from fewer than 1% of its workspaces were exposed, and Uber said personal details of about 150,000 drivers were stolen through a vendor TechCrunch identified as Fireworks. Zhao maintains the company's zero-retention promise for inference traffic was not compromised, and attributes the Notion exposure to a batch-prediction product that stored data at rest.

Money is chasing the category regardless: Fireworks was valued at $17.5 billion this summer and is reportedly in early talks for another round, Together AI and Baseten have raised large recent rounds, and Palo Alto Networks said Wednesday it will buy the year-old Applied Compute for $4.5 billion in stock.

Why it matters

The IPO puts hard numbers on what frontier AI costs: a record $104.8 billion raise still covers only a slice of commitments approaching $1.1 trillion, and that scale of capital is what separates Anthropic and its peers from everyone else. The open-source camp argues most workloads do not need that intelligence at all, and that routing to cheaper models changes the economics of enterprise AI. But Fireworks' breaches show the bargain option carries its own risks, leaving enterprises to weigh cost and differentiation against trust on both sides of the divide.

  • #anthropic
  • #ipo
  • #open-source
  • #ai-infrastructure
  • #enterprise-ai

Related posts