On Wednesday night, April 29, 2026, Bloomberg broke that Anthropic was weighing offers to raise capital at a valuation above $900 billion. By Thursday morning, TechCrunch had a sharper number: a $50 billion round, an 850 to 900 billion dollar valuation range, and a 48-hour window for institutional investors to submit their allocation requests. A May board meeting will decide whether to proceed.
If the round closes at the upper end, Anthropic will surpass OpenAI's $852 billion post-money valuation from earlier this year and become the most valuable private technology company in history. Up from $61.5 billion in March 2025. Up from $183 billion at its Series F in September 2025. Up from $380 billion in February 2026. A 15x markup in fourteen months on a company that crossed $30 billion in annualized revenue this quarter — with private sources citing the actual run rate at closer to $40 billion.
But the number that should be on every enterprise AI buyer's whiteboard this week is not 900. It is 46.
That is the percentage of Alphabet's $62.6 billion in Q1 2026 profit that came from mark-to-market gains on its 14% stake in Anthropic. Roughly $28.7 billion of Google's reported quarterly profit was not earned by Search, YouTube, Cloud, or Pixel. It was earned by holding paper that gets revalued every time Anthropic raises a round. Amazon's number is similar: $16.8 billion in pre-tax gains from its Anthropic position, comprising more than half of its Q1 pre-tax income.
Two of the three hyperscalers your enterprise depends on for compute, storage, and increasingly your AI inference now generate the majority of their reported quarterly profit from the valuation of a private company they don't control, can't sell easily, and whose paper value is set by the same funding rounds they themselves are participating in.
That is the structure your AI vendor strategy has to reckon with. Let me walk you through what changed this week, what the circular money flow actually means, and the three decisions your team should make before the next board update.
The Trajectory Is Faster Than the Bubble Math
Start with the velocity. Anthropic's valuation curve over fourteen months:
- March 2025: $61.5B
- September 2025 (Series F): $183B
- February 2026 (Series G): $380B
- May 2026 (proposed): $850B–$900B+
That is a 15x markup in fourteen months. To put that in perspective, OpenAI took roughly two years to go from $86B (October 2023) to $852B (early 2026). Anthropic is doing the same trajectory in a third of the time.
Revenue is keeping pace, which is the only reason this isn't a 2021 SaaS-style fever. Annualized revenue went from roughly $1B in January 2025 to a publicly stated $30B in April 2026 and a privately rumored $40B by quarter-close. Eight of the Fortune 10 are running production Claude workloads. Over 1,000 enterprise customers spend more than $1 million per year on the platform — a number that doubled from 500 in under two months after the February raise. Enterprise revenue is approximately 80% of the total, versus roughly 40% at OpenAI.
If you are a CFO reading this and your only frame is multiples, the new round prices Anthropic at roughly 22 to 30 times forward revenue depending on which run rate you believe. That is rich for a SaaS company. It is not unprecedented for a company tripling top line every four months.
The structural question is not whether the multiple is high. It is whether the revenue is durable and whether the buyer base is concentrated enough to revalue overnight if a few large customers churn.
The Circular Money Flow
This is the part most enterprise leaders are missing.
Amazon committed up to $25 billion in fresh capital to Anthropic on April 20. Google committed up to $40 billion on April 24. Both rounds happened before the $900B valuation talks went public. Together with prior commitments, Amazon now sits on roughly $33 billion of Anthropic exposure carried at over $70 billion in mark-to-market value. Google holds approximately 14% of the company.
Anthropic, in turn, has committed to spend over $100 billion on AWS over the next ten years and has agreed to use Google's TPUs at scale through a separate compute partnership. The structure is mechanically simple: hyperscalers wire dollars into Anthropic. Anthropic wires the dollars back as cloud spend. The cloud providers book the spend as revenue. They book the markup on the equity stake as profit.
Robert Willens, the tax consultant cited in Fortune's analysis this week, called the structure out plainly: "It's interesting that they're able to control or influence the value of one of their own assets" through business transactions with the same company. The hyperscalers are mark-to-market participants in a valuation they help underwrite.
This is not fraud. The accounting is legal and disclosed. But it has three consequences that matter for your enterprise:
One. A meaningful portion of the AI capex narrative — the headlines about how much money the hyperscalers are pouring into AI — is in fact recycled equity capital. The compute commitment is real. The chips are real. The data centers are real. But the financial signaling that "the hyperscalers can afford this" partially depends on continuing to mark up the stake whose dollars they're spending.
Two. Hyperscaler quarterly earnings now have a private-company valuation embedded inside them. If Anthropic's next round prints at a flat or down valuation in 2027, both Google and Amazon will report large mark-to-market losses that reduce reported earnings even if their core businesses are growing. Procurement teams negotiating multi-year cloud contracts in 2027 should expect cost-of-capital pressure that does not show up in the rate card.
Three. Vendor lock-in math is now bidirectional. You depend on AWS or Google Cloud to run your Claude workloads. AWS and Google Cloud depend on Anthropic to keep generating the headlines and the GPU consumption that justify their AI segment growth narratives. If Anthropic stumbles, your vendor stumbles. If your vendor pulls back, Anthropic loses a major customer of its own.
What Enterprise Leaders Should Do This Week
For boards, CFOs, and procurement heads, the action list is short and concrete.
Reprice your concentration risk. If Claude is now load-bearing in production — and for many of you it is — the relevant exposure is not just to Anthropic. It is to the financial structure tying Anthropic to your cloud provider. Treat your Anthropic dependency and your AWS/Google Cloud dependency as a single concentrated bet, not two independent vendor relationships. The correlated downside is that a slowdown in one cascades into the other.
Get pricing protection in writing. Anthropic raised commercial pricing twice in the last twelve months. The hyperscaler markup on Claude inference through Bedrock and Vertex has its own margin layered on top. With a $40-50B raise about to close, the company has every incentive and now the balance sheet to push price further to defend the multiple. If your contracts do not contain price protection clauses for the next 24 months, renegotiate now while you still have leverage from a usage commitment.
Stop treating model selection as a back-office plumbing decision. I have seen enterprise architecture reviews where the choice between Claude, GPT, and Gemini is delegated to a senior engineer making a benchmarks call. That was acceptable in 2024. With Anthropic now structurally entangled in the earnings of two of the three hyperscalers, the model selection decision is a CFO-level vendor concentration call. Run it like one.
Build a credible exit lane to a second model family. Not a backup that runs at 5% of traffic. A real failover that could take 30 to 50% of production load within 90 days if you needed to. The capability gap between Claude, GPT-5.5, and Gemini Enterprise is narrow enough today that this is achievable. It will not stay narrow if the gap between the leading lab and the rest widens after this round funds.
What AI Engineering Teams Should Build
For my fellow AI engineering leaders, this is where the work shows up in your roadmap.
Abstract your model layer hard. If your code still has anthropic.messages.create scattered through your application logic, you are accumulating switching cost every week. Wrap every model call behind an internal gateway that can route to Claude, GPT, Gemini, or an open model based on policy. We are running Zscaler's internal MCP Gateway pilot precisely so model substitution is a config change, not a refactor.
Instrument cost per outcome, not cost per token. Token pricing is the headline number Anthropic will use to defend the new valuation. The number that determines whether you can absorb a 20% price increase in 2027 is dollars per resolved support ticket, dollars per qualified lead, dollars per closed sale, dollars per security incident triaged. If your finance team cannot see those numbers in a dashboard today, your negotiating position next year will be weak.
Run an A/B harness for every workload. For each production use case, pick a Claude variant and one non-Claude variant. Run real traffic through both in shadow mode. Measure outcome quality, latency, cost per outcome, and failure modes. Publish the comparison internally every quarter. When the next renewal cycle hits, you will have data, not vibes.
Validate guardrails at the gateway, not the model. Anthropic's safety story is tightly coupled to Claude. If you are using Claude's built-in moderation, content filtering, or system prompt hardening as your only line of defense, switching to Gemini or GPT becomes a security project, not a routing project. Push your guardrails into your gateway layer so swapping the underlying model does not regress your prompt injection or DLP posture. We are building this exact pattern with AI Guard internally.
Pre-negotiate inference capacity, not just rate. When your usage spikes during product launches or peak business cycles, the bottleneck is often not your contracted price. It is whether you can get the throughput at all. Anthropic's roadmap to 5 GW of Trainium capacity through Project Rainier and similar buildouts at Google is real, but capacity allocation in the next twelve months will favor customers with the biggest commits. Lock in burst capacity guarantees in your contract, not just the per-token rate.
The IPO Question
Anthropic is now telegraphing an IPO later in 2026. The signals are clear: early investors from 2024 and earlier are skipping this round, waiting for the public market liquidity. Anthropic is raising the round size and the valuation with what reads as an explicit goal of setting a public-market floor.
For enterprise buyers, an IPO is not neutral. Public Anthropic has different incentives than private Anthropic. Quarterly earnings calls create pressure to hit revenue targets that do not exist today. Pricing actions, packaging changes, and aggressive cross-sell of newer products like Claude Agent and Claude Code Enterprise become quarterly story-arc events. Your vendor's behavior becomes more legible — and also more pressured.
Plan your 2027 contract cycle assuming Anthropic is public by the time you renew. That means tighter clauses on price protection, capacity guarantees, and exit terms. It means assuming that pricing pages will move more frequently. And it means treating the analyst day before each major Claude release as a leading indicator of where your costs are heading.
The Bottom Line
Anthropic is on the cusp of becoming the most valuable private technology company in history, with two of the three hyperscalers carrying its valuation as a load-bearing component of their own quarterly profits. The compute commitments are real. The revenue is real. The customer concentration is real. The circular financial flow is real.
If you are running Claude in production, treat the next two weeks as a forcing function. Reprice the concentration risk. Renegotiate price protection. Abstract your model layer. Instrument cost per outcome. Build a credible second-source lane.
The companies that come out of this cycle with the strongest AI cost discipline will not be the ones who picked the right model in 2026. They will be the ones who built the optionality to switch models in 2027 without rewriting their stack — and who negotiated their 2026 contracts knowing the vendor's valuation depends on hyperscaler earnings that depend on the vendor's valuation.
The flywheel is spinning fast. Make sure you are not the one strapped to it.
Want to calculate your own AI ROI? Try our AI ROI Calculator — takes 60 seconds and shows projected savings, payback period, and 3-year ROI.
