Google Cloud's 2027 Revenue Will Depend on Two Money-Losing AI Companies

By 2027, more than 48 percent of Google Cloud's revenue is projected to come from just two customers, and both of them are losing money at a scale most companies never see. That's the estimate industry commentator Ed Zitron laid out in a Bloomberg interview published July 31, 2026, citing figures from UBS analyst Stephen Ju: OpenAI and Anthropic together will account for 27 percent of Google Cloud revenue this year, rising past 48 percent in 2027, a combined $124 billion in payments, with Anthropic alone contributing $76 billion.

For anyone building products on top of OpenAI or Anthropic's APIs, this isn't background noise about the broader AI industry. It's a concentration number about the exact infrastructure your product runs on, and it's worth understanding what it does and doesn't imply.

What the estimate actually says

Zitron's argument, made on Bloomberg and reported by PPC Land, is that Google's massive cloud infrastructure buildout looks diversified from the outside but isn't, underneath, particularly diversified at all. "Everyone is buying into these stocks because they believe all of that CapEx is going towards diverse and spread out AI demand," Zitron said, "when in fact, what it's actually doing is helping create infrastructure for two unprofitable, unsustainable companies."

The jump from 27 percent to 48-plus percent in a single year is the number doing the real work here. It means Google Cloud's growth over the next twelve months is increasingly a bet on two specific companies continuing to scale their spending, not a bet on cloud demand broadly.

Why two unprofitable companies are carrying this much weight

The concentration matters more because of who the two customers are. OpenAI's own audited 2025 financials show a $20.9 billion operating loss against $13.07 billion in revenue, with $17.2 billion of its total spend going directly to Microsoft for Azure compute, figures independently reviewed by the Financial Times after Zitron first reported them. Anthropic, for its part, has grown its revenue run rate from roughly $9 billion at the end of 2025 to over $30 billion by April 2026, but the company has told investors it now expects to reach positive free cash flow in 2028, a year later than it previously projected, as its own costs keep climbing alongside that growth.

Neither company is close to funding its own infrastructure spend out of revenue. Both are scaling anyway, on the assumption that today's losses buy tomorrow's market position. Google Cloud is now underwriting a meaningful share of that bet, and by 2027, on UBS's numbers, nearly half its growth is tied to whether that bet keeps paying off.

None of this means either company is about to fail. Both are backed by investors willing to fund years of losses, and both have shown they can raise capital on demand. But it does mean the infrastructure underneath a large share of the AI tooling market is more concentrated, and more dependent on continued outside funding, than the size of the buildout suggests.

What it means if you're building on these APIs

If your product calls the OpenAI or Anthropic API directly, you're not exposed to Google Cloud's balance sheet, but you are exposed to the same underlying dependency: your cost structure and uptime rely on a company that is spending far more than it earns, for at least a few more years. That's a normal, common way to build in this market, and it isn't a reason to panic. It is a reason to build with the assumption baked in rather than ignored.

A few concrete things worth doing if you haven't already:

• Don't hardcode a single provider into your core logic. Route model calls through a thin abstraction layer (even a simple internal wrapper function, not necessarily a router service) so switching providers or adding a fallback doesn't mean rewriting your application.
• Track your actual per-request cost, not just your plan tier. Vendor concentration risk shows up first as pricing changes, not outages. If you don't know your current cost per completed task, you won't notice when it moves.
• Keep a second provider's credentials live, even if you rarely use them. A tested fallback path, exercised occasionally, is worth more than a theoretical one you'd have to build under pressure during an actual disruption.
• Read pricing and rate-limit change announcements as they happen, not after they've already affected your bill. A company running at a loss this large has more reason than most to revisit what it charges.

None of this requires predicting whether OpenAI or Anthropic succeeds. It requires treating vendor concentration as an operational risk to manage, the same way you'd manage a single-region cloud deployment or a single-supplier dependency in any other business.

The takeaway

The 48 percent figure is an estimate, not a certainty, and UBS's model could be wrong in either direction. But the direction of travel, from 27 percent to a near-majority share of Google Cloud's revenue in a single year, describes a real and growing dependency, not a hypothetical one. For anyone whose product is itself built on OpenAI or Anthropic's API, that dependency is already yours. The question worth asking isn't whether these companies will still exist next year. It's whether your system would keep working, and at what cost, if the terms you're building on changed.

Sources: Google Cloud faces 48% revenue reliance on OpenAI and Anthropic in 2027, PPC Land, August 2, 2026; Zitron's Bloomberg interview, "Zitron: 'Everyone Has Been Sold a Lie' on AI," Bloomberg Podcasts, July 31, 2026; OpenAI's audited 2025 financials as reported by Ed Zitron and independently reviewed by the Financial Times.