Key Takeaways

  • Anthropic's S-1 filing revealed roughly $10 billion in cumulative burn, which investor Gavin Baker calls an upside surprise given the company's strong unit economics and enterprise distribution.
  • Contrary to reports claiming 80 percent of Anthropic's revenue came from just eight clients, the filing shows 250 accounts spending more than $10 million each.
  • Compute has officially flipped into contango, meaning forward prices to lock down GPU capacity for 2027 now exceed current spot market rates.
  • Cloud operators like CoreWeave locked in long-term compute deals at fixed prices that sit well below market today, penalizing legacy infrastructure while newer players use upfront prepayments to fund hardware.

Anthropic's S-1 Defies the Cash-Burn Narrative

Public skepticism around foundation model labs assumes they are burning infinite capital on a handful of subsidized enterprise deals. Baker sees the actual filing as proof of a viable commercial business. A recent narrative claimed Anthropic depended almost entirely on eight customers for 80 percent of its revenue. The regulatory numbers told a completely different story: 250 accounts spend more than $10 million each, showing broad enterprise dispersion.

“The fact that they've only burned 10 billion, the profitability, the cash generation, the unit economics, they're all very encouraging,” Baker noted. When enterprise buyers integrate model APIs into their daily workflows, gross margins hold up better than tech critics projected. The cumulative $10 billion burned produced an asset base generating recurring, high-dollar annual contract value across hundreds of separate corporate accounts.

The GPU Market Flips Into Contango

Most commodity cycles follow a predictable path: high current demand produces a supply rush, which drives down prices in future years. Hardware markets normally sit in backwardation, where today's scarcity commands a price premium over next year's expanded production. In artificial intelligence infrastructure, that rule just broke.

Baker pointed to data from Carmen Lee, founder of Silicon Data, presented at Nvidia's GTC conference: “Carmen Lee, the founder, stood up at GTC and said that the compute market is now in contango.” Forward contracts to secure GPU clusters in 2027 are trading above current spot prices. Despite hundreds of gigawatts of planned datacenter construction worldwide, software demand is expanding faster than power and hardware delivery pipelines can support.

This pricing dynamic punishes operators who sold early capacity on multi-year agreements. CoreWeave and other early neoclouds won market share by guaranteeing large clusters at fixed rates to hedge their early capital buildouts. “They just sadly locked in a vast amount of compute and long-term contracts that are now way, way under market,” Baker observed. “They sold their GPUs for fixed prices and that's why they're getting hammered right now.”

Cloud Cash Flow Follows Hardware Scarcity

Modern infrastructure operators are abandoning fixed long-term discounts. Instead, they demand large upfront customer prepayments and float capacity on floating spot markets. This shifts balance sheet risk directly onto model builders while allowing hosting providers to self-fund their server purchases from actual customer cash.

Baker expects this dynamic to power financial results for hyperscalers and agile neoclouds alike over the next two years. “I think you're going to see a pretty continuous acceleration in the sum of the big hyperscalers cloud revenue growth,” he said. “That number is just going to continue to accelerate and then it's dropping down at very high margins so that operating cash flow is following.”

What to Do With This

Audit your cloud and model hosting commitments this week. If you rely on spot compute instances for training or high-volume inference, do not assume 2026 and 2027 capacity will get cheaper as new datacenters open. Price your software contracts with floating pass-through terms or lock in fixed-rate inference capacity right now before spot prices reset upward.