Akamai Gave Anthropic a 5% Warrant. AI Infra Is Now a Corp-Dev Question.

The rental frame just broke

In 36 hours, three moves pushed AI infrastructure from a rental market into a corporate-development market.

Akamai signed an $11.6 billion, seven-year cloud commitment with Anthropic. It also gave Anthropic a warrant for up to 5% of Akamai's common stock, or 7.7 million shares at an exercise price of $111.33. About 2% vests with the announced commitment. The other 3% vests only if the contract grows by another $9 billion, which would bring the total to about $20 billion. Akamai's stock jumped 22% after hours. Anthropic's contracted compute commitments have now passed $500 billion in under a year.

Twelve hours later, Databricks bought Row Zero, a cloud spreadsheet its own finance team had adopted. CEO Ali Ghodsi told TechCrunch, "we intend to do many more acquisitions like this in the future." Row Zero raised $10 million at a $40 million post-money valuation in May 2025, and the price wasn't disclosed. Databricks has $7 billion in annualized revenue and raised another $5 billion in August. Row Zero is its sixth acquisition of 2026.

Crusoe CEO Chase Lochmiller told Newcomer the company will make $2 billion in revenue this year. He told Bloomberg that growing political opposition to AI data centers is "a marketing issue." In a March 2026 Gallup poll, 70% of respondents opposed a data center in their own neighborhood.

The three moves belong in one story. Akamai used its own equity to lock in a customer at the model layer. Databricks skipped a build-versus-buy decision by acquiring a tool it already used internally. Crusoe is trying to get past community opposition with a better story. All three are corporate-development answers to a compute market that no longer behaves like a normal enterprise SaaS category.

For two years, enterprise buyers priced AI infrastructure the way they price cloud. They negotiated a discount on a multi-year commitment and passed any price changes on to customers. That approach assumed a utility-style supply chain, where many buyers share a few suppliers and no one holds equity behind the invoice.

The Akamai deal breaks that assumption. Akamai is a CDN (content delivery network) worth roughly $16 billion. It gave one AI customer a warrant for 5% of its stock in exchange for booked compute revenue. Every buyer downstream of Akamai is now tied into an equity arrangement it can't fully see, and that's a governance risk those buyers never agreed to.

Both sides do well. Anthropic's strike price is $111.33, and Akamai closed above that after hours. So any exercise buys Akamai shares below market price, while buyers downstream pay full price for compute. Akamai has roughly 145 million shares outstanding. A fully exercised warrant would dilute existing shareholders by about 5%, eating into the 22% jump the announcement gave them. Anthropic gets the option to turn compute spending into equity. Akamai gets revenue visibility no other CDN can match. The buyer three layers downstream signed nothing and still ended up with a new stakeholder in its supply chain.

Monday's piece came at the same pattern from another angle. It covered Jamie Dimon's $1 trillion figure and Verda's $189 million round, along with Harvey pulling OpenAI out of its stack. The argument then was that the AI compute market is being financed on token pricing its biggest customers are already walking away from. Akamai's warrant takes that a step further. The financing now includes equity in the infrastructure layer, and the customers with pricing power sit on both sides of the ledger.

The platform side is running the same play

Databricks doesn't need equity in its suppliers, because other companies sell into its platform. Its version of the trade is buying small companies whose products it already uses internally. Row Zero was in Databricks' finance stack before Databricks owned it. Ghodsi's "many more acquisitions like this" matters more to GTM founders than it does as corporate news. Platforms are treating internal adoption as their main acquisition pipeline. For enterprise data tooling, that limits venture-scale exits to categories the platforms haven't tried yet.

That's bad news for a founder pitching a Row Zero equivalent to Databricks, Snowflake or Confluent this quarter. If your headline win is "we're the tool Databricks uses internally," you're in a private auction where the buyer holds both bids. Faster ARR growth only brings that auction sooner. This is the second time this quarter the platform layer has moved to consolidate before prices settle. Last week's frontier-lab pricing story showed the model side doing the same thing, buying capacity to lock in position before the pricing curve was clear.

Cursor's July 2025 acqui-hire of Koala shows the other possible outcome. Cursor had been using Koala internally for prospecting. It bought the engineers and shut the product down nine weeks later. Databricks, by contrast, said Row Zero will keep serving outside customers. In TechCrunch, both look like ordinary acquisition announcements. For a founder pitching a strategic exit, what matters most is whether the buyer plans to keep the product or absorb the team. That decision usually isn't made until after the price is agreed.

Crusoe is the third example because it applies the same corp-dev instinct to reputation. Lochmiller is right that data-center opposition has a communications gap, but marketing won't close it. When a Gallup poll shows 70% opposition, the problem runs deeper than messaging. Every siting fight Crusoe loses in Q4 means capacity has to be built somewhere else at higher land and power costs. Those costs get passed on to customers whose renewals already assume inference costs are falling. Property-tax deals or a community stake in the facility might change that. So might direct payments to the counties.

The strongest objection is that none of this is new. Cloud contracts have always come with strategic sweeteners, and enterprise acquisitions have long followed internal use. Data-center projects meet local opposition in every cycle too. On this view, only the size of the checks has changed. That's mostly right, but two things make this week different. The Akamai warrant has a strike price and a vesting schedule. That makes it a contractual claim on equity that grows with bookings the buyer can't see, which is far more than a sweetener. The pace is new, too. Three of these announcements landed within 36 hours from three different parts of the stack. Google shipped Gemini 4 the same week, 12 days after Amodei's slowdown essay. This pattern is no longer rare.

What this means for CROs

Three concrete moves, in order of urgency.

Add a supplier-equity disclosure to the vendor packet. Ask every AI vendor with more than $250K in annual spend two questions. Does it hold equity in its compute supplier, or the right to acquire some? Does the supplier hold equity in the vendor or its parent company? If either answer is yes, ask for the strike price and vesting schedule, plus any change-of-control terms. Legal already asks similar questions about sub-processors during SOC 2 review. The commercial team should ask them now too.

Rebuild your inference-cost projection from a floor instead of a steadily falling curve. The Crusoe story shows that land and power for compute are getting harder to secure. Say your 2027 unit-economics model assumes inference costs fall 30% a year. If your AI vendor's supplier is losing siting fights in three counties, that decline may slow or stall. Model both scenarios and build the difference into your pricing now.

Plan for the acquisitions your vendors' sales teams can't talk about. Databricks, Snowflake, Salesforce and OpenAI now announce acquisitions in adjacent categories on a rolling basis. If a tool you rely on sits in one of those categories, price in that risk. Your 2027 renewal may be with a different company than the one you signed with this year. Every founder in those categories is on some platform's internal-use shortlist, whether they know it or not.

Our bet is that the supplier-equity question makes it into standard enterprise SaaS RFP templates within two quarters and gets its own SOC-style standard within four. The Akamai-Anthropic warrant won't be the last deal of its kind, and it's the first one clear enough for a procurement team to point to.

The buyer three layers downstream signed nothing and still ended up with a new stakeholder in its supply chain.

Ask every AI vendor with more than $250K in annual spend whether it holds equity in its compute supplier, or its supplier in it. If either answer is yes, get the strike price, vesting schedule, and change-of-control terms.

By Business of GTM