Akamai's $600m robotics deal comes with a warning label: its GPUs are already sold out

Akamai told analysts on its Q2 2026 earnings call that an unnamed US-based technology company has committed more than $600 million over four years to its cloud infrastructure services to power robotics development ^[1]. The more operationally useful disclosure came later in the same call, when CFO Edward McGowan said Akamai's GPU capacity was completely sold out and that the contract would require a further $500 million in capex to "replenish and expand our GPU capacity" ^[3].

Read those two statements together and the announcement stops being a scoreboard entry. The contract averages about $150 million a year ^[1]. Akamai's cloud infrastructure services segment booked $99.3 million in the quarter, up 39 percent year on year ^[7], which annualises to roughly $397 million ^[2]; the new commitment is therefore worth about 38 percent of the segment's current run rate ^[3]. Against that, the $500 million of replenishment capex is equivalent to roughly 83 percent of the contract's entire four-year value ^[4], before power, operations or financing. CEO Frank Thomson Leighton said the deal brings multiyear cloud infrastructure commitments signed so far this year to more than $2.8 billion, and that Akamai now expects overall revenue growth to accelerate into the low teens in 2027 ^[2][4][5]. Total company revenue for the quarter was $1.1 billion, up five percent ^[6], so the growth story rests on a segment that currently has nothing left to sell.

For anyone procuring GPU capacity outside the three largest clouds, the sequencing is the point. Capacity at these providers is being committed in multiyear blocks before the hardware behind it is bought, let alone racked. A price page or a "contact sales" form describes intent, not inventory.

The same source describes a second deal in the same month that fits the pattern. Hive Digital Technologies' Buzz HPC signed a five-year GPU cloud services agreement with an unnamed enterprise customer valued at $350 million ^[8][14]. The customer leases a dedicated cluster of 2,016 Nvidia Blackwell Ultra GPUs in Nvidia GB300 NVL72 systems, which works out to 28 of those racks ^[10][7], housed at the Bell AI Fabric data center in Merritt, British Columbia, with delivery and deployment expected by the end of this year ^[10]. Buzz says the contract takes its total annualized revenue to $180 million, of which about $35 million is active and realized today and about $145 million is contracted and expected to come online through Q4 2026 ^[9]. That is roughly 81 percent of the stated figure sitting in the future ^[6]. Buzz secured 6.5MW at the Merritt facility in March ^[12] and will run the cluster on Vast Data's operating system ^[11].

Two things are worth tracking. First, whether Akamai's $500 million actually converts into deliverable capacity on the timeline implied by a 2027 acceleration into low-teens growth ^[5], because sold-out capacity plus a capex plan is a promise about supply chains, not a fact about them. Second, whether Buzz's $145 million of contracted ARR lands by Q4 2026 ^[9]; contracted-to-live conversion is the only number in either announcement that tells a buyer when a cluster is available. Until then, assume the useful capacity is spoken for, and negotiate on delivery dates and power rather than on list terms.