Perspectives · Governance & Accountability

How a Target Becomes a Moonshot

A sustainability requirement usually enters a project the same way. It is written into the plan, it has a date, and it often has a senior name attached. Then the work reaches the points where money is committed and the next phase is approved, and the requirement is somewhere in the room without being the thing under decision. The decision is whether to proceed. The requirement is a condition someone is meant to check. The question that governs the outcome is who has the authority to stop the work if the condition is not met. The usual answer is no one.

Google’s 2025 Environmental Report is the clearest public version of this. In 2021 the company committed to net-zero emissions across its operations and value chain by 2030. That is the requirement. It is specific, dated, and positioned above everything the company builds. Over the same years the company has been building and powering AI data center capacity at scale. In the latest report, total emissions are up 51 percent against the 2019 baseline and rose 13 percent in a single year, driven by data center expansion and the hardware supply chain behind it. Data center energy emissions fell 12 percent while data center electricity consumption rose 27 percent year over year. The efficiency work is real. The total moves in the opposite direction from the commitment.

For project governance, the question is how the gap was produced. Each data center is a capital decision. Each one passed through whatever approval process Google uses to commit the money and open the next phase. None of those approvals carried a rule that a build had to clear the carbon commitment before it could proceed. The commitment had a date and a public owner and no authority over the decisions that determined whether it would be met. The people approving the builds were approving the builds. The commitment was a separate document held somewhere else.

This is the condition worth recognizing, because it does not look like failure while it is happening. No one rejects the target. The target stays published. Each build is defensible on its own terms. The market is moving, the capacity is needed, the efficiency numbers improve.

The tradeoff is worth stating plainly. Building the compute now is how a company stays in the AI market. Holding the carbon commitment on its original terms would mean slowing or refusing some of those builds. Both cannot happen at the current pace. An organization that wanted the requirement to bind would have to give someone the authority to hold a build at the gate until it cleared the commitment, and accept that some builds would be delayed or stopped. Google did not structure it that way, and the numbers show the result.

What happened to the requirement is the part to sit with. It was not enforced and the builds were not stopped. The language describing it changed. In the 2025 report the targets are called moonshots and described as ambition-based, softer terms than the company used before.

That is the integrity risk in its finished state. A requirement that is present on paper and optional in practice is harder to correct than one that was never set, because the published version absorbs the scrutiny. The target can still be cited. It still does the reputational work of a commitment while carrying none of the authority of one.

The version of this on most projects is smaller and quieter than the Google numbers, and it runs the same way. A sustainability requirement sits in the plan with a senior name and no gate it has to clear. It is treated as everyone’s responsibility and assigned to no one’s authority. The requirement gets marked complete while delivery does not match it, and when that gap surfaces, the available move is to change what the requirement is called.