Why Your AI Talent Doesn’t Want to Shovel: Blackwell, H100s, and the Cost of Enterprise Inertia

Let’s talk about shovels and backhoes.

Recently, I compared NVIDIA’s H100 GPUs to shovels—indispensable tools, but not exactly built for scale when you’re digging serious holes. My analogy sparked a deeper conversation. A friend offered a twist: what if we think of the new NVL72 Blackwell-based systems not just as a bigger, better shovel—but as the backhoe?

Suddenly, the conversation shifts.

Everyone Needs a Shovel, But Not Everyone Should Be Shoveling

H100s have been the go-to for enterprise AI workloads. And make no mistake—there’s still a lot of work to do that doesn’t require more than a “shovel’s” worth of compute. Fine-tuning small models, running inference on legacy workloads, or standing up initial PoCs—H100s are perfectly serviceable.

But here’s the kicker: even when the job can be done with a shovel, increasingly no one wants to do it that way.

Ever watch a contractor rent a backhoe to dig a hole that you and I could knock out in a day with a shovel? That’s not laziness—it’s economics. Labor is expensive. Time is limited. And talent? It’s nearly impossible to find people who are both willing and able to “shovel” in a world where the high-value talent wants to operate machinery.

The same is happening in enterprise AI.

The Talent Problem No One Talks About

Everyone’s chasing AI engineers, ML ops professionals, and data scientists. You’re paying top dollar to bring them in. And the first thing they ask? “Where’s the GPU cluster?”

They don’t want to SSH into half a rack of underutilized H100s buried in a colocation facility. They want a platform. They want scale. They want to run RLHF fine-tuning jobs without negotiating with IT for compute access. They want something that just works—like an NVL72 pod, with software-defined networking, optimized power and cooling, and integrated developer tooling.

In other words, they want a backhoe. And increasingly, so do the workloads.

The Inertia of the Enterprise

Here’s where things get real.

Most of the enterprise just signed off on their H100 purchases. CapEx committees reviewed them. Datacenter teams are still figuring out power and cooling. You think they’re going to turn around six months later and approve a shift to Blackwell?

Not likely.

This is where enterprise inertia kicks in. It’s not that they don’t see the value in Blackwell—it’s that they haven’t even digested the last meal. The shovel hasn’t been paid off yet. Meanwhile, the AI team is eyeing the construction site next door, where the competition just pulled up with three backhoes and a dump truck.

Being Right Doesn’t Mean Being Ready

This reminds me of what I wrote after the Southwest Airlines meltdown​. The CTO probably saw the risk a mile away. But if leadership says no, you don’t stop preparing. You start prototyping, modeling, building relationships, and laying the groundwork for when the shift becomes inevitable.

The same applies here. Smart CTOs aren’t waiting for the green light to move to Blackwell. They’re prototyping new workloads. They’re engaging with NVIDIA, Dell, HPE, and cloud providers. They’re understanding the operational model of NVL72s. They’re ready to move the moment leadership realizes that it’s not about shovels anymore—it’s about getting the job done, faster and with less human capital.

It’s Bigger Than GPUs

Look, the Blackwell announcement isn’t just a hardware refresh. It’s a signal. The same way DB2 on RDS wasn’t just “DB2 in the cloud”​—it was a shift in how we manage enterprise workloads. A backhoe isn’t just a faster shovel. It’s a different job site, with different risks, economics, and talent requirements.

That’s what Blackwell is. It’s a statement about where AI infrastructure is going. And it’s a challenge to enterprise IT leaders:

Are you going to keep asking your people to shovel? Or are you going to give them the machinery they need to move at the speed of business?