Account Executive - Frontier AI Labs

Overview

You're selling GPU compute infrastructure to frontier AI labs—companies building foundation models, training LLMs, and running large-scale AI research. You're talking to ML engineers, infrastructure leads, and CTOs who need hundreds of H100/H200 GPUs and know more about tensor cores than you ever will. Your job is to position GMI Cloud's infrastructure as a viable alternative to AWS, GCP, Azure, or building their own on-prem clusters.

Role Snapshot

Company Context

Stage: Series A ($82M raised Oct 2024, $15M equity + $67M debt)

Size: ~100 employees

Growth: Rapidly expanding - just announced $500M AI factory in Taiwan, partnership with NVIDIA as Reference Cloud Platform Provider. Opening new US data center.

Market Position: Challenger/upstart in crowded GPU cloud space competing against hyperscalers (AWS, GCP, Azure) and specialists (CoreWeave, Lambda Labs, Vast.ai). They're betting on being faster, more specialized, and offering better economics than hyperscalers for AI workloads specifically.

GTM Reality

Pipeline Sources:

• 30% Inbound - AI startups searching for GPU alternatives, developers finding them through comparison articles ("Best GPU Cloud 2025"), NVIDIA partner referrals
• 60% Outbound - You're building lists of AI labs from Crunchbase, AngelList, tracking Model releases on HuggingFace/Papers with Code, finding companies raising funding for AI
• 10% Partnerships - NVIDIA ecosystem, channel partners, VAST Data integration deals

SDR/AE Structure: Likely self-sourcing or minimal SDR support at this stage (100 employees, Series A). You're doing your own prospecting.

SE Support: You'll have Solutions Engineers/Architects to run technical demos, help with POCs, and explain GPU cluster configurations. But initial calls and discovery are on you.

Competitive Landscape

Main Competitors:

• Hyperscalers: AWS (EC2 P5 instances), GCP (A3 instances), Azure (ND-series) - established, trusted, but expensive and capacity-constrained for H100s
• GPU Cloud Specialists: CoreWeave (well-funded, NVIDIA-backed), Lambda Labs (developer-friendly), Nebius, Together.ai, RunPod
• On-Prem: Some labs just buy their own clusters if they have capital

How They Differentiate: NVIDIA Reference Platform status, claim better price/performance than hyperscalers, newer H200 GPUs, InfiniBand networking for distributed training, containerized deployment, faster spin-up than AWS quotas

Common Objections:

• "We're already on AWS/GCP and don't want multi-cloud complexity"
• "How do we know your capacity will be there when we need to scale?"
• "Your company is too small - what if you go out of business mid-training run?"
• "CoreWeave has more NVIDIA backing and proven scale"

Win Themes: Cost savings (20-40% cheaper than hyperscalers for same GPU), faster access to latest NVIDIA chips, better performance for training workloads, dedicated support, no quota battles

What You'll Actually Do

Time Breakdown

Prospecting/Research (35%) | Active Deals (40%) | Internal/Admin (25%)

Key Activities

• Prospecting Frontier Labs: You're building lists of AI startups that just raised funding (they need to burn through that capital on compute), research labs publishing papers, companies announcing new models. You send cold emails, LinkedIn messages, and maybe some calls, but these buyers are hard to reach by phone. You're tracking who's training what models and when their cloud contracts might be up for renewal.

• Running Discovery Calls: When you get a meeting, you're asking about their model architecture, training requirements, current infrastructure setup, pain points with existing providers. You need to understand their workload (training vs inference, model size, dataset size, distributed training needs) to position the right GPU configuration. This means learning enough ML terminology to have credible conversations.

• Coordinating Technical Evaluations: Most deals require a proof-of-concept. You're working with your SE to set up trial clusters, help them migrate a training job, prove performance benchmarks. You're chasing them for feedback, troubleshooting issues, and competing against them just doing nothing and staying with AWS. POCs can take 4-8 weeks and many stall out.

• Multi-threading Stakeholders: You're selling to ML engineers who care about performance, infrastructure/DevOps who care about reliability and integration, finance/procurement who care about cost, and a CTO or VP Eng who makes the final call. You're doing separate calls with each, building consensus, and navigating internal politics at companies that are often chaotic startups themselves.

• Negotiating Contracts: These deals involve commit terms ("We'll give you this rate if you commit to $X/month for 12 months"), custom SLAs, security reviews, sometimes MSAs that take weeks with their legal team. You're pricing deals with your sales engineer and leadership, figuring out margin, and trying not to give away too much discount.

The Honest Reality

What's Hard

• Your buyers are extremely technical and skeptical - They've seen every GPU cloud provider pitch. They will grill you on interconnect topology, GPU utilization rates, and how you handle node failures during multi-day training runs. If you bullshit, they'll know immediately.

• Deals slip constantly - "We'll start the POC next week" turns into next month. Training runs get delayed. Their model architecture changes. They go back to AWS because it's easier than getting a new vendor approved. Your forecast is always wrong.

• You're selling to companies that are themselves unstable - Half your prospects are startups that might run out of money. The AI hype cycle means some will pivot, some will get acqui-hired, some will just ghost you because their funding fell through.

• Capacity constraints and competitive pressure - If you promise H100 availability and can't deliver when they're ready to scale, you lose. If CoreWeave or AWS drops prices, you're scrambling. If NVIDIA prioritizes chips to competitors, you're stuck.

What Success Looks Like

• You close 6-10 deals per year at $200K-400K ACV, with 1-2 larger deals pushing you over $2M ARR
• You build a pipeline of AI labs who come back for more compute as they scale (today's $50K POC becomes next quarter's $500K production contract)
• You develop real expertise in AI infrastructure and become someone engineers want to talk to, not avoid

Who You're Selling To

Primary Buyers:

• VP Engineering / CTO (budget holder, final approver)
• Head of ML / AI Research Lead (technical evaluation, requirements)
• Infrastructure / DevOps Lead (integration, reliability concerns)
• Procurement / Finance (contract terms, cost analysis)

What They Care About:

• Performance: Training throughput (samples/sec), time-to-convergence, GPU utilization rates, network latency for distributed training
• Reliability: Uptime SLAs, what happens if nodes fail mid-training, data persistence, support response times
• Cost: TCO comparison vs hyperscalers, commit discounts, pricing predictability, no hidden egress fees
• Access: Can they actually get H100s/H200s when needed, or will they hit capacity limits
• Integration: How hard is it to migrate from AWS/GCP, do their existing tools/frameworks work, what's the onboarding like

Requirements

• 3-5+ years selling complex infrastructure, cloud, or technical products (ideally GPU compute, HPC, cloud infra, data platforms)
• Genuine curiosity about AI/ML and willingness to learn technical concepts (you don't need to be an ML engineer, but you can't fake it)
• Experience with full-cycle enterprise sales - you've built pipeline from scratch, run technical POCs, closed six-figure deals
• Comfortable selling to highly technical buyers who will test your credibility
• Track record in startup or high-growth environments where you had to figure things out without perfect enablement
• Self-motivated and competitive - this role has huge upside if you perform, but no one's going to hold your hand