SaaS Companies Face Extinction
Dark shadows are looming behind SaaS companies – Is this a bubble ?
Big software companies (Salesforce, ServiceNow, Workday, Adobe, etc.) need to add billions of incremental revenue to avoid decimation by GPU economics and investor expectations:
Part 1: The Core Problem – SaaS Can’t afford AI and be profitable?
AI workloads aren’t free. Whether companies buy GPUs (with rapid depreciation) or rent them in the cloud (paying hyperscaler margins), costs scale faster than SaaS pricing models allow. If margins fall, valuations collapse, because the market assumes SaaS margins remain flat or expand. To sustain valuations, AI usage must be monetised with software-like margins, not absorbed as infrastructure costs.
Without a viable downstream pricing model, infrastructure players (NVIDIA, CoreWeave, etc.) can’t support sustainable growth, and the current buildout risks becoming a bubble.
Part 2: Token Use by User Type
| User Type | Monthly Tokens (MT) |
|---|---|
| Classic SaaS seat (light AI, speech) | 0.6–1.5 |
| Enhanced SaaS seat (heavy copilots, call center) | 3–6 |
| Palantir / advanced analyst | 15–30 |
| Medical imaging (radiology) | 150–600 |
| Media analysis (surveillance, video) | 1,500–3,000 |
| Media generation (video/VFX) | 3,000–30,000+ |
Part 2a: Imputed GPU Costs (Blackwell at ~$1.26/MT)
| User Type | Monthly Use (MT) | Blackwell Cost / Month |
|---|---|---|
| Classic SaaS seat | 0.6–1.5 | $0.75–$1.90 |
| Enhanced SaaS seat | 3–6 | $3.80–$7.60 |
| Palantir / analyst seat | 15–30 | $19–$38 |
| Medical imaging seat | 150–600 | $190–$760 |
| Media analysis seat | 1,500–3,000 | $1,900–$3,800 |
| Media generation seat | 3,000–30,000+ | $3,800–$38,000+ |
Part 3: Examples
Let’s model costs and margin to work out the revenue that is needed to continue
- Classic SaaS (Salesforce/ServiceNow): small AI uplift per seat (summarisation, embeddings).
- Enhanced SaaS: reasoning agents, call center copilots.
- Palantir / analyst seat: orchestrated reasoning, multi-model queries.
- Medical imaging: each X-ray/CT study = tens of thousands of image tokens.
- Media analysis: surveillance or legal video review, millions of tokens per hour.
- Media generation: VFX or AI video, billions of tokens per project.
Part 4: Financial Lens
Additional Pressure on Top 10 Software Companies
The top ten Nasdaq-listed software companies together generate more than $350B in annual revenue. To maintain ~20% EBIT growth while covering AI compute, they will collectively need to add on the order of $70B+ new revenue each year. This is before accounting for incremental GPU costs, which will further raise the bar. AI seat pricing therefore becomes central not only to product strategy but also to market valuation.
These figures are based on current user numbers and revenue baselines assume similar number of seats (not growth of users). If this revenue is from more users, the required revenue growth would be even higher to preserve margins.
Breakdown of approximate revenue growth needed (20% p.a.):
| Company | Current Revenue | Est. Users | Growth Needed ($B) | Growth/User/Year |
| Microsoft (software segment) | ~$85B | ~400M | ~$17B | ~$42 |
| Oracle | ~$50B | ~50M | ~$10B | ~$200 |
| Adobe | ~$20B | ~30M | ~$4B | ~$133 |
| Salesforce | ~$35B | ~20M | ~$7B | ~$350 |
| Intuit | ~$16B | ~100M | ~$3B | ~$30 |
| ServiceNow | ~$10B | ~1M | ~$2B | ~$2,000 |
| Workday | ~$7B | ~60M | ~$1.4B | ~$23 |
| Autodesk | ~$6B | ~6M | ~$1.2B | ~$200 |
| Atlassian | ~$4B | ~20M | ~$0.8B | ~$40 |
| Snowflake | ~$3B | ~1.5M | ~$0.6B | ~$400 |
Together this aligns with ~$70B+ revenue growth per year required just to keep EBIT compounding at 20%.
Hardware Efficiency vs. Economics
| GPU Model | Approx Cost / MT |
|---|---|
| A100 | ~$3.82 |
| H100 | ~$3.04 |
| Blackwell | ~$1.26 |
Blackwell GPUs reduce cost per MegaToken significantly compared to A100 and H100, often by more than half. But even at ~$1.26/MT, the scale of token usage in advanced workloads means absolute costs remain material. For SaaS companies, efficiency gains ease the curve but do not remove the structural need to recover AI expenses through pricing.
Vulnerability of Infrastructure Providers
This dynamic also exposes risk for NVIDIA and cloud infrastructure companies. Once the upfront scale-out of GPUs is complete, if SaaS vendors fail to pass costs downstream, usage growth does not automatically translate into proportional revenue growth for infrastructure. Margins for the supply side could contract sharply, making the current buildout fragile if the economics of AI seats do not align with sustainable pricing. In other words, NVIDIA and infrastructure providers are highly exposed: once the upfront scale-out wave passes, their long-term growth depends on whether software companies can successfully monetise AI tokens without eroding their own valuations.
Discover more from Priory House
Subscribe to get the latest posts sent to your email.