Alison Smith

AI Capital Strategist - untangling tech and global markets | #GenerativeAI @BoozAllen | Tweets my own | RTs ≠ endorsements

[Alison Smith @AlisonTeeSmith]

AI model providers are discovering they're running commodities businesses rather than tech companies.

OpenAI's token processing jumped from 6B to 15B per minute in a few months. Cheaper access drove demand faster than infrastructure could scale.

The response? Classic commodity market behavior:
→ Rationing access through tighter limits
→ Retiring older, less efficient models
→ Shutting down products that burn too much compute (RIP Sora)

When your marginal costs are real and rising, the "build once, sell infinitely" SaaS model breaks down completely, and you're back to managing supply chains and capacity utilization like any resource-constrained business.

The winners will be the companies that optimize for compute efficiency rather than just model capability.

#AI #ComputeConstraints #AIStrategy

[Alison Smith @AlisonTeeSmith]

Most leadership teams haven't honestly assessed which DNA their organization can actually sustain.

Can your software company handle professional liability? Can your research team build reliable operations at scale?

The technical choice is secondary to the organizational one.

4/4

[Alison Smith @AlisonTeeSmith]

The "application layer" in AI could be a trap.

Staying purely in the middle—as an interface between models and users—leads to inevitable margin compression as both frontier labs and vertical specialists squeeze from either side.

1/4

[Alison Smith @AlisonTeeSmith]

These require completely different organizational DNA.

Moving down demands inference engineering culture and cost-to-serve optimization.

Moving up demands operations culture with liability management and human-in-the-loop governance.

3/4

[Alison Smith @AlisonTeeSmith]

@areedbuilds Really interesting thought on the "post-purchase" behavior. A lot of relationship-based transactions will also change - though I think the impacts of that will be more noticeable in enterprise purchases over consumer ones.

[Alison Smith @AlisonTeeSmith]

Something we haven't really thought about in product-market fit: does your market consist of humans or agents?

1/4

[Alison Smith @AlisonTeeSmith]

The real question isn't whether the stack will flip — it's whether a truly vertical AI company can challenge the infrastructure tax fast enough to make the 10-year timeline look conservative.

#AI #TechStrategy

4/4

[Alison Smith @AlisonTeeSmith]

The 10-year timeline for AI's economic "value inversion" assumes we're operating at the same cycle frequency as the early cloud era.

However, today's AI companies can potentially collapse entire value chains in ways that weren't possible during the Cisco-to-Amazon transition.

1/4

[Alison Smith @AlisonTeeSmith]

Yes, NVIDIA captures 79% of AI's gross profit today. But that concentration exists within a cycle that moves at much higher frequency than historical platform shifts.

We aren't waiting for a decade-long inversion. We're watching the compression of technology lifecycles in real-time.

3/4

[Alison Smith @AlisonTeeSmith]

So, do you optimize for human buyers who might get displaced, or start designing for agents that could represent increasing purchase volume?

Most companies are still pricing like it's 2019.

4/4

#AI #Pricing #B2B

https://t.co/s304lBPfui

[Alison Smith @AlisonTeeSmith]

Agents operate completely differently — they can process complex pricing matrices, want granular documentation, evaluate based on structured criteria. No charm pricing, no deal psychology.

3/4

[Alison Smith @AlisonTeeSmith]

4/ CIOs think they're streamlining budgets by consolidating vendors, but they're moving costs around while making their AI initiatives more vulnerable to execution risk.

The integration work still needs to happen. Someone still has to do it.

#AI #Enterprise

[Alison Smith @AlisonTeeSmith]

71% of CIOs plan to cut systems integrator budgets to fund AI spending, according to new a16z data (https://t.co/VEq7omuAAK).

This seems backwards to me. AI is harder to integrate reliably than traditional software, not easier.

Thread 🧵

[Alison Smith @AlisonTeeSmith]

3/ What I think will actually happen: the professional services component doesn't disappear, it just gets repackaged.

Instead of paying systems integrators, you'll pay AI vendors for "success teams," implementation fees, and ongoing calibration support.

[Alison Smith @AlisonTeeSmith]

Historically I've avoided "responsible AI" conversations because they lacked technical substance.

But when your model provider can silently change behavior and tank your SLA overnight, architecture becomes the primary concern rather than ethics.

4/4

#AI #RiskManagement

[Alison Smith @AlisonTeeSmith]

Most AI teams are building a monoculture without realizing it.

If 70% of Fortune 500 companies end up running on the same GPT version, we're one bad model update away from 2008-style systemic collapse.

1/4

[Alison Smith @AlisonTeeSmith]

Three things I'm considering in every production deployment:

• Model diversity across providers
• Frozen weights for high-stakes decisions
• Third-party verification for procurement/insurance

Basically: technical mechanisms that reduce priced risk.

3/4