Intelligence Mode?Why AI Isn't Everything in Construction Tech

This week on Practical Nerds - tl;dr:

• AI hype hasn't changed our investment thesis
• Defensibility of AI-only companies remains questionable
• Building authoring tools or data infrastructure is more valuable than AI alone
• Thoughtful AI applications in outcome-as-a-service models show promise
• Energy costs of AI could outweigh benefits in some cases
• Intelligence is bigger than artificial intelligence

Three kilowatt hours to generate one effing image.

Correction: We said on the show that Spanish is the most spoken native language in the world. This is Mandarin Chinese. Spanish is the second most spoken native language in the world.

?? Listen to this Practical Nerds episode

AI hype hasn't changed our investment thesis

We're back after a short break, and it's time to revisit a topic we discussed about 18 months ago: AI in construction tech. Has our perspective changed? Not really. We're still not jumping on the AI hype train.

In the past year and a half, we've seen some interesting developments. There have been AI companies, mostly outside of architecture, engineering, construction, and supply chain (AECS), that have shown rapid growth in annual recurring revenue (ARR). These are often in areas like automating outbound sales. But in our space? The impact hasn't been as dramatic as some might have hoped.

We've noticed a trend of indexing behavior among venture funds. They're spreading their bets across multiple AI companies in the same space, often with similar ideas. As founders, we'd be wary of accepting checks from investors with this strategy. It shows a lack of conviction and reliability.

The narrative around AI companies seems to be shifting. There's a growing fatigue among investors when it comes to backing companies that simply slap "AI" on their pitch deck. This isn't necessarily because VCs don't want to invest in AI. It's more about the pressure to see returns on the massive investments made in 2023 and early 2024.

Defensibility of AI-only companies remains questionable

One of our main concerns about AI-focused companies in our space is the question of defensibility. How unique and protected is your AI solution? This becomes even more pertinent when we consider recent developments.

Take Nvidia, for example. They recently published an open-source model rivaling GPT-4. Think about it: companies have spent hundreds of millions on AI talent and GPUs (mostly from Nvidia), and now Nvidia is using that cash to build potential rivals. It's a bit ironic, isn't it?

We're also skeptical about the proliferation of language-specific large language models (LLMs). While we understand the need for models trained on languages with unique characteristics, like Korean or some Indian languages, we question the long-term viability and necessity of these niche models.

The reality is, having an LLM isn't as defensible as some might think. It's becoming more democratized, almost like an arms race. The value and defensibility will likely come from integrations, partnerships, and distribution – not from the AI model itself.

Building authoring tools or data infrastructure is more valuable than AI alone

When we look at the potential of AI in construction tech, we're seeing that the real value might not be in standalone AI solutions. Instead, it's in how AI can be integrated into existing workflows and tools.

Let's take the example of auto-generating buildings. If you want to do this at a high level of detail, you'll need to interact with existing software and infrastructure. Why? Because either the authoring tool or the data infrastructure is what AI runs on. There's nothing else.

This realization leads us to an important conclusion: building the authoring tool or the data infrastructure is likely more valuable than building the AI itself. Full stop.

We're still in an environment where a lot of construction data isn't in a consistent, computer-readable format. This data is only now being created or has only been created in recent years. Expecting the industry to suddenly become AI-ready might be putting the cart before the horse.

Thoughtful AI applications in outcome-as-a-service models show promise

While we're skeptical of companies that lead with AI as their main selling point, we do see potential in businesses that thoughtfully apply AI to solve specific problems. We're particularly interested in how AI can be applied to what we've previously discussed as "outcome as a service" models, or what we've termed "Palantir for construction" or "Infosys for construction".

In these models, AI can be both a brilliant enabler of growth and a driver of profitability. But here's the key: you're not selling the AI. You're not even putting it front and center. Instead, AI allows you to do something else, which is what you're selling – the value that your customer gets.

We've seen some interesting applications in this vein. One company we encountered was using AI to automate repetitive tasks in drawing annotation and tagging. Another was applying AI to dramatically speed up concept material design and research, potentially saving significant time and money in the testing and iteration process.

These are the kinds of applications that excite us – where AI is being used as a tool to solve real, specific problems in the industry, rather than being touted as a cure-all solution.

Energy costs of AI could outweigh benefits in some cases

One aspect of AI that often gets overlooked in the hype is the energy cost. This could be a significant factor in determining the practical applicability of AI solutions in our industry.

A recent study analyzed the energy required per query for various tasks in large language models. While simple text classification might use about 0.05 kilowatt-hours, generating an image takes about 3 kilowatt-hours. That's a significant amount of energy for a single image.

Now, extrapolate that to generating a 3D model with millions of objects. The energy requirements could be astronomical. This raises questions about the economic viability of using AI for certain tasks in construction and design. In some cases, it might actually be more cost-effective to use human architects and designers.

This is a crucial consideration that needs to be factored into any AI solution for our industry. The energy efficiency per query or per generation needs to be such that it's actually cheaper than traditional methods. Otherwise, the ROI simply won't be there.

Intelligence is bigger than artificial intelligence

After revisiting this topic 18 months later, our stance remains largely unchanged. We're not dismissing AI outright – we recognize its potential to transform many aspects of our work in the coming years. But we're also not buying into the hype uncritically.

We're still focused on first principles. We're looking for companies that are building real businesses, solving real problems, and creating tangible value. If they're using AI thoughtfully as part of their solution, great. But we're not interested in companies that are AI for the sake of AI.

In the end, we believe that intelligence – human intelligence, business intelligence, emotional intelligence – is bigger than artificial intelligence. It's about applying the right tools, including AI where appropriate, to solve real-world problems in our industry.

So, while others might be chasing the latest AI trend, we're sticking to our guns. We're looking for founders who understand the nuances of the construction industry, who can identify real pain points, and who can build sustainable, defensible businesses to address those pain points. Whether that involves AI or not is secondary.

As we move forward, we'll continue to evaluate AI applications in construction tech with a critical eye. We're open to being surprised and excited by novel applications. But we're also committed to maintaining a level-headed approach, focusing on the fundamentals of good business rather than getting caught up in the hype cycle.

In the end, it's not about having the most advanced AI – it's about building solutions that truly make a difference in the complex, challenging world of construction. That's what we're here for, and that's what we'll continue to pursue.Find more analysis on the Practical Nerds podcast

You Can Find More Analysis On The Practical Nerds Podcast

Spotify: https://open.spotify.com/show/1Q86tEwusNGwAmRdDqjFL4

Apple: https://podcasts.apple.com/de/podcast/practical-nerds/id1689880222

Web: https://www.foundamental.com/perspectives?tag=Practical+Nerds

Companies mentioned in this episode

OpenAI: https://openai.com/

Nvidia: https://www.nvidia.com/

ChatGPT: https://chat.openai.com/

Follow the Practical Nerds

Patric Hellermann

Shubhankar Bhattacharya

Tags

#AIinConstruction #VentureCapital #ConstructionTech

#practicalnerds #aec #founders #startup