What is deeptech? | Signals from the frontier | Deeptech reads | Deeptech deals
Speaking at IIIT Delhi Sam Altman recently said
Honestly, I feel so bad about the advice I gave while I was running YC I’ve been thinking about deleting my entire blog
He was of course referring to his previous advice of launching quickly, raising as little capital upfront as possible, and avoiding R&D risk in what you are building. Advice he himself ignored in the foundation of the now hugely massively OpenAI.
Thanks for reading Deep in the Chasm! Subscribe for free to receive new posts on Deeptech, product management and innovation.
But I believe Sam Altman is wrong. His previous advice is valuable. The proof of this is that Y Combinator is the most successful accelerator partly off the back of that advice.
However, where I think Sam Altman is mistaken is that his previous advice is contextual. Most YC startups are “shallow tech” where his advice makes sense. But OpenAI is a deeptech startup and deeptech is different.
What makes sense for most YC startups does not necessarily make sense for a deeptech startup like OpenAI.
What is deeptech?
Deeptech is often defined in terms of certain technologies, or in terms of spinning out from university research, or innovation based on high tech engineering and science. However, I think a much better definition is to think of deeptech in terms of risk.
Innovation face three types of risk:
- Market: The customers a business serves, the markets they operate in and the competitors they face.
- Implementation: The business model, they team and how they execute
- Technology: The product and technology stack that supports it
For most new ventures the biggest area of risk lies with the market. Sam Altman’s advice to YC startups was based on minimising risk in implementation and technology and then doing everything possible to manage the large inherent risk of the market. Get clear on who you customer is, understand their problem or need, build, learn iterate.
For deeptech ventures, their biggest area of risk lies in technology. That doesn’t mean there isn’t any risk in the market or implementation. There is always some. But if a deeptech startup succeeds the market is much more likely to be guaranteed.
If you are successful in building a human level intelligence AI model people will want what you are selling. If you are able to create commercially viable fusion power your technology will become foundational to the energy transition. If you can scale your quantum computing technology then you will revolutionise whole industries.
Deeptech is then a new venture where the majority of the risk lies in technology. Therefore the YC/lean startup approach focused on addressing market risks is not appropriate. A different approach oriented towards minimising the risk in market and implementation and manage the technological risk.
Clearly this is a spectrum and there is no hard line between shallow and deep tech but a continuum over which the risks of market, implementation and technology will change and evolve. To be successful in deeptech requires an awareness of these risks and a range of tools and techniques that can be applied depending on the context.
Lessons for investors
Reward follows risk and early innovation is always highly risky. But it is important to understand where that risk lies. If risk is higher in technology for deeptech then for a similar valuation as a standard startup, risk should be lower in market or implementation to compensate.
This means deeptech investments are best focused on technologies that reduce the cost of a commodity or improve the productivity of an existing asset to the point where it gives the customer an unfair competitive advantage in the market against those who don’t adopt that technology. This lowers the risk of market and implementation which balances the higher risks associated with the technology.
But prices are not dictated by risk but by supply and demand. With fewer investors focused on deeptech because of concerns around the larger capital requirements and time to market, supply of investment capital is lower. As a result deeptech investors can often get preferable terms over the more crowded standard technology investor even in today’s times.
Why is deeptech important?
Deeptech ventures are not just building apps or websites, they are often addressing big societal, economic or environmental challenges. They have the potential to create entirely new markets or disrupt existing ones. Deeptech is important because it has the potential to transform the world for the better by solving many of the challenges that humanity faces today.
The past 15 years of easy money has meant that most innovation has focused on “shallow tech”. It may have provided the customer with utility but it did so whilst creating very little utility at a broader societal level. At the same time the customer utility that was created was often done so at the cost of subsidized customer acquisition and low profits. The poster child for this being Uber, founded in 2009 and not making a profit until the 4th quarter of 2021.
Innovation is now deepening because the cost of money is increasing, governments are shifting their focus to industrial policy, and companies are becoming more aware of the transformative impact of emerging technologies and combinatorial effects,. This will have broad and beneficial impacts. But it is important to recognise that the innovation lessons of the past 15 years may not necessarily apply in the realm of deeptech.
With Deep in the Chasm I aim to explore the challenges and rewards of deeptech. What technologies are out there, who is creating change, what tools, tips and techniques are useful and what will be the impacts.
Dive DeeperLinks to dive deeper:
- Deep Tech and the Great Wave of Innovation – BCG
- What Is Deep Tech Anyway? A Guide to Deep Tech Startups in CEE from the Wolves Summit
Signals From The Frontier
- Robots are growing green fingers with autonomous robots armed with computer vision to destroy pests: The TartanPest and scientists are finding these robots can be more efficient than humans:
- IBM claim breakthrough in quantum computing: Evidence for the utility of quantum computing before fault tolerance. Classical computers have a probability of a bit flipping from 0 to 1 when hit by a cosmic ray of 10^-27. This can be easily detected and corrected using cyclical redundancy codes. But a quantum computer is much more error prone with a probability of error occuring of 10^-4. To correct these errors using traditional approaches would require the majority of the quantum computers capacity to solve. IBM have developed a novel fault tolerant approach by manipulating the noise in the quantum system.
- NVIDIA announce the GH200 Grace Hopper “superchip” combining both a CPU and GPU and specifically designed for large scale AI applications. The chip will be used in the new DGX GH200 supercomputer providing 1 exaflop of performance and 144TB of memory but at an estimated cost somewhere in the low 8 digits! Meanwhile NVIDIA Neuralangelo model can reconstruct 3D geometry from 2D video captures.
- Ransomeware attacks on K-12 schools rose 43% in 2022 with an average recovery cost of $1.59M!
- Infleqtion announces a quantum-enabled high-performance accelerometer designed for positioning, navigation and timing that can work even under high acceleration. Meanwhile research is underway at DTU and ONERA on their Green Quantum project that could provide high precision navigation
- Samsung has become the world’s second-largest advanced chipmaker, by focusing on memory chips. However, with memory chip prices declining, the company is shifting its attention to its foundry business, which manufactures custom chips for major clients like Qualcomm, Tesla, and Intel. Samsung is now investing $17 billion in a chip fabrication plant in Texas and expanding capacity in South Korea as it aims to surpass Taiwan Semiconductor Manufacturing Company (TSMC) and become the industry leader by 2025
- Manufacturing in space takes off although expensive there may be benefits of producing some high-value products such as pharmaceuticals in a micro-gravity environment
Deeptech Reads
- Tesla’s EV charging standard is quickly becoming the industry default: How Tesla is calling the shots in the EV shift
- A deep dive into the space economy: The Satellite Renaissance
- McKinsey estimates that generative AI could add $2.6 trillion to $4.4 trillion to global GDP with biggest impacts across sales and marketing, customer operations, software engineering and R&D: Economic Potential of Generative AI
- 11 myths about Silicon Photonics: A technology that could help extend Moore’s law beyond the physical limits of current chip manufacturing
- Inside the quest to engineer climate-saving “super trees”: Deep dive into Living Carbon a startup genetically modifying trees to capture more carbon
- Hyperdimensional Computing Reimagines Artificial Intelligence Hyperdimensional computing should provide more transparency and greater abstract reasoning powers as well as being more fault tolerant and energy efficient
- What are Cobots used for? Cobots are designed to work alongside humans and improvements in their design and the integration of AI is making them ever more capable. Examples include the recently announced TM20 Palletising robot from Reeco Automation with a 20kg carrying capacity and ABB’s Swifti CRB1300 with a 11kg payload capacity and speeds up to 6.2m/s.
- The State of Indiana has awarded a total of 425 smart manufacturing matching grants totaling $45 million with an average internal rate of return of 26%. Lessons learned from 50 case studies are not surprising: Assign an accountable leader, take time to identify strategic business objectives and pair them with use cases, engage the ecosystem of technology integrators, and invest in your workforce. Read the full impact report here.
- What is IOTA? A distributed ledger technology designed for the Internet of Things
- For real time AI to work at scale you need a database technology that is highly scalable, reliable for continuous data access, fast enough to capture large data flows and flexible enough to deal with various data types. For many organisations this means the right database is the open-source NoSQL Apache Cassandra. Meanwhile Vector databases are attracting increasing attention for AI workloads and this article does a great job at breaking them down.
- These Careers Are at the Forefront of the Deep Tech Revolution
- In 2008, the EU’s $16.2 trillion economy was 10% bigger than the USA. By 2022, the $23 trillion US economy was 15% bigger than the combined EU-UK economy. With more risk averse investors and less funding available one area the EU is falling behind in is deeptech: The Deeptech Conundrum: Funding Europe’s Cutting Edge Startups
Deeptech Deals
- Four-week-old AI startup raises record $110M (FT Paywall) as the EU looks for its homegrown alternative to OpenAI
- Additional ventures spins out Carbon to Sea which utilizes ocean alkalinity enhancement to remove atmospheric CO2.
- Voima ventures raises ~$100M for third fund which will be the first Nordic based fund focused on deeptech spinouts
- NASA has spent $45M seeding 249 space tech businesses and 39 research institutions under its Small Business Innovation (SBIR) and Small Business Technologry Transfer (STTR) programs. Beneficiaries include
- Florida Funders and Leonardo DiCaprio have invested in Nuview who intend to annually map the entire Earth’s land surface with LiDAR
- The EU has announced $8.7 Billion in state aid to be allocated to 68 projects across advanced materials, chip design, and manufacturingSubscribe