Lightmatter

Source: CNBC via GV

Founders: Darius Bunandar, Thomas Graham and Nicholas Harris
Founding: 2017
Mission:  Accelerate human advancement through computing
Employees: 165 & ~35% Local
Workplace: Hybrid
Stage & Capital Raised: Series C & $450M+ raised
Investors: Fidelity, GV, Hewlett Packard, Sequoia, Spark Capital, Viking Global
Key Customers: Some of the largest chip makers & cloud service providers
Glassdoor Rating:
4.2
Valuation (estimated): $1.2B (reported as part of their Q4 ‘23 $155M Series C fundraise)
^ this is a useless number from MGMT Boston. There is no tangible valuation until the business is sold or goes public. Don’t forget it!

It’s officially summer so this week I’m bringing you a lighter version of one of the freaking coolest technology companies building here in Boston: Lightmatter. Here are 10 things you need to know about this late stage, space age hardware company..

  • Lightmatter is building a photonic supercomputer company, reinventing how chips interact using optics instead of electrical transistors to power the GPU led AI revolution 
  • Founder & CEO Nicholas Harris has over 50 patents and 80 publications to his name, joined by Chief Scientist Darius Bunandar and CFO Thomas Graham. The trio met at MIT
  • Major bottlenecks in AI development include hardware horsepower limitations and the electricity it takes to run the models. Lightmatter aims to use light (or photons) to better connect chips in a datacenter to help them run more efficiently and at a much larger scale
  • The global semiconductor sales forecast for 2024 is $611B (src) and NVIDIA is the largest player in the GPU market (I know you knew that!)
  • Lightmatter offers three products: Passage, Envise & Idiom
    • Passage is hardware that enables arrays of different chips to communicate with each other using photonics (light) instead of traditional electrical transistors and without fiber optic cable connections. This drives 5x more energy efficiency & 200x more bandwidth
    • Envise is a more efficient hardware server that combines photonic & transistor based systems to save power and deliver increased compute speed & performance in data centers
    • Idiom is a software platform that automates deployment of AI models to Lightmatter hardware & optimizes neural network performance too
  • Lightmatter’s flagship Passage hardware product aims to be a more transcendent traffic control officer in data centers, allowing GPUs to communicate directly with each other instead of through layers of switches
  • With Moore’s Law, based on transistor technology, finally slowing down (and arguably broken), a new promise of photonic computing allows for step function higher bandwidth and better environmental efficiency in semiconductor development
  • OpenAI’s GPT-4 was estimated to have been trained on ~20,000 NVIDIA GPUs. Lightmatter has a goal for 2026 to build Passage to enable capacity for similar models to run on 1M+ GPUs
  • The startup has raised $450M+ from investors like Fidelity, GV, Hewlett Packard, Sequoia, Spark Capital, and Viking Global to make their vision a reality
  • Boston is Lightmatter’s HQ and they have offices in Toronto and Mountain View

Operators to Know:

My investigative powers continue to need work so apologies to the Lightmatter team I know I missed many up & coming operators internally

Key Roles To Be Hired (Locally):

If I were interviewing here are some questions I’d ask:

  • What are the key technology milestones that must be reached to make photonic supercomputing a reality?
  • What are the company & strategic goals for 2024? 2025?
  • Who are the best Lightmatter customers?
  • What are the most important roles you’ll be looking to add in 2024 // teams that need the most help?

To learn more about Lightmatter you’ll have to D.Y.O.R. I’m excited to watch this team bring us closer to AGI. All of humanity applauds your efforts. See you around town!