Customers who require dedicated, managed hardware for compliance purposes
Customers who prefer part self-hosting, part groq-hosting
Customers who want to put groq hardware in their data-centers
Since they own the whole stack, from hardware to inference API, they are able to offer inference cheap on their lightning fast Language Processing Units (LPUs).
They’ve adopted a direct sales pipeline like Tesla.
Very few companies which are Fabless designers actually use their own hardware. This gives them an unprecedented advantage with customers like Aramco who intend on using groq LPUs in their data-centers, as groq’s team understands the pain-points of working with their own hardware.
Groq chips have very little SRAM which means they need many more chips to run the same LLMs as everyone else.
The company plans on going from 14nm to 4nm chips in 2025.
What is a metaverse ? The apps we use today are already metaverses - social media, games like GTA, etc. When we refere to the metaverse today, we really just mean even more immersive experiences, in larger physical worlds.
The first family of apps that’s likely to gain extremely large usage will be mass content - sports, concerts, other global events. The challenge with doing this is handling the concurrency of sheer number of people. Today, even 50 users in a group-chat or room starts to cause serious network issues.
The general public is betting on specific people / leadership within companies to drive metaverse growth. Ex : John Carmack who created the first game engine, which is why we have 3-D games today.
5G/6G development is essential to metaverse progress : once we can consistently support very large bandwidths on 5G, we can decouple the processor from VR headsets which would make them way less bulky, which is a central problem today.
Gaming companies/engines have a huge advantage in the race of the metaverse since they’ve solved low-latency updates for hundreds of simultaneous players on their own infrastructure (without relying on cloud providers).
Why MSFT » Meta - Meta has acquired/purchased everything they have been good at. Meta is also likely to take away a huge chunk of creator profits.
Why do skin’s for guns or character skins work ? People associate time spent with value. In this metaverse, skins are an asset, its value is recognized by other people involved in the metaverse. This is why people are buying land/buildings in certain metaverse worlds.
People are following virtual characters on social media - Kaira. Why? Why do people follow & care about celebrities?
Other interesting things going on in industry - haptics / full-body suits for interaction in the metaverse.
The cyber-security life cycle : prevent, detect, respond.
Initially we aimed for tower & moat (prevent intruders + assume internal-network is safe). But networks became increasingly complex - we started increasing their size, the kinds of devices, their extent. We now follow a zero-trust policy where-in we assume the network is compromised and entirely focus on limiting extend of damages by early detection and appropriate response.
Where do different companies/players fit into the whole ecosystem ?
Edge security: The network edge - phones, pcs, other devices (user & employee)
Identity & Access Management (IAM) + Endpoint Detection & Response (EDR, device specific) + Extended DR (XDR, across an organizations devices) + Anti-virus
Network security: Internal networks within a company - where codebases, documents, services etc exist.
Firewalls (source-destination control / access-control list) + Intruder Detection Services (IDS) + Secure Web Gateways (SWG) etc
Cloud security: Cloud environments in which microservices / are hosted and offered for public access
Karthik: Did a double MBA. Went to Wharton. During his second MBA, he had a lot of time on his hands so he was able to experiment a lot. Lost his job in the dot com burst. Came back to India and joined banking in the media domain. He later joined TOI. In the meantime he started investing his friend’s money. He is the founder of Bloom.
Rajan: Srilankan born. Went to MIT & Stanford (mechanical), then joined McKinsey. He then joined Dell, Microsoft and then ran Google India. He has a tough childhood in the middle of the Tamil-Sinhala conflict. He unfortunately saw his friends and teachers get killed before his eyes as he was returning from school on his bus. He’s invested in over 200 companies as an angel. He now runs Sequoia Surge.
Prashanth: A great friend of Nikhil. He is always busy and always finds new things for himself to do. He runs Accel Ventures India.
Seems like everyone studied engineering, worked for a bit, progressed through the ranks and eventually ended up in VC.
5 companies no matter the portfolio size give you 80% of your returns as a VC. About 20% of your picked ventures 5x.
It is very easy to exit as an angel because VC’s or larger investors want to buy out your share. As an angel you can’t expect much more than a 5x. That being said, very few angels are worth taking money from. They need to be deeply networked or very very knowledgeable for the angel to actually add value to your company.
When collecting the total fund, VC’s heavily rely on referrals. More than the returns you have generated before, the most important thing in the industry in trust.
Generally VCs will promise investors (Like Endowments or Limited Partners) ~20% compounded, after charges.
Micro-VCs is pooling personal money ~100 Cr funds, they are early investors.
VC’s are generally CAT 1 or CAT 2 under the Alternative Investment Fund Regulations (AIF) -
CAT 1 - no public markets
CAT 2 - public + private markets
CAT 3 - only public markets
What is Private Equity (PE)? VC’s generally drive all the Series rounds, until profitability. Once the company is established and profitable, that’s when PE’s step in.