⚛️ Quantum Computing — The biggest application of AI 🔌 Networking— Broadcom Carries the Torch
Our weekly digest is intended to keep you on the cutting edge of investments in data infrastructure, software, and cybersecurity.
Quantum Computing Google wow-ed the market with a BIG breakthrough in quantum computing; here is everything you need to know 👇 Quantum computing is not new – the idea has been around since the 1980s. The big bottleneck in advancing quantum computing has been error correction, as the extreme environment in which these models operate makes them highly susceptible to errors due to decoherence and noise. This is where accelerated computing comes in. With the advancements in hardware and software, companies can now simulate how a system would run and detect and eliminate errors.
This is all made possible with accelerated computing and Nvidia’s CUDA Q platform, which provides quantum computing companies with frameworks that enable them to remove errors. ⚛️Different Approaches to Quantum Computing While errors and fidelity are the most important challenges for quantum computing applications to advance, there are several others: Environmental factors: quantum computers need to operate near freezing temperatures. Qubit quality & scaling: improving qubit coherence times and gate fidelities when scaling up the number of qubits Systems: Designing control systems that can manage millions of qubits simultaneouslyCreating software on top of it that can leverage the quantum hardware and integrate with classical systems. Different companies have taken distinct approaches, with the main two technologies being super-conducting Qubits, which IBM, Google, and a newer entrant, Rigetti, have been focusing on, and Trapped-Ion technology, where IONQ is the leader. Superconducting QubitsThis technology uses superconducting circuits as qubits, operating at extremely low temperatures. The advantages of this technology are that it is scalable and has faster gate times. The disadvantage is that it has higher fidelities (more errors) and requires an extreme operating environment (near-freezing temperatures). Key players:IBM: Leader in superconducting technology with their 1,121-qubit Condor processor.Google: Developed the Willow chip with 105 qubits.Rigetti: Specializes in superconducting quantum processors, with an 84-qubit system launched in 2023.
Trapped-Ion TechnologyThis approach uses charged atomic particles suspended in electromagnetic fields as qubits. The advantages of this technology are a less extreme operating environment, higher facilities and slower delivery times. The disadvantage is a less proven scaling roadmap. Key players:IonQ: Leader in trapped-ion quantum computing, with systems available on major cloud platforms.Quantinuum: Formed by merging Cambridge Quantum Computing and Honeywell Quantum Solutions.
Quantum AnnealingYet another technology specialized for optimization problems is quantum annealing, which takes a different approach from universal quantum computers. Key player:D-Wave Systems: Pioneer in quantum annealing technology, with over 5,000 qubits in their Advantage system24.There are several other technologies, such as Quantum Annealing, Silicon-Based, and Photonic-Based Technology, that offer their own benefits and challenges. 💡Why was the Google announcement a breakthrough? Earlier this week, Google introduced its new quantum computing chip, Willow, and did a particularly good job communicating its breakthrough achievements. Willow solved a benchmark task in under 5 minutes that would take one of today’s most powerful supercomputers approximately 10 septillion years (10^25 years) to complete. Furthermore, as encoded qubits were increased from a 3×3 to a 5×5 to a 7×7 lattice of physical qubits, the encoded error rate was suppressed by a factor of two. This demonstrates the exponential error suppression promised by quantum error correction, a nearly 30-year goal for quantum computing and the key element to unlocking large-scale quantum applications. All of this was accomplished on a 105-qubit processor; can they achieve the same performance on a 1,000-qubit processor?
What may have been missed by analysts is that Google is using Nvidia’s CUDA framework to achieve this, something that will likely be available to others. Consequently, we expect significant development on this front in the next several years, with the entire ecosystem, from the Cloud Service Providers such as Google, Amazon, and Microsoft to specialized quantum computing companies to benefit. 🕰️Timeline Quantum computing today is in its infancy with limited commercial use cases but offers promising applications in many fields, ranging from chemistry and drug discovery to optimizations and cryptography. Most of these applications require billions, if not trillions, of operations to execute, which is why scaling is particularly important. Specialized quantum computing hardware companies have very little revenue today. IONQ is one of the larger ones, expecting ~$40 million in revenues, and is not surprisingly cash flow negative. Today’s customers are still very research-based rather than commercial-based, consisting of research labs and governments. However, as the technology has become available on all major cloud platforms, there has been increased interest in commercial applications. We believe that the entire ecosystem will have to invest in developing use cases and frameworks to make the technology more easily accessible, which is very similar to what Nvidia did for accelerated computing in 205-17. Cloud service providers are likely to lead the way with significant announcements from both Google and Amazon over the past several weeks.
Networking – Broadcom Carries the Torch Broadcom reported in-line results and a blowout guide yesterday after the close. Many expected a weak quarter due to fewer Google TPUs, but the long-term guide took the spotlight. Here are the key points: Guided to $60-90B in AI SAM (Servicable Market) from its three largest customers by 2023 from $15B today (60% CAGR).Expect its customers to move to 1M XPU-sized clusters into the FY27 timeframe.The company’s three largest customers are Google, META, Bytedance. Expect XPU to be 75-85% and Networking 15-25%Expanding the custom chip ASIC AI to 5 vs. 3 major customers previously. Apple’s latest addition announced earlier this week, implying a significant upside. Custom chips and networking are two of the most attractive areas of data center hardware as we look into ’25.To learn more, check out our last week’s newsletter: 🔌Networking Cornucopia🔒 Cybersecurity Inflection or tune or access out latest Data Center Webinar & Slides below 👇
