• Maintain core exposure to NVIDIA (NVDA) to capture massive near-term momentum following strong forward guidance of $108 billion for next quarter, while keeping a close eye on TSMC supply chain risks.
• Buy Broadcom (AVGO) to capitalize on the industry's shift toward custom AI silicon, highlighted by its breakthrough Jalapeno inference chip co-developed with OpenAI that significantly undercuts competitor power consumption.
• Accumulate Tesla (TSLA) to trade the near-term catalyst of deploying 2,500 CyberCabs in Las Vegas over the next year as the company works toward a high-margin, $100 billion robotaxi revenue target.
• Invest in Alibaba Group (BABA) to gain low-cost exposure to the commercial AI media boom, powered by its newly launched WAN 3.0 video generation model serving the 70% of Chinese compute demand driven by visual content.
• Establish long-term positions in Nuclear Energy and Small Modular Reactors (SMRs) to profit from the critical domestic HALEU fuel breakthroughs needed to power AI data centers ahead of target 2030 commercial grid rollouts.

NVIDIA (NVDA)

• NVIDIA reported quarterly revenue of $96.2 billion (up 106% year-over-year), with forward guidance of $108 billion for the following quarter and long-term 2028 growth guidance of 70% (versus Wall Street consensus of 44%).
• The company maintains high gross margins between 80% and 85%, driven by near-total market share (90–95%) in AI foundation model training compute.
• Strategic vulnerabilities and market risks were highlighted:
  • TSMC remains the sole fabricator for NVIDIA's high-end chips, creating critical supply chain concentration.
  • A growing portion of AI hardware demand is suspected to be driven by circular financing and private credit backstopped by chipmakers.
  • The software moat provided by CUDA is weakening for inference workloads as custom ASICs emerge, pushing NVIDIA to rely on its high-speed networking interconnects (Mellanox) to maintain its training moat.

Takeaways

• While NVIDIA continues to dominate AI training clusters, its long-term inference market share faces margin pressure from custom in-house chips developed by major customers.
• Investors should monitor customer-financing disclosures and hardware concentration at TSMC as potential macro risk factors.

SpaceX (Private)

• Elon Musk issued long-term revenue targets of $1 trillion by 2029–2030 and $3.5 trillion by 2033 (compared to Amazon's current baseline of ~$828 billion).
• Key drivers projected to reach this scale include:
  • Global AI inference delivery and connectivity through Starlink.
  • Humanoid robotics via Optimus (if merged with or licensed alongside Tesla's assets).
  • Terrestrial and orbital AI compute infrastructure, including long-term space-based data center deployments.
• Private market analysts highlighted that SpaceX functions as a diversified infrastructure play spanning launch, satellite communications, orbital data centers, and advanced manufacturing.

Takeaways

• SpaceX represents a broad, vertically integrated bet across physical compute infrastructure, robotics, and global communications.
• Key risks remain execution timelines, regulatory barriers, and high key-man dependency on Elon Musk.

Broadcom (AVGO) / OpenAI (Private)

• OpenAI unveiled benchmark performance for Jalapeno, a custom AI inference chip co-developed with Broadcom (AVGO).
• Jalapeno delivers 1.5x to 1.9x more AI compute per watt and up to 3.6x lower latency compared to NVIDIA's GB200/GB300 architectures, while consuming 700 watts (half the 1,400 watts required by GB300).
• OpenAI reported up to a 54x user throughput increase when hosting open-source models (such as GPT-OSS) on this architecture.
• This shift signals OpenAI's transition from being purely a compute buyer to a custom silicon designer and potential independent cloud provider ("OpenAI Compute Cloud").

Takeaways

• Broadcom (AVGO) continues to benefit from the hyperscaler and frontier lab shift toward custom ASIC design.
• As inference shifts away from general-purpose GPUs toward specialized ASICs, operating costs for frontier AI labs are poised to drop significantly.

Tesla (TSLA)

• Tesla expanded its autonomous rideshare fleet testing, registering 79 Model Y vehicles in Texas and targeting the deployment of 2,500 CyberCabs in Las Vegas over the next year.
• The company's long-term commercial goal is deploying 1 million robotaxis, which could generate an estimated $100 billion in high-margin, recurring transportation revenue.
• Tesla's vision-only autonomy approach (relying on 8 cameras and onboard neural networks) provides a significant manufacturing cost advantage over competing sensor-heavy, LiDAR-based platforms (such as Waymo).

Takeaways

• Commercial robotaxi deployment represents Tesla's primary transition from a traditional auto manufacturer into an autonomous software platform.
• Investors should track municipal regulatory approvals and revenue-per-mile efficiency as key deployment milestones.

Apple (AAPL)

• Apple introduced the Mac Studio equipped with the M5 Ultra, supporting up to 512 GB of unified memory, alongside the announcement of the M6 chip built on a 2-nanometer manufacturing process.
• The expanded memory bandwidth allows desktop-class machines to run large open AI models locally on-premises.
• Target use cases include privacy-sensitive sectors (such as legal and healthcare/HIPAA compliance) seeking to avoid perpetual cloud token fees.
• Despite leading hardware capabilities, Apple faces criticism for lagging behind its Big Tech peers in software-level AI execution and foundation model integration.

Takeaways

• Apple is positioning itself as the enterprise on-premises and privacy-first local AI hardware alternative.
• Long-term valuation upside depends on management's ability to introduce software and services that fully monetize its custom silicon.

Alibaba Group (BABA)

• Alibaba launched WAN 3.0, an AI video generation model capable of producing 30-second single-pass videos from documents, spreadsheets, and slides at a price point of $0.05 to $0.20 per second ($20,000 to $60,000 for a feature-length film equivalent).
• The product release followed a $10.2 billion private follow-on offering in Hong Kong.
• Video generation and physical "world models" now represent approximately 70% of AI token consumption in China, driven by short-form digital media, commerce, and robotics simulations.

Takeaways

• Chinese AI developers are prioritizing world models, video generation, and manufacturing automation over pure enterprise text LLMs.
• Rapidly collapsing video generation costs create near-term commercial opportunities in digital media production and synthetic content creation.

Nuclear Energy & Small Modular Reactors (SMRs)

• Advanced fission startup Actinide demonstrated enrichment of natural uranium into high-assay low-enriched uranium (HALEU) up to 15.38% purity, exceeding traditional commercial enrichment levels (3% to 5%).
• Fuel supply remains the primary bottleneck for the small modular reactor (SMR) sector, which targets commercial grid deployment around 2030.
• Modernizing enrichment techniques using electromagnetic systems and advanced power electronics is bypassing legacy post-WWII centrifuge infrastructure.

Takeaways

• Reliable domestic HALEU enrichment is a critical precursor for the commercial viability of Gen 4 nuclear power and SMR developers.
• Energy generation capacity is increasingly recognized as the primary bottleneck for expanding AI data center infrastructure.