Breaking the Bottleneck | Issue 80
[5/5/2024] π0.5, BYD Deep Dive, Robots in US Manufacturing & Deloitte's 2025 Smart Manufacturing and Operations Survey!
Breaking the Bottleneck is a weekly manufacturing technology newsletter with perspectives, interviews, news, funding announcements, manufacturing market maps, and a startup database!
💥 If you are building, operating, or investing in manufacturing, supply chain, or robots, please reach out. I’d love to chat [Aditya Raghupathy]!
🏭 If you were forwarded this and found it interesting, please sign up!
Content I Enjoyed Last Week 🗞️🔬 📚
News:
P-1 AI Launch [Fortune]
Former Airbus CTO Paul Eremenko, former DeepMind researcher Aleksa Gordic, and engineering leader Adam Nagel have founded P-1 AI, which emerged from stealth with a $23 million seed funding round led by Radical Ventures. The startup is developing Archie, an AI-powered engineering assistant designed to handle routine but critical engineering tasks, interpret requirements, generate preliminary designs, and ensure compliance, thus embedding itself as a junior member of engineering teams. Initially trained on synthetic data generated by realistic physics-based simulations of components like motors and pipes, Archie aims to surpass current large language models by deeply understanding engineering-specific instructions, adapting through human feedback, and improving with real-world usage. While incumbents such as Autodesk, Siemens, and IBM explore elements of engineering-focused AI, P-1 differentiates itself by pursuing a broad, ambitious vision: an AI capable of autonomously designing complex machines. Inspired by strategies similar to DeepMind’s AlphaGo, the startup is leveraging synthetic training data to achieve a practical, near-term rollout rather than a distant research goal. Eremenko and Gordic assert this approach is feasible and pragmatic, positioning Archie as an immediate productivity enhancer while methodically paving the way toward a future where AI independently engineers advanced structures.
π0.5: A VLA with Open-World Generalization [Physical Intelligence]
Robotics researchers have developed a new vision-language-action (VLA) model, π0.5, that significantly improves robots’ ability to generalize physical skills and semantic understanding to unfamiliar environments. Unlike typical robotics methods that operate effectively only in controlled, predictable settings, π0.5 uses co-training on diverse data, including multimodal web data, verbal instructions, and various robotic experiences, to enhance adaptability. Tested on tasks such as cleaning kitchens and bedrooms in previously unseen homes, π0.5 achieved strong out-of-distribution performance, demonstrating flexibility in identifying appropriate actions (like picking up unfamiliar items) and successfully carrying out complex, multi-step activities. The π0.5 model employs hierarchical inference by first generating high-level textual action plans, which it then translates into low-level motor commands through flow matching, enabling both conceptual reasoning and precise physical execution. Ablation studies showed that multimodal web data significantly boosted generalization to novel objects, while diverse robotic data improved overall policy effectiveness. Scaling studies further revealed that increasing the number of training environments rapidly enhanced generalization, with only about 100 unique environments needed for performance nearing models trained explicitly on test scenes. These results highlight π0.5’s promise as a foundation for broadly generalizable robotic intelligence capable of navigating messy, real-world contexts.
China Export Orders Plunge [WSJ]
China’s manufacturing sector showed significant signs of strain in April as U.S. tariffs, recently escalated to 145% by President Trump, sharply reduced export orders and factory output. China’s official manufacturing purchasing managers’ index dropped to 49, indicating a contraction, while the sub-index for new export orders fell even more drastically to 44.7, the weakest level since December 2022. The data underscores how Trump’s intensified tariffs are beginning to materially impact China’s economy, with economists revising China’s annual growth forecasts down to around 3.5%, well below Beijing’s 5% target. In response, China has adopted a defiant posture, pledging increased stimulus and government spending to counteract the impact. However, economists caution that these measures may not fully offset the damage from declining trade. Goldman Sachs estimates that between 10 and 20 million Chinese manufacturing jobs depend directly on U.S. imports, highlighting the high stakes involved. While both countries have signaled openness to limited concessions, such as tariff exemptions on electronics and aviation components, substantive negotiations have yet to materialize, leaving the long-term trade outlook uncertain and further straining global economic stability.
Why Robots Are Not the Answer to US Manufacturing Reshoring Hopes [FT]
Bringing manufacturing jobs back to the US through robotics, a key goal of Trump’s economic agenda, is encountering significant hurdles. Despite half a million manufacturing jobs remaining vacant as of March, experts note that high initial costs and the complexity of integrating automation into diverse manufacturing processes limit rapid adoption. While becoming cheaper, industrial robots can still cost tens of thousands of dollars, with full integration, including infrastructure, often totaling up to $150,000 per unit, placing automation out of reach for smaller factories and businesses facing uncertain economic conditions due to tariffs. Additionally, the US lacks sufficiently skilled workers for programming, operating, and maintaining advanced robotic systems, with factories often abandoning robotics projects due to this skill gap. The Trump administration’s tariffs, reliance on foreign-made robots, lack of national robotics strategy, and reduction in scientific research funding contrast sharply with proactive initiatives from China and South Korea, whose robust government incentives have significantly boosted robot adoption. Labor union resistance, exemplified by strikes at US ports and efforts to secure limits on automation, further complicates a transition toward widespread robotic automation.
Trump’s Tariffs Are Lifting Some U.S. Manufacturers [WSJ]
Small and midsize U.S. manufacturers are experiencing increased demand as companies seek domestic suppliers to avoid paying new tariffs, particularly the 145% levies on Chinese imports. For example, Jergens Inc., a tool manufacturer, operates at full capacity due to rising orders. Grand River Rubber & Plastics expects new orders worth around $5 million annually, representing 10% of its revenue. SafeSource Direct, a U.S.-based PPE producer, has also seen orders surge, enabling the restart of production lines that produce millions of gloves monthly, making its products competitive with Asian prices. Similarly, AccuRounds and Whirlpool report increased business and anticipate long-term benefits if tariffs remain stable. However, manufacturers dependent on imports face increased production costs, complicating their outlook. Grand River must pass higher costs onto customers. At the same time, Husco International, despite halving its reliance on Chinese imports, still imports critical components and is forced to raise prices to offset tariffs. Whirlpool expects tariff-related costs to rise by $400 million this year (2.4%), offset by price hikes and expense cuts. Although tariffs provide immediate gains for some, uncertainty about tariff longevity makes many manufacturers cautious about long-term investments in hiring and expanding production.
Great Blogs:
BYD Semiconductor Deep Dive
Trump’s Semi’s Trade Policy [China Talk]
Research:
2025 Smart Manufacturing and Operations Survey [Deloitte]
Deloitte’s 2024 survey of 600 US-based manufacturing executives shows significant momentum toward adopting smart manufacturing, driven primarily by operational (49%) and financial (44%) benefits. Key impacts reported include an average of 10–20% higher production output, 7–20% improved employee productivity, and a 10–15% increase in unlocked capacity. The survey also highlighted that investments in automation are increasing rapidly, with 78% allocating over 20% of improvement budgets to innovative initiatives, and 88% planning to maintain or grow investments. Priorities for the next two years are process automation (46%), physical automation (37%), and factory synchronization (24%), aimed at addressing skilled labor shortages and enhancing productivity. Manufacturers are investing heavily in foundational technologies like sensors, cloud computing, data analytics, and AI, with 57% adopting cloud solutions and analytics, and significant adoption of industrial IoT (46%) and 5G (42%). However, manufacturers face notable challenges in their clever manufacturing transitions, with leadership buy-in, technology investment, skilled workforce availability, change management, and cybersecurity cited as major hurdles. Talent shortages remain a pressing issue: 48% of respondents report significant difficulty filling roles in production and operations management, and upskilling workers for the “Factory of the Future” is a top human capital concern (35%). Cybersecurity is another critical priority, with respondents dedicating roughly 16% of their IT budgets to it, highlighting concerns about unauthorized access, intellectual property theft, and operational disruption. Despite progress, respondents acknowledge significant room for improvement, particularly in technology maturity, workforce upskilling, and cybersecurity preparedness, signaling ongoing investment and organizational focus.
The H20 Problem and China [IFP]
The U.S. faces a critical gap in enforcing and adapting export controls for advanced AI chips, inadvertently aiding China’s advancement in frontier AI. Despite restrictions, Chinese firms, including ByteDance, Alibaba, and Tencent, have placed massive orders, totaling over 1.3 million NVIDIA H20 inference chips worth more than $16 billion, that likely violate existing U.S. controls designed to prevent powerful AI capabilities from being built in China. Past enforcement failures, including allowing NVIDIA’s H800 and H20 chip exports and TSMC’s unauthorized sales of advanced components to Huawei, have significantly reduced America’s strategic advantage in AI. To address this urgent issue, the U.S. government must immediately halt NVIDIA’s H20 shipments to China, enforce compliance through stringent “Know Your Customer” obligations, and update export control regulations to explicitly include inference chips due to their increasingly crucial role in advanced AI. Additionally, establishing a specialized technical team within the AI Safety Institute (AISI) at NIST would enhance proactive monitoring and forecasting of AI threats, helping policymakers keep pace with rapidly evolving technologies. Finally, implementing chip geolocation features and creating clear incentives for their adoption will significantly improve oversight and enforcement, curbing the illicit flow of advanced AI hardware to adversaries.
Podcast/Video:
Reindustrializing Through Robotics [Hill & Valley Forum]
Finance & Transactions 💵
Venture Capital:
VSORA - A French company designing and manufacturing AI inference chips
€40 million [Series F] - Led by Otium and joined by Omnes Capital, Adélie Capital, and the EIC Fund
P-1 AI - A company that building general engineering intelligence
$23 million [Seed] - Led by Radical Ventures and joined by Village Global, Schematic Ventures, and Lerer Hippeau
EdgeRunner AI - A company air-gapped, on-device AI agents for military and industrial organizations.
$12 million [Series A] - Led by Madrona and joined by Four Rivers Ventures, HP Tech Ventures, and Alumni Ventures
Planned Downtime 🧑🔧
Highest 2 Lowest Trailer