Julius Lilly

Japan is the world's leading maker of factory-automation robots and ranks fourth in robot density, yet sits near the bottom on generative-AI adoption.

• Their explanation for the gap is placement: Japanese plants tend to deploy AI beside the lane as dashboards, copilots and inspection assistants, with a person still acting as the cognitive node, while Chinese dark factories put AI inside the lane, where software senses, predicts and acts without waiting for a human.
• They frame automating the operator's mind as an extension of jidoka — Toyota's automation-with-a-human-touch principle — rather than a rejection of it.

Xiaomi's Changping plant, described in the post as a US$330M facility with 11 lines and around 10 million phones a year, runs on software they call HyperIMP, which detects micro-defects and anticipates failures.

• They contrast it with Omron's Shanghai factory — roughly 100 production lines against about ten at its Japanese plants, with productivity tripled through unmanned inspection — and Toyota's wholly owned Lexus EV plant under construction in Shanghai for 2027.
• They present the plant as a base for mastering AI-driven manufacturing at Chinese speed and transferring proven methods home.

In a follow-up, they apply the same logic to the United States and Europe, laying out a six-step chain: implement AI in manufacturing, recover productivity and competitiveness, then reshore, rebuild clusters and regain industrial sovereignty. The Western version leans on US factory construction tripling to a $235.6bn peak in 2024 and a projected shortfall: 3.8 million manufacturing workers needed by 2033, with about 1.9 million roles potentially unfilled.

Read original

The bottleneck they identify is organisational latency between design and analysis, not solver time.

• Their proposed fix is surrogate models trained on prior solver output that return feedback inside the CAD viewport instead of across departmental handoffs.
• They argue Japan has an unusual advantage here, because surrogates need high-quality labelled data and Japanese manufacturers have spent decades building simulation histories validated against physical testing.
• The constraint, in their view, is upstream: data capture without usable results, stalled cross-company linkage, and design instructions still issued to suppliers as drawings.

Nissan said in June it had cut vehicle development from 55 months to 26, targeting 90% of programmes within fiscal 2026, while Chinese EV entrants reach launch in 18 to 24 months.

• They cite a 2026 SimScale survey of 350 engineering leaders finding autonomous agents in 19% of CAE workflows but only 11% of CAD, and a Honda example from June describing what the post calls the first foundation-model-orchestrated workflow for crash-safety design, producing 35 compliant bumper alternatives from one exploration against weeks of conventional CAE.
• Their summary line: Japan does not have a design-competence problem but a design-data-liquidity problem.

Read original

They cite Ronald Coase's argument that firms integrate when internal coordination becomes cheaper than buying externally, then claim three forces are pushing that tipping point back inside the firm: automation arbitrage replacing labour arbitrage, AI removing the coordination penalty that once made integration bureaucratic, and geopolitics turning every supplier into a risk. Their prescription is deliberately narrow — own the bottlenecks, key nodes, IP and learning loops, not everything — and they argue the biggest payoff is speed, because engineering, manufacturing and software inside one organisation collapse iteration from weeks to days.

Boeing outsourced about 70% of the 787, after which wing and actuation know-how sat with Mitsubishi Heavy Industries and Nabtesco; Boeing's answer was US$4.7bn to buy back Spirit AeroSystems.

• On the other side, the BYD Seal is roughly 75% in-house parts with pack costs 20–25% below rivals, Tesla refines its own lithium, and Apple builds its own chips.
• They keep two counterexamples in view: Northvolt, where they say integration without execution destroyed capital, and TSMC, where a world-class specialist outperforms integrated manufacturers.

Read original

They map the supply chain by layer — structure, materials, systems, cabin, propulsion and space — and describe the position as prime-contractor-level risk and responsibility without the prime contractor's name on the fuselage. They attribute the gap between capability and reputation to a visibility problem: success is invisible while failure is advertised.

Mitsubishi Heavy Industries, Kawasaki Heavy Industries and Subaru hold 35% of the Boeing 787 airframe, the highest foreign work share on any Boeing programme.

• Toray supplies 100% of the carbon fibre in the 787's primary structure as exclusive supplier since 2004, with cumulative 787 and 777X contracts over ¥1.3 trillion.
• Nabtesco holds near-100% of flight-control actuation on Japanese-built aircraft and actuation for eight flight-control surfaces on the 777X, and JAEC — IHI, Kawasaki and MHI — holds 23% of the PW1100G-JM engine programme for the A320neo.
• In space, MHI operated the H-IIA rocket to 49 successes in 50 flights before its 2025 retirement.
• They point to the cancelled MRJ SpaceJet as an example of the reputational cost, saying it shapes foreign perception more than 35% of every 787 in the sky.

Read original

The earlier machines proved the shape but ran ahead of the technology needed to scale it; the brains arriving now come from foundation-model AI, sensors, actuators and batteries, with EV and drone supply chains cutting costs and China adding scale through dense supplier networks. They are explicit that shipments and choreography do not equal general autonomy, and label Japanese production ambitions as targets rather than contracted output.

Waseda's WABOT-1 walked in 1973, Honda began biped research in 1986 and unveiled ASIMO in 2000 before ending development in 2018, and about 27,000 SoftBank Pepper units were produced before production stopped in 2020.

• Today Omdia estimates 13,318 humanoid shipments in 2025, up about 480%; Unitree launched the R1 from RMB39,900; and Figure says its Figure 02 loaded more than 90,000 parts over 1,250 hours at BMW Spartanburg.
• On China's cluster effect, they credit XbotPark, co-founded in Dongguan in 2014 by HKUST professor Li Zexiang — an early mentor to DJI founder Frank Wang — with an ecosystem that regional accounts put at more than 80 companies and six unicorns.
• Japan's re-entry example is Highlanders Inc., a 2023 University of Tokyo spin-off, signing a memorandum with Mitsubishi Motors to explore joint development at unused buildings in Mitsubishi's Kyoto plant.

Read original

L-band, narrowband and rain-resistant on small battery-powered antennas, keeps the machine layer alive through the disaster — sensor telemetry, distress messaging, positioning.

• Ka-band, using high-throughput spot beams, reconnects evacuation centres and backhauls mobile base stations after the shock.
• Their strategic conclusion: satellite connectivity in Japan is no longer a product category but a procurement standard, regulation-driven and budget-resilient.

They say the Noto earthquake killed 732 people and disrupted all four major mobile carriers across Ishikawa, while roughly 150 satellite terminals were rushed to Kanazawa in the first week and 550 routers were ultimately deployed to shelters and medical teams.

• Japan's QZSS (Michibiki) broadcasts tsunami warnings and emergency alerts every four seconds on the L-band layer.
• KDDI's au Starlink Direct reached about four million users in its first year, covering the roughly 40% of Japan's land area that terrestrial networks cannot economically reach.
• They also cite about US$922M committed by Tokyo in June 2026 to a domestically owned low-earth-orbit constellation called J-LEO.

Read original

Fibre still reaches almost the whole population, but its net additions are falling while home-router wireless grows several times faster. They identify two unmet gaps — enterprise-grade fixed wireless with multi-carrier management, and private local 5G networks — and note that economic-security screening of infrastructure suppliers by country of manufacture and foreign-government ties is opening a route for suppliers from trusted countries.

Fibre net additions collapsed 28.9% in the first half of fiscal 2025, and to March 2028 fibre-to-the-home grows at a 1.1% compound annual rate against 7.0% for home-router fixed wireless access, per MM Research.

• All four carriers are committed: NTT Docomo's home 5G added 120,000 subscribers in six months, Rakuten Turbo monetises spare radio capacity, SoftBank raised its fixed wireless hardware price 33% — which they read as evidence of a winning product — and KDDI with Soracom has entered the segment.
• They name three routes to market: carrier-branded supply through certification, OEM and ODM partnerships, and non-incumbent operators such as Soracom and IIJ's full-MVNO enterprise offering.

Read original

It is the clearest statement of Lilly's central argument and supplies the framing vocabulary — AI beside the lane versus AI inside the lane — used across the other posts.

Lilly argues that Japanese manufacturing rested for fifty years on the supreme execution of the human operator, and that this premise is failing because operators are scarce, not because the production system failed.

• The figures cited: manufacturing jobs at 10.46 million and falling, about 1.7 production-job openings per applicant against a 1.25 average, and wages up more than 5% for three straight years.
• The paradox they identify is that Japan is the world's leading factory-automation robot producer and fourth in robot density at 446 robots per 10,000 workers, yet near the bottom on AI adoption, with 26.7% generative-AI experience against 81.2% in China.
• The key distinction is where the intelligence sits: Japanese plants deploy AI beside the lane as dashboards, copilots and inspection assistants, keeping a human as the cognitive node, while Xiaomi's Changping plant in China runs AI inside the lane through software the post calls HyperIMP, detecting micro-defects and correcting them without waiting for a person.
• Supporting cases include an Omron Shanghai factory running roughly 100 production lines against about ten at its Japanese plants, with productivity tripled through unmanned inspection, and a wholly owned Toyota Lexus EV plant under construction in Shanghai for 2027 as an emulation base for transferring proven methods home.
• Lilly connects the argument to jidoka, Toyota's automation-with-a-human-touch principle, and to Tokyo's AI Basic Plan, which names physical AI and vertical AI as national priorities backed by about one trillion yen.
• The closing line states that the objective is not a factory without humans, but keeping the factory in Japan.

For brands planning a campaign

For an industrial-strategy client, this is the anchor piece for a campaign targeting Japanese manufacturing automation demand.

• The distinction between AI beside the lane and inside it gives clients a concrete commentary angle, while the adoption-gap statistics support a section on where Japanese small and mid-sized manufacturers are likely to invest next.
• The post's comments contain the underlying data and sources; ask for them before relying on the figures.

Read it after Part I, since it explicitly builds on the Japan case.

Where Part I covers Japan, Part II extends the same logic to the United States and Europe.

• Lilly lays out a chain of six consequences: implement AI in manufacturing, recover productivity, recover competitiveness, reshore manufacturing, rebuild GDP-generating clusters, regain industrial sovereignty.
• The figures include US factory construction tripling to a 235.6 billion dollar peak in 2024, and a projected shortfall in which 3.8 million US manufacturing workers are needed by 2033 while roughly 1.9 million positions may go unfilled.
• The organising concept is cognitive latency — segmented decisions, post-process inspection and lost handover knowledge — which they contrast with dark factories that automate execution.
• McKinsey's Lighthouse research is cited for roughly 40% average labour-productivity gains among a recent cohort.
• Two points give the argument its economic edge: a domestic plant need not match offshore wages if it wins on cost per unit through yield, cycle time and uptime, and each remaining manufacturing job is said to support about five jobs across the wider economy, making the real comparison an automated domestic factory versus no factory at all.

For brands planning a campaign

Useful when a client is weighing reshoring or European industrial policy, because it makes yield and uptime more important than labour cost. The five-jobs-per-manufacturing-job multiplier and the productivity-gain figure are quotable starting points for a scenario section, but should be attributed to Lilly's synthesis of cited third-party research, not presented as measurements by Lilly.

A good entry point for a client focused on competitiveness rather than technology fashion.

Lilly separates engineering quality from engineering velocity — how many times a design can be revised before the calendar runs out.

• The comparison: Nissan said in June it had cut vehicle development from 55 months to 26, targeting 90% of programmes within fiscal 2026, while Chinese electric-vehicle entrants reach launch in 18 to 24 months and many legacy manufacturers still need 40 to 50.
• The bottleneck they identify is organisational latency between design and analysis, not solver time.
• Their proposed remedy is surrogate models trained on prior solver output that return feedback inside the CAD viewport instead of across departmental handoffs.
• Cited figures include a 2026 SimScale survey of 350 engineering leaders in the US, UK and Germany finding autonomous agents in 19% of CAE workflows but only 11% of CAD, and METI's 2026 Monozukuri White Paper showing manufacturing workers under 35 falling from 3.84 million in 2002 to 2.58 million in 2025, with 54.8% of respondents still naming re-employment or extension of older workers as the skills-transfer answer.
• A Honda example from June describes what the post calls the first foundation-model-orchestrated workflow for crash-safety design, producing 35 compliant bumper alternatives from one exploration against weeks of conventional CAE.
• The constraint Lilly stresses is upstream: two-thirds of manufacturers capture process data but fewer than half get a result, cross-company data linkage has not moved in two years, and as of 2020 over half still issued design instructions to suppliers as drawings — no surrogate trains on a PDF drawing.
• Their summary: Japan does not have a design-competence problem but a design-data-liquidity problem.

For brands planning a campaign

This is the strongest piece for an advisory brief about engineering software, simulation or supplier data practices in Japan.

• It supports angles on treating simulation archives as capital assets, using multi-fidelity workflows where AI screens and the solver validates, and digitising tier-2 and tier-3 supplier interfaces.
• Attribute the vendor survey and white-paper figures to their publishers if reusing them with a client.

It supplies the historical and economic reasoning behind the manufacturing arguments in the other posts, making it the best single read for a client thinking about supply-chain structure rather than a specific technology.

Lilly traces a forty-year outsourcing consensus back to arithmetic: early-2000s Chinese labour below one US dollar per hour against 20 to 30 dollars in the OECD, negligible freight, and offshoring that cut landed costs by 20 to 40%.

• They cite Ronald Coase's argument that firms integrate when internal coordination becomes cheaper than buying externally, and say three forces are now pushing that tipping point back inside the firm.
• First, automation arbitrage is replacing labour arbitrage: Chinese factory wages have risen to roughly 5 to 6 dollars per hour while an amortised industrial robot runs below 2 dollars per hour anywhere.
• Second, AI is removing the coordination penalty that once made vertical integration bureaucratic.
• Third, geopolitics: Red Sea reroutings add 10 to 14 days and Asia–Europe rates are 25 to 40% above pre-crisis levels.
• The IP argument is illustrated by Boeing outsourcing about 70% of the 787, after which wing and actuation know-how sat with MHI and Nabtesco; Boeing's answer was 4.7 billion dollars to buy back Spirit AeroSystems.
• Speed is presented as the largest advantage: when engineering, manufacturing and software sit in one organisation, iteration collapses from weeks to days.
• Examples of re-integration include the BYD Seal at roughly 75% in-house parts with pack costs 20 to 25% below rivals, Tesla refining its own lithium, Apple building its own chips, and Amazon owning warehouses, aircraft and AI chips.
• Lilly notes the counterexamples: Northvolt shows integration without execution destroys capital, and TSMC shows specialists can outperform integrated manufacturers.
• A Dykema 2026 survey is cited for supply-chain resilience as the top driver of automotive deals at 55%.

For brands planning a campaign

Use this for clients assessing make-versus-buy, supplier concentration or reshoring risk in industrial sectors.

• The selective-ownership conclusion — own the bottlenecks, key nodes, IP and learning loops rather than everything — is a defensible client-facing thesis that avoids the simplistic onshoring story.
• The counterexamples let the commentary acknowledge where integration fails.

It is the most concrete supply-chain piece in the material and a strong example of Lilly's argument from named companies and market shares rather than national sentiment.

The premise is that Japan never tried to win commercial aerospace by owning the logo on the tail, instead securing irreplaceable pieces of aircraft carrying other companies' names.

• The named shares: Mitsubishi Heavy Industries, Kawasaki Heavy Industries and Subaru hold 35% of the Boeing 787 airframe in wings, forward fuselage and centre wing box, the highest foreign work share on any Boeing programme, plus 21% of the 777 and, with ShinMaywa and Nippi, 21% of the 777X.
• Toray supplies 100% of the carbon fibre in the 787's primary structure as exclusive supplier since 2004, with cumulative 787 and 777X contracts over 1.3 trillion yen, and also supplies Airbus.
• Nabtesco holds near-100% of flight-control actuation on Japanese-built aircraft and supplies fly-by-wire systems on the 737, 747-8, 757, 767 and 777, plus actuation for eight flight-control surfaces on the 777X as Boeing's first Japanese systems partner.
• Jamco supplies 100% of 787 lavatories, galleys, bars and flight-deck interiors as sole supplier.
• Panasonic Avionics is one of two players dominating global in-flight entertainment.
• JAEC, made up of IHI, Kawasaki and MHI, holds 23% of the PW1100G-JM engine programme for the A320neo, with IHI alone at 15% of that engine, 11% of the GE9X and 9% of the GE90, and MHI Aero Engines as a risk-sharing partner across three Rolls-Royce Trent programmes.
• Lilly notes that Japan sits inside Pratt & Whitney, GE and Rolls-Royce simultaneously, a position they say no other nation holds.
• Bridgestone is one of four manufacturers holding a combined 80 to 85% of the global OEM aircraft tyre market.
• MHI operated the H-IIA rocket to 49 successful flights of 50, including 44 consecutive successes, before its 2025 retirement in favour of H3.
• The closing observation is that success is invisible while failures are advertised, with the cancelled MRJ SpaceJet shaping foreign perception more than 35% of every 787 in the sky.

For brands planning a campaign

A client brief on aerospace supply chains, industrial risk or Japanese industrial policy can use this as a supplier-mapping starting point.

• Verify the market-share figures against primary sources before contractual use; Lilly aggregates them from public information.
• The perception-versus-substance argument is useful for commentary on how investors and journalists misread industrial capability.

Useful for a client who needs the history and current shipment and funding numbers in one place.

Lilly's frame is bodies awaiting brains: Waseda's WABOT-1 walked in 1973, Honda began biped research in 1986 and unveiled ASIMO in 2000 before ending development in 2018, Sony's QRIO ran from 2003 to 2006, and about 27,000 SoftBank Pepper units were produced before production stopped in 2020.

• Those machines proved the form factor ahead of the intelligence, reliability and cost curves.
• The brains arriving now come from foundation-model AI, sensors, actuators and batteries, with electric-vehicle and drone supply chains cutting costs and China adding scale.
• XbotPark, co-founded in Dongguan in 2014 by Hong Kong University of Science and Technology professor Li Zexiang, an early mentor to DJI founder Frank Wang, is credited by regional accounts with an ecosystem of more than 80 companies and six unicorns, and a local director said good ideas could typically find implementation suppliers within half an hour.
• The market figures: Omdia estimates 13,318 humanoid shipments in 2025, up about 480%; Unitree launched the R1 from RMB39,900, then roughly 5,600 US dollars; Morgan Stanley forecasts 50,000 China shipments in 2026 and 446,000 by 2030; a 2026 Goldman Sachs outlook projected about 502,000 global shipments in 2032 with a published 2035 total addressable market of 38 billion dollars.
• Figure says its Figure 02 loaded more than 90,000 parts over more than 1,250 hours at BMW Spartanburg.
• On capital, XPeng said on 24 August its robotics unit raised over 900 million dollars at a post-money valuation above 6.3 billion, described as China's largest single private embodied-AI financing, with IRON expected to enter mass production by end-2026 and external deliveries planned for 2027; PitchBook reports 6.1 billion dollars across 139 humanoid start-up deals in 2025.
• On Japan, Highlanders Inc., a 2023 University of Tokyo spin-off, signed a 9 July memorandum with Mitsubishi Motors to explore joint development and production at unused buildings in Mitsubishi's Kyoto plant, potentially from early 2027, with separately reported ambitions of 1,000 units a month by late 2027 that Lilly labels targets rather than contracted output.
• Lilly's caution is explicit: shipments and choreography do not equal general autonomy.
• A Part 3 on safeguards is referenced but not included in the posts here.

For brands planning a campaign

Suitable for a client tracking physical AI, robotics investment or Japanese re-entry into a field it once pioneered.

• The post separates company-reported figures, analyst forecasts and targets, making it a sound basis for a scenario table.
• Carry forward its distinction between autonomy and demonstration.

The strongest example of Lilly linking a technology shift to procurement and regulation.

Lilly contrasts the everyday network with the network that operates in a disaster, using two events: the 2011 earthquake and tsunami, and the Noto earthquake of 1 January 2024, which they say killed 732 people and disrupted all four major mobile carriers across Ishikawa prefecture.

• Satellite terminals did the work — roughly 150 rushed to Kanazawa in the first week and 550 routers ultimately deployed to shelters and medical teams.
• The technical argument rests on two frequency bands with complementary physics: L-band survives, being narrowband, rain-resistant and usable on small battery-powered antennas, keeping sensor telemetry, distress messaging and positioning alive through the disaster, with Japan's QZSS Michibiki broadcasting tsunami warnings and emergency alerts every four seconds; Ka-band restores, using high-throughput spot beams to reconnect evacuation centres and backhaul mobile base stations after the shock, with vehicle-mounted and ship-based satellite-backhauled base stations now a standing carrier capability.
• Supporting numbers include Japan's satellite communications market growing from 3.84 billion dollars in 2025 to a forecast 7.76 billion by 2030 at roughly 15% compound annual growth; KDDI's au Starlink Direct reaching about four million users in its first year and covering the roughly 40% of Japan's land area that terrestrial networks cannot economically reach, against 99.9% population coverage but about 60% area coverage; NTT Docomo launching its own direct-to-satellite service in April 2026 and Rakuten with AST targeting late 2026; about 922 million dollars committed by Tokyo in June 2026 to a domestically owned low-earth-orbit constellation called J-LEO; and 3GPP NTN standards letting ordinary cellular chipsets roam onto satellite, which Lilly says collapses a cost barrier that kept satellite IoT niche for two decades.
• Their conclusion: last-resort connectivity in Japan is no longer a product category but a procurement standard.

For brands planning a campaign

Strong material for a client brief on telecommunications, critical infrastructure or disaster-resilience procurement in Japan, including the point that regulation and public budgets, not consumer demand, sustain the category.

• The L-band-versus-Ka-band framing gives clients a simple technical structure for evaluating vendors.
• Trace market-sizing figures to their original publishers before using them in client deliverables.

A useful complement to the satellite piece for clients examining how Japanese broadband demand is being served.

Lilly argues that Japan's next connectivity contest is not about smartphones but the self-installed 5G box: the country's fastest-growing broadband product.

• The figures: fibre net additions collapsed 28.9% in the first half of fiscal 2025; to March 2028, fibre-to-the-home is forecast to grow at a 1.1% compound annual rate while home-router fixed wireless access grows at 7.0%, about six times faster, according to MM Research; and Japan's 5G fixed wireless market is heading toward 12.1 billion dollars by 2030 at a 42.2% compound annual rate, per Grand View.
• Lilly attributes the shift to a lasting constraint: Japan is running out of installation engineers, while a plug-in box needs none.
• All four carriers are committed.
• NTT Docomo's home 5G added 120,000 subscribers in six months; Rakuten Turbo monetises spare radio capacity; SoftBank raised its fixed wireless hardware price 33%, which Lilly reads as a sign of a winning product; and KDDI with Soracom has entered the segment.
• Lilly identifies two remaining gaps: enterprise fixed wireless, where branch failover, construction sites, retail and logistics are largely served with consumer packages while machine connections move toward 69.2% of all mobile virtual network operator lines by 2030; and private networks, with Japan's local 5G market forecast at 55.8 billion yen by fiscal 2030 from 6.3 billion in fiscal 2023, a route that needs no mobile network operator certification.
• They also note that Japan's economic-security regime now screens infrastructure suppliers by country of manufacture and foreign-government ties.
• Lilly closes by asking who will bring this architecture to Japan, and from where.

For brands planning a campaign

Directly relevant to a client considering market entry into Japanese telecoms equipment or enterprise connectivity.

• The post identifies three entry routes: carrier-branded supply through certification, OEM and ODM partnerships, and non-incumbent operators such as Soracom and IIJ's full mobile virtual network offering.
• The economic-security screening point should be checked against current policy before being repeated to a client.

The posts contrast Japan's robot leadership and low generative-AI adoption with Chinese dark factories that put AI inside the production lane, then frame cognitive automation as an extension of jidoka rather than a rejection of it. A follow-up applies the same logic to the US and Europe, linking AI deployment to productivity recovery, reshoring, and industrial sovereignty.

Combines labour and adoption statistics, robot-density comparisons, named factory examples such as Xiaomi, Omron, and Toyota, policy references including Japan's AI Basic Plan and METI material, and company-reported figures. Forecasts and targets are generally labelled as such.

For brands planning a campaign

Useful for factory-automation, industrial-AI, simulation, and reshoring campaigns. The strongest angle is practical adoption: where AI changes supervision, inspection, maintenance, and data readiness—not a generic AI-revolution message.

Examples: Source 1 · Source 2

The bottleneck is how quickly designs can be revised; AI surrogates trained on solver output can return answers inside CAD and CAE workflows. Lilly points to simulation archives as underused capital, weak supplier data linkage, and drawing-based processes as reasons the advantage is not automatic.

Uses development-time comparisons for Nissan and Chinese EV entrants, survey data from SimScale and METI, and a Honda crash-safety example to develop three executive implications.

For brands planning a campaign

Relevant to CAD, CAE, simulation, PLM, engineering-data governance, and supplier-digitisation campaigns. A brand can ask for an executive explainer on data liquidity or a client-facing note on accelerating design cycles without overpromising autonomous engineering.

Examples: Source 1

The prescription is not to own everything; it is to own bottlenecks, key nodes, IP, and learning loops. Examples include BYD, Tesla, Apple, Amazon, Boeing's Spirit AeroSystems buyback, and cautionary cases such as Northvolt.

Uses Coase's theory of the firm as a frame, then tests it with wage benchmarks, freight and supplier-risk data, company examples, and M&A survey evidence.

Lilly notes that integration without execution can destroy capital and that specialist manufacturers such as TSMC can outperform integrated firms.

For brands planning a campaign

Suitable for supply-chain resilience, industrial M&A, automation, and manufacturing-location briefs. Campaigns should focus on selective ownership and execution risk, not a blanket reshoring-is-back story.

Examples: Source 1

Japanese suppliers hold major shares in Boeing and Airbus programmes: airframes, carbon fibre, actuation, cabin interiors, engines, tyres, landing gear, and batteries.

• The same industrial base appears in space through H-IIA and Hayabusa.
• Lilly says success is invisible while failures, such as the cancelled SpaceJet, shape foreign perception.

Maps the supply chain by layer: structure, materials, systems, cabin, propulsion, and space. Gives programme shares, supplier names, and contract figures.

For brands planning a campaign

Strong for aerospace, defence, advanced materials, and industrial-strategy content aimed at buyers who need supplier intelligence rather than prime-contractor branding. It also supports a counter-narrative on hidden supply-chain power.

Examples: Source 1

The historical form factors ran ahead of intelligence, reliability, and cost curves; those curves are now converging, with Chinese clusters, EV and drone supply chains, new investment, and early factory deployments. Japan is re-entering through Highlanders and a Mitsubishi Motors MOU.

Balances history with current shipment estimates, funding data, company-reported deployment figures, and forecasts from Omdia, Morgan Stanley, Goldman Sachs, and PitchBook.

The post cautions that shipments and choreography do not equal general autonomy, and labels some Japanese production ambitions as targets rather than contracted output.

For brands planning a campaign

Useful for robotics, physical AI, advanced manufacturing, logistics, and care-sector briefs. A campaign can focus on realistic deployment stages and supply-chain readiness, distinguishing demonstrations from scaled autonomy.

Examples: Source 1

The post distinguishes L-band as the layer that survives, carrying sensor telemetry, alerts, and positioning, and Ka-band as the layer that restores evacuation centres and mobile base stations. It cites QZSS, direct-to-satellite services, J-LEO, and 3GPP NTN standards.

Combines disaster case evidence, frequency-band physics, market forecasts, carrier subscriber figures, and policy commitments into a procurement-standard argument.

For brands planning a campaign

Relevant to satellite operators, NTN, telecom infrastructure, disaster-resilience, and public-sector technology campaigns. The angle is operational resilience and procurement, not consumer satellite broadband.

Examples: Source 1

Fibre net additions are falling while self-install 5G fixed wireless access grows faster because Japan is running out of installation engineers.

• All four major carriers are active, and the unmet gaps are enterprise-grade CPE, multi-carrier management, and local 5G private networks.
• Economic-security screening is opening a route for trusted-country suppliers.

Uses market forecasts, carrier product moves, global FWA CPE shipment data, and regulatory analysis to identify routes to market: carrier certification, OEM and ODM partnerships, and non-incumbent operators.

For brands planning a campaign

Good for telecom hardware, private 5G, enterprise connectivity, and Japan market-entry campaigns. A brand can ask for a channel map or procurement-readiness brief instead of a broad telecom trend piece.

Examples: Source 1

Formats range from numbered explainers and multi-part series to supply-chain maps and market-entry notes. Forecasts, company-reported figures, and targets are usually labelled, and each post carries a source note.

The recurring move is to identify a structural constraint, such as scarce operators, engineering latency, supplier risk, or installation capacity, and show how technology or policy changes the economics around it.

Some posts rely on third-party forecasts and company-reported numbers; commissioning brands should ask for source-by-source confidence labels and a distinction between estimates and contracted output.

For brands planning a campaign

Julius Lilly is best used for explainer-led campaigns that need a specific, comparative argument with concrete examples. Briefs should ask for named sources, a clear separation of fact and forecast, and a practical implication for the client's decision.

Examples: Source 1 · Source 2 · Source 3 · Source 4 · Source 5 · Source 6 · Source 7 · Source 8

Yu focuses on infrastructure finance: private credit, fundraising, portfolio strategy and assets such as transmission lines, transport modernisation and Vision 2030 projects. Lilly analyses the technology and industrial capabilities within sectors, rather than the capital structures funding them.

Read original · Original 2

Tohumcu writes from a cloud vendor’s customer-facing position, mixing in event and account material. Lilly writes as an independent analyst, focused on sector-level arguments.

Read original · Original 2

Yan approaches the field as an investor weighing where capital and the data layer will land, framing China, the US and Japan as different layers of one build-out. Lilly explains the industrial shift and where adoption, policy and capability lag.

Read original · Original 2

Their posts put Asia-Pacific and Middle East infrastructure private debt growing from $12.9bn in 2020 to $24.9bn in 2025 and name concrete Japanese needs: more than 30% of bridges and tunnels past 50 years of age, grid-modernisation and digital-connectivity gaps, and private lenders stepping into long-tenor project finance as banks pull back.

• A companion post reports management meetings on a listed alternative asset manager, covering fundraising targets, secondaries, private-equity deployment and valuation.
• For a client asking who funds the next round of Japanese transmission, data-centre or transport assets, and on what terms, this is closer to the question than an operations-led view.

The tradeoff is the other half of a Japan industrial campaign.

• Kit Yu's material sits with allocators and lenders: asset-class sizing, financing structures, valuations and fundraising.
• It does not cover the manufacturing floor, robotics, aerospace or technology procurement, so it cannot support the factory-automation, engineering-data or supplier-mapping angles a Japan industrial campaign usually needs.
• The analysis is sector- and region-wide financing commentary, not detail on Japan manufacturing operations.

Examples: Source 1 · Source 2

A post recounts a week in Tokyo and Nagoya with automotive and manufacturing leaders and draws conclusions a strategist can use: software-defined-vehicle teams splitting what to keep in-house versus partner on; a real compute crunch compounded by the weak yen; AI-partner choice rising to CEO and board level; and physical AI moving from pilot to the factory floor, with the humanoid market framed at roughly $3bn in 2025 toward $15bn+ by 2030.

• It also cites Toyota's record fiscal-year revenue near $323bn and electrified vehicles past half of domestic sales.
• For a client campaign on how Japanese manufacturers are actually making AI and cloud decisions, this is the closest of the alternatives.

The tradeoff is framing and scope.

• The posts carry an explicit technology-vendor identity: the author writes as a cloud-provider executive, using ‘our investment’ and ‘we bring’.
• This is a practitioner and vendor perspective rather than independent sector analysis, which clients comparing sources should weigh.
• Japan appears as one leg of a wider tour (Tokyo and Nagoya one week, Korea flagged for the next), so depth per sector is thinner than a dedicated Japan analyst's.
• For a Japan industrial campaign, ask for additional coverage of connectivity, aerospace and financing.

Examples: Source 1 · Source 2

Posts map China's most-funded embodied-AI companies into a modular chain — components, embodiment, motion and control, foundation models, data and simulation, applications — name the data-and-simulation layer as the biggest white space, and frame Japan's potential as ‘the world's best place for Physical AI to learn physical work’ through its factory knowledge, robotics hardware and dexterous manipulation.

• A second post extends the same investor lens to AI's physical build-out, using Generac's data-centre power backlog (around $1.6bn) and Modine's multi-billion data-centre cooling agreement as examples of industrial beneficiaries.
• For a client weighing embodied-AI and AI-infrastructure investment theses, this supplies the capital-allocation view.

The tradeoff is that this is thesis-and-opportunity analysis rather than Japan industry reporting.

• The Japan material is forward-looking positioning — investment opportunities and a proposed moat — not evidence of what Japanese firms are doing now; the AI-infrastructure examples are US-listed industrials rather than Japanese ones.
• For a Japan operations brief, ask for coverage of connectivity, aerospace and manufacturing operations.
• Trace any figure carried into a client document to the primary sources the posts point to.

Examples: Source 1 · Source 2