Tag: NVIDIA

  • The Semiconductor Supply Chain in 2026: Why Chips Are Still a Geopolitical Weapon

    The global semiconductor industry is expected to hit $975 billion in revenue in 2026—a 26 percent increase over 2025, which itself grew 22 percent. The combined market capitalization of the top 10 chip companies reached $9.5 trillion by December 2025, up 181 percent from two years earlier. TSMC introduced the world’s most advanced 2-nanometer chip, promising 10 to 15 percent faster speeds and 20 to 30 percent lower power consumption than its 3-nanometer predecessor. And the United States and China are engaged in a technology control regime that a Texas National Security Review analysis compared, unfavorably, to Cold War-era CoCom—the multilateral export control system that tried and largely failed to prevent the Soviet Union from accessing Western technology.

    The semiconductor supply chain was the most globally integrated industrial system ever built. It is now fragmenting along geopolitical lines, and every major government on earth is treating chip access as a national security priority rather than a commercial one.

    The chokepoints

    The semiconductor supply chain has a concentration problem that makes OPEC look diversified. Three American companies—Nvidia, Qualcomm, and Broadcom—account for over 75 percent of advanced chip design. TSMC in Taiwan manufactures 80 to 90 percent of the world’s sub-7-nanometer chips. Two Korean companies, Samsung and SK Hynix, plus one American company, Micron, produce essentially all the world’s high-bandwidth memory. ASML, a single Dutch company, manufactures the extreme ultraviolet lithography machines that are required to produce chips below 7 nanometers—and ASML is the only company on earth that makes them.

    Each of these chokepoints is a potential geopolitical weapon, and several have already been deployed as one. The U.S. began restricting semiconductor exports to China in October 2022, targeting advanced AI chips and the equipment used to manufacture them. Those controls were tightened in October 2023, again in December 2024, and again in March 2025, when the Trump administration blacklisted dozens of additional Chinese entities. The Biden administration’s January 2025 AI Diffusion Rule proposed a three-tiered global framework that categorized every country on earth by its access to advanced chips—essentially creating a semiconductor caste system aligned with U.S. strategic interests. The Trump administration rescinded parts of that rule but imposed its own restrictions. The Netherlands, under sustained U.S. pressure, restricted ASML’s sales of advanced lithography equipment to China. Japan implemented similar controls on semiconductor manufacturing equipment.

    China responded with its own export controls on critical minerals—gallium, germanium, and other materials essential to chip manufacturing—explicitly leveraging its dominance of the mineral supply chain as a countermeasure. The tit-for-tat is ongoing, escalating, and structurally embedded in both countries’ industrial strategies.

    What the controls actually accomplished

    The honest assessment, three years into the U.S. export control regime, is that the controls disrupted China’s semiconductor industry without stopping it. CSIS analysis found that the restrictions created equipment shortages for Chinese chipmakers, produced severe bottlenecks, limited manufacturing yields, and forced workforce reductions across China’s chip sector. Chinese manufacturing yields for advanced chips reportedly run 30 to 50 percent, compared to over 90 percent for U.S.-allied manufacturers. Huawei’s Ascend 910C AI processor, China’s most advanced domestically produced AI chip, is limited to an estimated 250,000 to 300,000 units in 2026 production, bottlenecked primarily by high-bandwidth memory availability. For comparison, U.S. production of Nvidia B300-equivalent chips reached 3.67 million units in 2025—and each B300 is roughly five times more powerful than a 910C.

    But China adapted faster than the controls’ architects expected. Cut off from ASML’s state-of-the-art EUV lithography machines, China’s Semiconductor Manufacturing International Corporation (SMIC) used older deep ultraviolet machines to produce 7-nanometer and even 5-nanometer chips—behind TSMC’s leading edge of 3 nanometers, but far more advanced than the controls were designed to allow. Huawei reportedly used shell companies to trick TSMC into manufacturing 2 million chiplets for its Ascend 910 processors. China is investing in domestic lithography equipment, recruiting former ASML employees by the thousands, and pursuing alternative chip architectures—including a 2D transistor from Peking University researchers that reportedly operates 40 percent faster than TSMC’s 3-nanometer devices while consuming 10 percent less energy.

    The CSIS report summarized the fundamental problem: chipmaking equipment is heavy, produced in small lots, and hard to smuggle. Chips are tiny, produced by the millions, and easily concealed. Design software can be moved across borders undetected. Export controls can restrict equipment. They struggle to restrict everything else. The Texas National Security Review analysis drew the Cold War parallel explicitly: CoCom did not prevent the Soviet Union from accessing key technologies, and China is a “more adept target” than the USSR was.

    The cost of the controls to the U.S.

    The restriction regime isn’t free for the restrictor. An ITIF economic model estimated that full U.S.-China semiconductor decoupling would cost American chipmakers approximately $77 billion in first-year revenue losses. U.S. semiconductor R&D investment could decrease by 24 percent, or $14 billion. Over 80,000 American semiconductor jobs would be at risk. Korean firms would gain roughly $21 billion of that lost U.S. business; EU firms would pick up $15 billion; Taiwanese firms $14 billion; Japanese firms $12 billion.

    Nvidia has already raised prices on nearly all its AI GPUs—gaming cards up 5 to 10 percent, high-end AI accelerators up 15 percent—citing increased manufacturing costs and tariff impacts. TSMC is considering a 10 percent price increase on advanced wafers. The semiconductor industry was built as a globally interdependent system where each region specialized in what it did best. Breaking that interdependence doesn’t just hurt the target. It raises costs for everyone, reduces R&D reinvestment for the companies leading innovation, and creates market share opportunities for competitors in countries that aren’t implementing controls with the same rigor.

    The geopolitical imperative and the economic imperative are pulling in opposite directions, and no government has figured out how to resolve the tension. Restrict too aggressively and you damage your own industry. Restrict too loosely and you fund your adversary’s military modernization. The U.S. government approved Nvidia to sell H200 AI chips to selected customers in China in December 2025—the same government that had blacklisted dozens of Chinese entities months earlier. The policy is simultaneously hawkish and permissive because the constraints are genuinely contradictory.

    The Taiwan variable

    Underlying all of this is a single geographic fact: the island of Taiwan, 180 kilometers off the Chinese coast, with a population of 24 million, manufactures the overwhelming majority of the world’s most advanced semiconductors. TSMC’s fabrication facilities in Taiwan represent a concentration of strategic capability that has no parallel in any other industry. If those facilities were destroyed, captured, or rendered inoperable by a Chinese military action—or by the threat of one—the global technology supply chain would experience a disruption that would make the COVID-era chip shortage look trivial.

    This is why the U.S. is funding TSMC’s construction of fabrication plants in Arizona under the CHIPS Act. It’s why Japan, the EU, and South Korea are all building or expanding domestic chip manufacturing. The entire reshoring effort is an insurance policy against a Taiwan contingency—and it’s going to take a decade to meaningfully reduce the concentration risk, because building a leading-edge fabrication facility takes three to five years and costs $15 to $20 billion per facility.

    The semiconductor supply chain in 2026 is not a market. It’s a battlefield where the weapons are export controls, lithography machines, rare earth minerals, fabrication capacity, and the strategic ambiguity surrounding a 180-kilometer strait. The $975 billion flowing through it annually isn’t just commerce. It’s the material substrate of AI development, military capability, and economic power for every country on earth—and the fight over who controls it is the defining industrial conflict of the decade.

    We cover the semiconductor supply chain alongside rare earth monopolies, conflict minerals, and the full landscape of critical material geopolitics across our Rare Earth Elements course—including why the most important factory on earth is on an island that one country claims as its own and another has promised to take.

  • Can Robots Replace Nurses? The Realistic Case for Robots in Healthcare

    In 2023, MultiCare Health System in Tacoma, Washington, purchased 14 Moxi robots—five-foot, 300-pound autonomous machines with blinking blue eyes that turned heart-shaped when greeting people—and deployed them across its hospitals to deliver supplies, transport lab samples, and collect soiled linens. The idea was straightforward: nurses spend up to 30 percent of their time on non-value-added tasks, and a robot that handles the fetching and carrying gives that time back for patient care. By early 2025, MultiCare pulled the plug. Nurses described the robots as “annoying” and said they “got in the way.” Hospital administration said the program didn’t make financial sense. Moxi, the robot that was supposed to help solve the nursing shortage, passed peacefully to what the Washington State Nurses Association called “the AI beyond.”

    Meanwhile, at Cedars-Sinai in Los Angeles, three Moxi robots are operating across neurology, orthopedic, and surgical units, and the nursing staff describes them with genuine affection. Nearly 100 Moxi robots currently operate across more than 25 hospital facilities nationwide. Diligent Robotics, Moxi’s creator, was acquired by Serve Robotics in January 2026 and unveiled Moxi 2.0 in October 2025—a next-generation platform with ten times the compute power, built on 1.25 million deliveries of proprietary real-world data. Foxconn’s Nurabot, built with Kawasaki hardware and NVIDIA AI infrastructure, is being piloted in Taiwan and is slated for commercial launch in early 2026, with early results showing a 20 to 30 percent reduction in daily nursing workload. Changi General Hospital in Singapore has more than 80 robots assisting doctors and nurses with everything from administrative work to medication delivery.

    The Moxi story contains both realities simultaneously: in one hospital system, the robot was a $1.5-million failure that nurses wanted gone. In another, it’s a beloved teammate that staff say makes their shifts better. The difference isn’t the technology. It’s implementation, workflow integration, hospital layout, staffing culture, and whether the robot was solving a problem the nurses actually had.

    The shortage the robots are supposed to address

    The U.S. nursing shortage is not speculative. It’s structural, worsening, and quantified in detail. An estimated 200,000 to 450,000 nursing positions are currently vacant. Over 6.5 million healthcare professionals may exit the workforce by 2026, creating a projected shortfall of more than 4 million workers across physicians, nurses, and support staff. In 2024, national RN turnover ran at approximately 16 percent, with more than 287,000 staff RNs leaving their positions and hospitals hiring roughly 385,000 to backfill and grow. Nearly one million registered nurses are over 50, signaling a massive retirement wave. Between 2024 and 2025, more than 65,000 qualified applicants were turned away from nursing programs due to faculty shortages, limited clinical sites, and budget constraints.

    The pipeline is fragile, demand is surging (five of the 20 fastest-growing occupations in the latest BLS statistics are nursing roles), and the burnout driving the exits is self-reinforcing—fewer nurses means higher patient ratios, which means more burnout, which means more exits. One hundred thousand nurses left the profession during the pandemic alone. The nursing shortage is not a problem that can be solved by hiring faster. There aren’t enough nurses being produced, and the ones who exist are leaving.

    This is the context in which robots enter the conversation. Not as a replacement for nurses—no serious roboticist or hospital administrator frames it that way—but as a tool to reduce the non-clinical workload that burns nurses out and pushes them toward the exit.

    What robots actually do in hospitals right now

    The taxonomy of healthcare robots in 2026 is broader than most people realize, and the category “nurse robot” is mostly a media invention. Robots in hospitals today fall into distinct functional classes, and understanding what each does—and doesn’t do—is essential to answering the replacement question.

    Logistics and delivery robots, like Moxi and Nurabot, transport medications, lab specimens, linens, and supplies between departments. They navigate hallways, operate elevators, avoid obstacles, and complete deliveries autonomously. They do not touch patients. They do not make clinical decisions. They are, functionally, autonomous supply carts with better navigation software and the emotional intelligence to wave hello in the hallway. The value proposition is time savings on the walking-and-fetching that consumes a third of a nurse’s shift.

    Surgical robots are the most established category and the least relevant to the nursing question. The da Vinci Surgical System has been in use for over two decades, and roughly three out of four prostate cancer surgeries in the U.S. are now performed using it. But da Vinci doesn’t replace surgeons—it extends their precision. A surgeon operates the robot’s arms through a console. The robot doesn’t make decisions about incision placement or tissue handling. It’s a tool that makes the surgeon more accurate, not a replacement that makes the surgeon unnecessary.

    Pharmacy automation systems dispense, sort, and track medications with higher accuracy than manual processes. These are well-established, relatively uncontroversial, and meaningfully reduce medication errors—one of the leading causes of preventable hospital deaths.

    Companion and therapeutic robots occupy a small but growing niche. Paro, a therapeutic baby harp seal robot developed in Japan, is used in hospitals and nursing homes to provide emotional support for dementia patients. In Scotland, the National Robotarium trialed an ARI robot to assist patients with rehabilitation exercises, addressing physiotherapist shortages. Japan’s AIREC humanoid can reposition patients, cook, and do laundry in aged-care settings—addressing a demographic crisis where the elderly population is growing faster than the workforce that cares for them.

    Disinfection robots became ubiquitous during the pandemic, using UV-C light to sterilize rooms between patients. Telepresence robots allow remote physicians to “visit” patients via a screen-on-wheels, expanding specialist access in rural hospitals.

    What robots cannot do

    The list is long, and it maps almost perfectly onto the things that make nursing a profession rather than a job.

    Clinical assessment—the ability to look at a patient and recognize that something is wrong before the vitals confirm it. The pattern recognition that comes from thousands of patient interactions. The judgment call about whether a change in a patient’s behavior warrants a page to the physician or a note in the chart. The capacity to hold a dying patient’s hand and know when to stop talking and when to say something. The ability to advocate for a patient who can’t advocate for themselves—to push back on a physician’s order, to escalate a concern, to notice the subtle signs of abuse or neglect or depression that don’t appear in any data stream a robot can access.

    Nursing is a knowledge profession built on a foundation of physical tasks, and the physical tasks are the part robots can help with. The knowledge, judgment, empathy, and advocacy are the part they can’t. The Washington State Nurses Association, in its statement about Moxi’s discontinuation at MultiCare, put it simply: “Nurses are, and will always be, MultiCare’s most critical resource.”

    The honest market

    The global medical robotics market was valued at roughly $19 billion in 2025 and is projected to reach $74 billion by 2034—a 16 percent compound annual growth rate. The smart hospital sector hit $72 billion in 2025. Diligent Robotics expects to double its hospital footprint annually and deploy thousands of Moxi units by 2030. These numbers are real. The investment is substantial. The trajectory is clearly toward more robots in more hospitals doing more tasks.

    But the trajectory is also clearly toward robots as teammates, not replacements. Moxi 2.0’s roadmap includes expansion into senior living facilities, where the robot would greet residents by name, remember their preferences, and eventually hold basic conversations. The co-founder of Diligent Robotics describes the goal as “combining useful help with genuine human connection”—which is either a touching aspiration or a fundamental misunderstanding of what human connection actually is, depending on your tolerance for Silicon Valley framing of emotional labor as an engineering problem.

    The realistic near-term future is hybrid: robots handling logistics, pharmacy automation, disinfection, supply transport, and basic monitoring, while nurses handle everything that requires judgment, assessment, empathy, advocacy, and the irreplaceable capacity to be a human being in a room with another human being who is scared, in pain, or dying. The question “can robots replace nurses?” has a definitive answer in 2026: no. The better question—can robots make nursing sustainable as a profession by absorbing the non-clinical workload that’s burning nurses out faster than schools can train new ones?—has a more interesting answer: maybe, if the implementation doesn’t end up like MultiCare, and if the investment goes into solving nurses’ actual problems rather than building photogenic machines that wave hello in the hallway.

    We cover healthcare robotics alongside humanoid manufacturing, autonomous drones, and the full landscape of robots entering human workspaces across our Humanoid Robots & Drones course—including why the robot most likely to change your life won’t look anything like the ones in the movies.