TSMC: The $0 Bet That Changed Computing
In the summer of 1997, a four-year-old semiconductor startup with fewer than sixty employees and a bank account drifting toward zero received an unannounced telephone call. Jensen Huang, the company’s founder, had written a letter requesting a meeting with Morris Chang, the founder of Taiwan Semiconductor Manufacturing Company. Huang proposed a collaboration that seemed, by any conventional standard, disproportionate: his failing firm, NVIDIA, would design graphics chips; TSMC would manufacture them. Huang included a prediction. The chip would not merely save his company from bankruptcy. Within two or three years, it would make NVIDIA a major TSMC customer.
Chang called Huang directly. He asked questions. He was struck by the clarity of Huang’s optimism, the coherence of his technical vision. Then he agreed. The decision was not a calculation of immediate revenue, the volumes were trivial, but a recognition that TSMC’s business model required exactly this kind of asymmetric bet. While competitors manufactured for the present, TSMC would manufacture for a future that did not yet exist. The call established a pattern that would define the company for the next two decades: an institutional willingness to commit resources before the market validated the commitment, sustained by a founder’s personal conviction that process technology, when pushed relentlessly forward, created its own demand.
This is the central mechanism of TSMC’s economic engine. The company captures value not merely by owning fabs, but by occupying a temporal position. Through sustained research and development spending fixed at eight percent of revenue, non-negotiable through booms and busts, TSMC creates a two-to-three-year lead on each new process node. This lead generates pricing power because the yield maturity of a node is not a commodity. A 28-nanometer process that functions reliably is worth exponentially more than one that does not, and TSMC’s organizational discipline ensures it reaches yield first. The volume compounds. Each successful node captures the next wave of demand, smartphones, then artificial intelligence, while the capital intensity of the business, running to six billion dollars annually by 2010, erects barriers that competitors cannot scale without matching both the R&D and the capital expenditure simultaneously. Most cannot. They cut R&D during downturns. They lay off engineers when margins compress. TSMC does neither.
The durability of this advantage rests on a set of policies that Morris Chang institutionalized early, derived from scars accumulated during his previous career. At Texas Instruments, where he had risen to head worldwide semiconductor operations, Chang had repeatedly requested to raise R&D spending from 4.8 percent of revenue to 5.5 percent. He was denied every time. When he founded TSMC, he set the floor at eight percent and made it invariant. He told the board and his executives that this was not a budget line but a covenant. The percentage remained fixed even when revenue collapsed during the 2008 financial crisis. This was not financial conservatism; it was financial aggression. By maintaining R&D while competitors retreated, TSMC ensured that when demand returned, it alone possessed the process technology to meet it.
Yet the model contains a structural constraint that shapes every major capital allocation decision. To maintain the two-to-three-year lead, TSMC must commit to massive capital expenditures for new nodes before demand is proven. The bet must be placed while the outcome remains uncertain. In 2010, as the global economy trembled, Chang proposed tripling capital spending to nearly six billion dollars, concentrated on the 28-nanometer node. The board revolted. Directors called the general counsel, demanding an audience with Chang before the formal meeting. They argued that the company could not sustain such outflows while revenue declined. Chang listened. Then he told them: “I heard you, but I am still the guy that’s responsible for the operation of the company. You need to let me go ahead with this one.”
The board relented. Chang had identified 28-nanometer as the “sweet spot”: he node where power efficiency and cost would align to capture the emerging smartphone market. He later quoted Shakespeare to explain the logic: “There is a tide in the affairs of men which, taken at the flood, leads on to fortune.” The metaphor reveals the economic reality of semiconductor manufacturing. If TSMC had waited for market validation, waiting for smartphone demand to appear in the financials before building capacity, the tide would have passed. Competitors would have captured the node. The “sweet spot,” as one R&D director described it, was like a tennis racket: you had to hit the ball precisely there, or lose the point. Missing the node would not merely mean idle capacity; it would mean ceding the process leadership that defined TSMC’s entire value proposition.
This tension, between committing billions to unproven demand versus preserving cash for validation, never resolves. It is the permanent condition of the business. The 28-nanometer bet succeeded, capturing the smartphone boom and generating returns that funded the next node. But the risk is existential. If Chang’s business development director had misread the market, if smartphones had not become ubiquitous, if AI demand had not materialized for the subsequent nodes, the six billion dollars would have become a stranded asset, and the board’s fears would have materialized as a liquidity crisis.
What allowed Chang to override the board was not merely ownership structure or formal authority, but a specific form of leadership capital accumulated over decades. He possessed the credibility to enforce counter-cyclical discipline because he had founded the company, survived previous cycles, and demonstrated that his bets on process technology eventually converted to revenue. This leadership functioned as an economic variable in the strict sense. Without Chang’s authority, the eight percent R&D covenant would have broken under shareholder pressure during the 2009 crisis. The six billion dollar capex line would have been vetoed. The moat would have collapsed.
The economic value of this leadership manifested most clearly in moments of organizational crisis. In 2009, with the financial crisis eroding revenue, TSMC’s then-CEO (Chang had stepped back to chairman, a distinction that proved meaningless when operational decisions conflicted with his principles) laid off six to seven hundred employees based on performance ratings. The layoffs occurred without board approval, circumventing Chang’s explicit directive that any workforce reduction required board review, which he would deny. The dismissed workers appeared at Chang’s home at dawn to protest. His wife brought them breakfast at six in the morning. Chang reversed the terminations. He rehired the workers.
The decision appears, on a spreadsheet, as an indefensible subsidy of fixed costs during a revenue contraction. It was precisely the opposite. In semiconductor manufacturing, process leadership does not reside in patents alone, but in the accumulated, tacit knowledge of yield engineering: he specific adjustments required to move a node from functional to mature. When competitors laid off engineers during downturns, they lost the specific individuals who had solved the yield problems of the previous node. Those engineers carried the debugging knowledge that prevented the next node from repeating the same failures. By retaining his workforce, Chang preserved an institutional memory that competitors could not replicate through hiring. The “no layoff” policy created a talent moat. Competitors who cut heads found, when demand returned, that they had amputated their capacity to learn.
This organizational stability intersected with Chang’s personal relationships with customer CEOs to create a switching cost that no contract could duplicate. When the 2009 crisis deepened, TSMC faced a hundred-million-dollar dispute with NVIDIA over yield and quality issues on the 40-nanometer node. The previous CEO had refused compensation, asserting TSMC bore no fault. The relationship fractured. NVIDIA considered arbitration, potentially switching fabs. Chang resumed active management of the CEO role, a signal in itself that the matter required his personal attention, and spent four to five weeks negotiating directly with Huang. He calculated a number, north of one hundred million dollars, and gave Huang a forty-eight-hour deadline to accept or proceed to arbitration. Huang accepted within two days.
The economics of the settlement were complex. Paying too much would signal weakness, inviting other customers to demand similar concessions. Refusing to settle would have cost TSMC its largest customer at that node and the volume required to mature the process. Chang’s intervention resolved the dispute through trust rather than contract. Huang accepted the terms because Chang offered them, and Chang’s track record, beginning with that 1997 phone call to a failing startup, had established that his commitments extended beyond quarterly results. The relationship reduced the transaction costs of dispute resolution and secured capacity commitments that market mechanisms could not enforce. Competitors could not buy these relationships; they had to be earned through decades of execution that demonstrated TSMC would absorb pain to preserve partnerships.
These personal relationships and organizational policies constitute an underappreciated structural advantage. Most analyses of TSMC focus on capital intensity, economies of scale, or technological sophistication. These are necessary but not sufficient. The moat is maintained by the intersection of fixed R&D spending, workforce continuity, and founder-level customer diplomacy, echanisms that require a specific type of governance to sustain. Chang’s authority allowed him to treat R&D as a fixed cost and human capital as an appreciating asset when conventional corporate governance would have optimized both for short-term margin preservation.
Yet the model remains fragile. The most significant risk is the possibility of a missed node bet. If Chang’s R&D organization had failed to identify 28-nanometer as the sweet spot, if the market had shifted to a different geometry or if demand had fragmented across multiple nodes, the six billion dollar commitment would have left TSMC with idle capacity and a board convinced that Chang’s judgment was unsound. A single major error in node selection would break the credibility required to enforce the eight percent R&D covenant or the no-layoff policy. The board, already restive in 2010, would have installed a cost-cutting successor. The organizational culture that generates process leadership would unravel within two business cycles.
A second fragility lies in customer concentration. TSMC’s economics depend on a small number of firms, NVIDIA, Apple, Qualcomm, consuming massive volumes at the leading edge. If any of these customers diversifies to competing fabs or builds internal manufacturing capacity, TSMC loses the volume that justifies its capital intensity. The 2009 NVIDIA dispute illustrated how proximate this risk is. The previous CEO’s refusal to compensate NVIDIA nearly triggered a defection that would have degraded TSMC’s yield learning curve and revenue base. Chang’s settlement preserved the relationship, but the dependency remains. These customers tolerate single-sourcing because of Chang’s personal relationships and TSMC’s technical lead. If either factor erodes, if a new CEO at Apple or NVIDIA lacks history with Chang, or if a competitor achieves process parity, the switching costs dissolve.
The third fragility is the sustainability of Moore’s Law itself. The entire economic model assumes that semiconductor complexity will continue doubling every two to three years, requiring new process nodes that command premium pricing. If physical limits or economic constraints slow this progression, TSMC’s R&D and capex strategy becomes uneconomical. The eight percent spending would generate diminishing returns; the six billion dollar bets would produce capacity for improvements that customers no longer value sufficiently to pay premiums. The foundry model works because the next node is always sufficiently better to justify migration. If that stops, TSMC becomes a commodity manufacturer with unbearable fixed costs.
These fragilities expose the implicit assumption underlying Chang’s strategy: that the board and shareholders will continue to accept counter-cyclical discipline even during prolonged underperformance. The 2009 board rebellion over capex, and the forced layoff reversal, demonstrated that this acceptance is contingent on results. Chang’s authority derived from a track record of correct bets. Had the 28-nanometer node failed to capture the smartphone transition, the institutional culture he built, the fixed R&D, the employment guarantees, would have appeared as stubbornness rather than wisdom. The moat is maintained by success; it is not structurally self-sustaining.
Chang’s management of the 28-nanometer centralization decision further illustrates how governance structure affects economic returns. Early in the node’s development, TSMC had split its organization into advanced and mainstream groups with separate profit and loss statements. Chang reversed this, centralizing all leading-edge manufacturing under a single structure. The move removed internal competition that might have generated parallel innovation paths, but it prevented the dilution of resources and political friction that had slowed the 40-nanometer ramp. The decision treated the foundry not as a portfolio of businesses but as a single technical undertaking requiring unified focus. Separate P&Ls would have optimized individual group profitability at the expense of node yield maturity. Centralization sacrificed internal market efficiency for organizational coherence: trade only a founder with Chang’s authority could enforce against internal resistance.
The Shakespearean tide that Chang referenced in his autobiography was not merely poetic license. It described the irreversibility of node transitions in semiconductor markets. Once a process node becomes the industry standard for a given application, 28-nanometer for smartphones, 7-nanometer for AI inference, the volume concentrates at the leader. Customers design their architectures around that node’s specific characteristics. Switching to a second-source fab requires redesigning chips for different process variations, delaying time-to-market by quarters or years. TSMC captures this volume because it reaches yield maturity first, and it reaches yield maturity first because it never stopped investing when others retrenched. The tide flows in one direction.
But tides also recede. The semiconductor industry has witnessed the collapse of seemingly permanent technical leads before. Chang’s strategy requires that each generation of leadership maintain the discipline he institutionalized, even as the founder’s personal authority dissipates. The relationships with Huang and Apple’s leadership must survive executive transitions at those companies. The board must continue to accept eight percent R&D and no layoffs during the next downturn, even if the previous node was slightly late or slightly over budget. The assumption is that the organizational culture Chang built has become self-sustaining, that the policies have hardened into infrastructure.
This is the wager. TSMC’s value derives from being the inevitable choice for leading-edge manufacturing, a position achieved through capital commitments made in defiance of present demand and employment policies maintained in defiance of present margins. It is a business built on the conviction that process technology, when pursued with sufficient discipline and capital, creates its own market. The model works as long as the physics of silicon continues to reward density, as long as customers remain concentrated enough to fund the next node, and as long as the institution retains the nerve to bet billions before the bet is obviously safe. When Chang placed that call to Jensen Huang in 1997, he was not merely cultivating a customer. He was demonstrating that TSMC’s business was time itself, anufacturing the future before anyone else could prove it existed.
The irreversibility Chang described operates not merely at the market level but within the physics of the fabrication process itself. When TSMC committed to the 28-nanometer node, it was not simply betting on smartphone demand; it was betting that the node would exhibit the specific electrical characteristics, leakage current, gate control, power scaling, that would make it the terminal point for planar transistors before the industry transitioned to FinFET architectures. If the physics had proven recalcitrant, if the node had suffered from excessive leakage or variability, the six billion dollars would have purchased not capacity but a lesson in physical limits. The “sweet spot” was therefore a technical prediction as much as a market one. The tennis racket metaphor invoked by Chang’s R&D director implied a precision that belied the underlying uncertainty: the industry had to hit the node exactly when the economics of silicon scaling intersected with the power constraints of mobile devices. Miss that intersection by six months, and the foundry would be manufacturing high-performance chips for a market that had already moved to lower-power alternatives, or low-power chips for applications that demanded performance TSMC could no longer deliver cost-effectively.
This temporal specificity explains why Chang reversed the organizational split between advanced and mainstream manufacturing groups. The separation had seemed logical under standard corporate finance theory: distinct P&Ls would drive competition, allowing the advanced group to pursue leading-edge geometries while the mainstream group optimized mature nodes for cost. But in semiconductor manufacturing, the distinction between “advanced” and “mainstream” is fluid and dangerous. A node transitions from experimental to commercial not through a discrete handoff but through a continuous process of yield learning that requires the same engineering teams to shepherd designs from pilot production to high-volume manufacturing. By splitting the organization, TSMC had created a structural incentive for the advanced group to prioritize technical milestones over manufacturability, while the mainstream group lacked the technical depth to solve the yield problems that determined whether the node achieved economic viability. The political friction manifested in delayed decisions about resource allocation, tooling purchases, personnel assignments, customer prioritization, that consumed weeks when the market window was measured in months. Chang’s centralization removed these veto points. He treated the foundry as a single technical system where the velocity of learning, not the optimization of individual group margins, determined returns. The economic variable was not the efficiency of the organizational chart but the speed at which the institution could convert R&D expenditure into yield maturity.
The cost of this centralization was the elimination of redundancy. When the organization pursued parallel paths, competing internal groups exploring different process variations, failure of one approach did not doom the node. Under Chang’s unified structure, TSMC bet on a single technical architecture for each node. If that architecture encountered fundamental physical barriers, there was no fallback. The centralization worked because Chang’s R&D organization possessed sufficient technical judgment to select the correct path, and because the fixed eight percent spending had accumulated enough process knowledge to avoid blind alleys. But this concentration of technical risk remains a hidden fragility. A competitor with a more distributed R&D structure might stumble upon a process variation that solves a critical yield issue, while TSMC’s unified approach commits fully to a solution that proves non-viable late in the development cycle. The organizational efficiency gains come at the cost of optionality.
Chang’s willingness to accept this concentration of risk derived from his experience at Texas Instruments, where he had learned that distributed, cautious R&D produced mediocrity rather than safety. During his tenure heading worldwide semiconductor operations at TI in the early 1970s, Chang had repeatedly requested authorization to increase R&D spending from 4.8 percent of revenue to 5.5 percent. The requests were denied every time by a finance organization that viewed research as a variable cost to be minimized against current profitability. The denial was not malicious; it was the rational outcome of a corporate governance structure that prioritized quarterly earnings and risk-adjusted returns on capital. Chang later described this period as formative, noting that the refusal to fund marginal R&D increases represented “the best thing that we did for R&D”, ot because the denial was correct, but because it taught him that the finance function, when given veto power over technical investment, systematically undervalues the compounding returns of process leadership. When he established TSMC, he inverted this relationship. He told the R&D director that the eight percent floor was non-negotiable not because he possessed superior forecasting ability, but because he understood that the alternative, variable R&D subject to quarterly review, would produce a competitor that always arrived two years late to each node.
The economic logic of this fixation becomes clearer when examining the specific mechanics of yield improvement. In semiconductor manufacturing, yield, the percentage of functional dies per wafer, follows a learning curve that is neither smooth nor predictable. Each new node introduces novel physical phenomena: line-edge roughness at extreme ultraviolet wavelengths, variability in gate oxide thickness, electromigration in copper interconnects. Solving these problems requires iterative experimentation that consumes wafers, chemicals, and engineering hours without immediate revenue. A competitor that cuts R&D during a downturn does not merely slow its development schedule; it interrupts the learning curve. The engineers who were debugging the previous node’s contact resistance problems are reassigned or dismissed. The tacit knowledge, specifically, which process parameters actually controlled the variation, dissipates. When demand returns, the competitor must reconstruct this knowledge, but the engineers who possessed it have moved to firms that did not lay them off. TSMC’s no-layoff policy, maintained through the 2009 crisis when revenue collapsed and the previous CEO attempted to dismiss six to seven hundred workers, preserves this learning curve as an organizational asset. The workers Chang rehired were not generic labor; they were the specific individuals who had solved the 40-nanometer yield issues that had nearly destroyed the NVIDIA relationship. Their retention ensured that when TSMC began ramping 28-nanometer production, it did not repeat the defects that had plagued the previous node.
The specificity of this knowledge explains why the settlement with NVIDIA exceeded one hundred million dollars yet represented rational economic behavior. When Chang took back the CEO role in 2009 to resolve the dispute, he was not merely buying NVIDIA’s continued patronage; he was purchasing the forensic data that NVIDIA possessed regarding the 40-nanometer failures. NVIDIA had designed its chips to TSMC’s process specifications, and the yield shortfalls represented a mismatch between design assumptions and physical reality. By settling rather than proceeding to arbitration, Chang gained access to NVIDIA’s failure analysis, which revealed which process modules, implant energy, anneal temperature, etch chemistry, were actually causing the defects. This information was worth more than the settlement amount because it accelerated TSMC’s learning curve for the subsequent node. The alternative, winning in arbitration while losing the customer, would have saved cash but destroyed the information exchange that made TSMC’s process technology improve. The relationship with Huang was therefore not merely social; it was an information channel that reduced the cost of yield learning. When Huang accepted Chang’s forty-eight-hour ultimatum, he was acknowledging that the information asymmetry favored continued collaboration: NVIDIA needed TSMC’s future capacity more than it needed the settlement proceeds, and TSMC needed NVIDIA’s diagnostic data more than it needed legal vindication.
Yet the durability of these information channels depends on continuity at both organizations. The fragility that most analyses underweight is the dependency on specific individuals at customer firms. Chang’s relationship with Huang began in 1997 and was tested in 2009; it survived because both founders remained in control of their respective companies. But semiconductor firms experience leadership transitions, and the personal trust that permitted rapid dispute resolution does not automatically transfer to successor executives. If NVIDIA or Apple were to install CEOs who viewed supplier relationships through purely transactional lenses, evaluating TSMC solely on price and delivery rather than technical collaboration, the switching costs that currently bind these customers to TSMC would diminish. The foundry model assumes that customers cannot easily replicate TSMC’s process knowledge elsewhere; it implicitly assumes that customers will not attempt to develop that knowledge internally or distribute it across multiple foundries to reduce single-source risk. Intel’s periodic attempts to enter the foundry business, however unsuccessful, represent a persistent threat that customers might subsidize alternative capacity if TSMC’s pricing or delivery becomes unacceptable. The concentration of TSMC’s revenue in a handful of firms creates a bargaining asymmetry that will intensify as these customers grow larger and more sophisticated in their supply chain management.
The bear case for TSMC therefore rests on the convergence of three trends: the slowing of Moore’s Law, the maturation of customer diversification strategies, and the eventual departure of Chang’s governing authority. The physics of silicon scaling approaches fundamental limits. As transistors shrink below five nanometers, quantum tunneling and power density create thermal constraints that node shrinks no longer solve cost-effectively. If the industry transitions to three-dimensional packaging or specialized architectures rather than pure geometry scaling, TSMC’s advantage in process node leadership becomes less relevant. The eight percent R&D spending produces diminishing returns if the next node offers only marginal improvements over the previous one. Customers might then extend the lifecycle of mature nodes, reducing the volume that justifies TSMC’s massive capital expenditures. The foundry model works when the transition from 28-nanometer to 16-nanometer to 7-nanometer delivers discrete improvements in power and performance that compel designers to migrate. If those improvements become incremental, the economic pressure to remain at the leading edge dissipates, and TSMC’s capital intensity becomes a burden rather than a moat.
This physical constraint intersects with the organizational fragility of Chang’s institutional culture. The policies that define TSMC’s competitive advantage, the fixed R&D percentage, the employment guarantees, the centralized technical authority, depend on a governance structure that accepts lower short-term margins in exchange for long-term technical leadership. Chang implemented these policies through personal authority derived from his founding role and his track record of correct bets. When he told the board in 2010 that they must accept the six billion dollar capex increase despite their objections, he could do so because he had accumulated sufficient credibility reserves to override their risk aversion. A successor CEO, lacking this founding authority and facing activist shareholders or a board accustomed to quarterly performance metrics, might find the eight percent R&D floor unsustainable during the next industry downturn. The policy is durable only as long as the CEO possesses the political capital to defend it against finance organizations that will inevitably argue for variable spending tied to revenue. Once the policy breaks, once R&D becomes a variable cost optimized for current margins, the two-to-three-year lead begins to compress. Competitors who maintained their R&D through the downturn (or who benefit from state subsidies that insulate them from market pressure) catch up. The yield advantage disappears. The pricing power evaporates.
The employment policy faces similar erosion risk. A successor CEO, facing a prolonged recession and board pressure for “operational efficiency,” might classify workforce reductions as “performance management” rather than layoffs, circumventing Chang’s prohibition exactly as the 2009 CEO attempted. The six to seven hundred workers Chang rehired represented specific institutional knowledge; their dismissal would not merely reduce headcount but would amputate the debugging capability for the current node. Yet this economic reality is obscured by accounting conventions that treat labor as a variable cost. Without Chang’s explicit intervention, his personal appearance at dawn to speak with protesters, his reversal of terminations without board consultation, a future CEO might optimize the income statement while destroying the balance sheet of human capital.
The most acute risk, however, remains the temporal mismatch between capital commitment and demand validation. The 28-nanometer bet succeeded because smartphones created a massive, synchronous demand for exactly that geometry’s power-performance characteristics. But semiconductor markets fragment. Artificial intelligence workloads may demand specialized architectures, chiplets, analog computing, optical interconnects, that do not map cleanly to traditional node scaling. If the next wave of computing diverges from the path of pure CMOS scaling, TSMC’s six billion dollar bets on advanced lithography may purchase capacity for a market that never materializes. Chang’s Shakespearean tide assumes that the flood always arrives, hat there is always a next node that captures the next platform. But tides are tidal; they recede as well as advance. The semiconductor industry has experienced periods of architectural consolidation before, during which leading-edge capacity became commoditized and margins collapsed.
Chang’s strategy assumes that TSMC can perpetually outrun commoditization through technical leadership. But technical leadership in a slowing Moore’s Law regime becomes a matter of incremental optimization rather than generational breakthroughs. In such an environment, the capital intensity that currently excludes competitors becomes a millstone. The fabs built for six billion dollars cannot be repurposed; they are specific to particular process technologies. If demand fragments across multiple specialized geometries rather than consolidating on a single sweet-spot node, TSMC’s centralized, high-volume model faces competition from smaller foundries serving niche markets with lower capital requirements. The economics of the foundry business invert: flexibility becomes more valuable than scale, and the fixed-cost structure of massive leading-edge fabs becomes a liability.
This possibility remains distant but not theoretical. The transition from planar to FinFET transistors at the 16-nanometer node required TSMC to master new device physics; the transition to gate-all-around transistors at three nanometers presents similar challenges. Each transition consumes more R&D resources for diminishing improvements in performance per watt. At some point, the economics of node migration fail: the cost of designing chips for the new node exceeds the value of the performance improvement. When that point arrives, TSMC’s customers will extend their stay at mature nodes, and the volume that justifies the eight percent R&D and six billion dollar capex will dissipate. Chang’s model works only if the tide continues to rise; it has no contingency for a high-water mark.
The question that will determine TSMC’s longevity is whether the institutional culture Chang built can adapt to a world where the tide recedes, here process leadership means cost optimization at mature nodes rather than relentless pursuit of the next geometry. The policies that served TSMC during the smartphone boom, centralized authority, fixed R&D, employment guarantees, may prove maladaptive in a fragmented market requiring product diversity rather than process uniformity. Chang’s genius was recognizing that semiconductor manufacturing is a business of time, of capturing the flood at its peak. But floods eventually reach their limit. The test of the institution he built will be its ability to survive the ebb.