The data is unambiguous. Goldman Sachs entered 2026 with a bullish outlook on three Asian currencies—the South Korean won, the Taiwanese dollar, and the Malaysian ringgit—citing AI-driven export surpluses as the structural catalyst. By mid-year, all three were down against the US dollar. The won fell 2.1%, the ringgit dropped 1.8%, and the Taiwanese dollar posted the worst performance among the group, losing 3.05%. The US dollar index had risen nearly 3% over the same period.
The ledger remembers what the market forgets.
As a DeFi security auditor who has spent years stress-testing the assumptions behind protocol incentives, this pattern is painfully familiar. A narrative—AI exports will drive currency appreciation—is adopted by a prestigious institution. The market initially trades on that narrative. Then, an external variable intervenes, and the model breaks. The same thing happens in crypto when a protocol assumes its high APY will attract permanent liquidity, only to find that a shift in market volatility or a competitor's incentive program drains its TVL overnight.
Goldman's framework was built on a simple, elegant logic: AI capital expenditure creates a surge in semiconductor exports from Korea, Taiwan, and Malaysia, generating record current account surpluses. These surpluses, in theory, should push the currencies higher. The data partially supported the premise. Korea's current account surplus was projected to nearly double to around $300 billion, representing 13.9% of GDP. Taiwan's surplus was even more extreme at 25% of GDP. Malaysia was attracting consistent foreign direct investment as part of the China-plus-one supply chain shift. Yet the currencies did not appreciate. They weakened.
I have seen this fracture before. In my early work auditing Compound's interest rate model in 2020, I wrote a Python script that simulated 10,000 random liquidity events. The model assumed that lenders would always be willing to supply at a given rate curve, but my stress tests revealed a theoretical insolvency path under extreme volatility. The market ignored the post-audit gist I published—until a real black swan event later proved the simulation correct. Goldman's model made a similar error: it treated the current account surplus as a deterministic driver, while ignoring the systemic variable that the Federal Reserve's monetary policy exerts on all dollar-denominated assets.
The dollar index rising 3% may not sound dramatic, but for emerging market currencies, it is a force that can swamp even the strongest fundamentals. The won, the ringgit, and the Taiwanese dollar were not falling because their exports were weak. They were falling because global capital was flowing into the US, seeking yield and safety. The capital account overwhelmed the current account. This is a textbook lesson that every macro trader knows, but Goldman's research note, according to the BeInCrypto report, did not adequately weight the Fed's influence.
Stress tests reveal the fractures before the flood.
When I analyzed the Terra/Luna collapse in 2022, I spent 72 hours tracing the exact sequence of oracle manipulations and liquidation logic. The protocol's architecture assumed that demand for the stablecoin would remain stable under all conditions. It did not. Goldman's thesis made a similar assumption about AI investment continuity. The report explicitly stated that the wedge between AI exporters and energy importers would persist "as long as AI investment remains intact." That is a conditional clause that should trigger every auditor's alarm. In smart contracts, we write require() statements to enforce preconditions. If the precondition fails—if AI capex slows—the entire thesis collapses.
And there is a deeper structural concern. The concentration of export earnings in semiconductors creates a single-point-of-failure for these currencies. Korea and Taiwan derive an outsized share of their current account surplus from chips. If the AI investment cycle enters a correction—if hyperscalers like Microsoft, Google, and Meta reduce their capex guidance by even 10%—the impact on currency flows could be immediate and severe. The report itself flagged this risk in its final paragraph: "if capex slows, this distinction matters because a cycle shift could move them in tandem." Yet the bullish positioning on the won, the ringgit, and the Taiwanese dollar was based on the expectation that such a slowdown would not happen.
Simplicity in logic, complexity in execution.
Let me be precise about what the data does not show. The three currencies did not perform equally poorly. The ringgit held up better than the won and the Taiwanese dollar, partly because Malaysia's FDI inflows provide a more diversified base. And the currencies did outperform the energy-importing ASEAN peers—the Thai baht fell 4.48%, the Indonesian rupiah dropped 3.8%, and the Philippine peso weakened 4.2%. So the relative ranking Goldman predicted was correct: AI exporters beat energy importers. The absolute direction was wrong because the tide of the dollar lifted all boats—or rather, dragged all currencies down, but some less than others.
This is where a DeFi auditor's perspective adds value. In crypto, we constantly measure alpha against a base layer. A liquidity pool might generate positive yield relative to ETH, but if ETH itself declines by 50%, the absolute return is negative. The question is not whether the trade generates relative outperformance—it is whether the underlying asset has sufficient structural protection to weather a systemic drawdown. Goldman's trade was a relative-value trade dressed as an absolute-value call. The market forced a reevaluation.
The one outlier in Asia is the renminbi. It strengthened against the dollar by 3.32% in 2026, making it the only Asian currency to gain. Goldman maintained its USD/CNY forecast at 6.50. This anomaly cannot be explained by the AI-export framework; China is the world's largest energy importer. The renminbi's strength is a policy-driven phenomenon—the People's Bank of China has the tools and the determination to manage the exchange rate through fixing, offshore bill issuance, and reserve manipulation. In crypto terms, it is a fully controlled asset with a central oracle that overrides market forces. It is not a free market signal. Using it as a benchmark for Asian FX analysis would be like using USDC as a proxy for stablecoin market health while ignoring that Circle's redemptions are policed by bank compliance.
I recently audited an AI-agent smart contract protocol that allowed autonomous agents to execute trades based on natural language prompts. The vulnerability was a classic injection attack: a carefully phrased input could override the agent's access controls and drain the treasury. The agent's reasoning engine was powerful, but it lacked deterministic verification of its outputs. Goldman's model is not so different. The AI export narrative is the prompt. The output—currency forecasts—looked convincing, but no one built a verification layer to check what would happen if the Fed changed its own prompt (its dot plot). The lesson is that narrative-driven models, whether in macro or in DeFi, need a formal stress-testing layer. Auditors call this "formal verification. " In market analysis, it means running scenarios where every key variable is shocked—not just the ones you like.
Verification precedes value.
What are the actionable signals now? I have been tracking the same set of risk indicators I would use for a protocol audit: capital expenditure guidance from the major US tech firms, semiconductor export data from Korea and Taiwan, the DXY index, and the Renminbi central parity rate. The most important leading indicator is the US tech earnings season. If Amazon, Microsoft, Google, and Meta collectively signal a capex reduction of more than 10% for the next quarter, the currencies that rely on AI exports will lose their final anchor. At that point, the relative outperformance will vanish, and they could fall as hard as the energy-importing peers.
Second, watch the Bank of Korea and the Taiwan central bank. If either cuts rates unexpectedly, it will signal that domestic economic weakness is overriding the export boom. That would be the equivalent of a protocol disabling its own security checks to maintain liquidity—a red flag for any auditor.
Third, do not underestimate the oil price. Brent crude above $120 would punish the Thai baht, the Indonesian rupiah, and the Philippine peso, but it would also indirectly pressure the AI exporters through global demand destruction. A prolonged energy shock could tip the global economy into recession, collapsing the AI capex cycle that Goldman bet on.
The renminbi remains a separate ballgame. The Chinese central bank has the balance sheet to defend a target, but the cost is mounting. If the dollar continues to strengthen, maintaining the 6.50 level would require accelerating reserve drawdowns. History suggests that currency pegs eventually break under sustained pressure, even with the most capable central bank. The difference is that China can impose capital controls—the equivalent of a smart contract that pauses all withdrawals. That kind of emergency stop exists only for the renminbi. No other Asian central bank can legally halt capital outflows. The won, the ringgit, and the Taiwanese dollar must float.
From my perspective as someone who has seen code violate assumptions, the Goldman episode is a warning to the entire crypto ecosystem. We are already seeing similar narratives form in DeFi—projects that claim their revenue model is "AI-driven," "real-world asset-backed," or "sustainable because of staking yields." The same danger applies: a single factor narrative that ignores the base layer volatility of the underlying asset (ETH, BTC, or stablecoin supply). When the Fed shifts, or when a competitor protocol launches a liquidity mining campaign with higher rewards, the TVL exit can be as dramatic as a currency sell-off.
In my 2025 audit of an AI-agent protocol, I discovered that the agent could bypass access control logic through a prompt injection—a simple linguistic tweak. The code assumed the natural language parser was secure. The market making similar assumptions about the macroeconomic "parser" of AI investments. The Federal Reserve is the ultimate prompt. When it changes, the entire context shifts.
Let me be explicit about what I am not saying. I am not arguing that the AI export thesis is wrong. The semiconductor data from Korea and Taiwan continues to show robust demand. Samsung and TSMC are running at high utilization. The current account surpluses are real. But in financial markets, intrinsic value is only one input. The other is the price of the asset relative to the dollar, which is determined by global liquidity conditions. The dollar's dominance is a systemic factor that cannot be hedged with a simple current account model. It requires a portfolio-level approach that accounts for cross-currency correlations, interest rate differentials, and risk appetite.
If I were to construct a trade today based on this analysis, I would consider a relative-value pair: long the Korean won versus short the Thai baht, hedged by a short dollar position or a long yen position (since the yen tends to move inversely to risk). This captures the AI/energy wedge while reducing the dollar beta. But I would size it small, because the next major catalyst could come from either direction. The US election, the Fed's September meeting, and the tech earnings season are all potential turning points.
Chaos is just unverified data.
Goldman's report was not wrong in its relative ranking, but it was wrong in its absolute conviction. The market, as always, had the final word. The ledger does not forget. The data from the first half of 2026 is now recorded on the global book of capital flows. Future analysts will reference it when they build their own models. The question for us, as participants in markets that are even more volatile than Asian currencies—crypto markets—is whether we are willing to stress-test our own narratives before the flood arrives.
I have been an auditor long enough to know that the safest contract is the one that has been formally verified, stress-tested across multiple scenarios, and then subjected to independent review. The same discipline should apply to investment theses. Goldman's thesis had no formal verification. It had no scenario analysis that incorporated a strengthening dollar. It had no independent audit. And it failed.
In crypto, the cost of a failed assumption is a permanent loss of funds. In macro, it is a quarter of negative performance. Both are avoidable. The tool kit exists: quantitative simulation, historical backtesting, and the humility to admit that the most elegant narrative is still just a narrative until it is verified against the one truth that matters—the data.
Formal verification is the only truth in code. And in markets, the only truth is the price.