IMF · Artificial intelligence · Global economy
The global economy apparently has an AI dependency now.
At Monday’s IMF–World Bank meetings in Bangkok, IMF Managing Director Kristalina Georgieva said AI infrastructure investment was helping sustain global growth around 3% despite an energy supply shock. Reuters reports that an updated economic outlook is due Tuesday. This is her assessment of AI’s economic contribution, not a controlled experiment measuring how much growth would vanish without the investment. The distinction matters when a compelling account of a growth engine starts sounding like proof that the engine can never stall.
The employment figures need similar care. IMF research published in 2024 estimated that almost 40% of global employment is exposed to AI, rising to about 60% in advanced economies. Exposure does not mean that 40% of workers will lose their jobs. It includes tasks AI may complement as well as replace. The analysis suggests roughly half the exposed jobs in advanced economies could benefit from integration, while the other half could face reduced labor demand. Skills, institutions and the way employers introduce the tools will help determine the outcome.
World Bank President Ajay Banga emphasized smaller AI applications for farmers, health clinics and teachers, and disputed the assumption that entry-level jobs must be destroyed. That perspective leaves room for practical benefits beyond the largest models. Our broader takeaway is that investment and productivity arrive on different schedules: building facilities creates economic activity before every promised efficiency gain materializes. Workers then have to adapt to the systems those facilities support. The industry promised to change everything. The economists have started adjusting their models.
Source: Reuters on Monday’s assessment and differing employment views ↗Source: IMF research on job exposure, published in 2024 ↗
TSMC · Semiconductor manufacturing · AI infrastructure
The company making the AI chips would like to report another record.
TSMC is scheduled to report third-quarter earnings Thursday, October 15. An LSEG SmartEstimate drawn from 16 analysts forecasts net profit of NT$746.3 billion, about US$23.4 billion, up roughly 65% year over year. Reuters says that would be a sixth consecutive quarterly profit record. Those earnings remain a forecast. The company’s third-quarter revenue, by contrast, has already been reported at a record level, up about 50% from a year earlier. A successful sales quarter is established; the anticipated profit record still has to survive the earnings report.
The manufacturing story extends beyond the processor itself. TSMC’s CoWoS, short for Chip-on-Wafer-on-Substrate, integrates processing chips and high-bandwidth memory into a package. TSMC describes that integration as a way to deliver both computing performance and memory bandwidth. An AI accelerator needs to move data as well as perform arithmetic, which makes packaging part of the useful capability rather than a decorative box around it. The supply chain has to deliver a working combination, not simply an impressive isolated component.
That gives the manufacturer a distinctive position in the boom: competing chip designers still need somebody to make and package their products. Rising sales show that the hardware economy is substantial, while an earnings forecast describes what analysts expect that business to deliver next. Neither guarantees that every customer’s eventual AI investment will pay off, nor tells us that frontier intelligence advanced overnight. Record revenue is confirmed. Record profit is expected. The gold rush continues, and the company making the shovels has another report due Thursday.
Source: Reuters on the earnings forecast and record revenue ↗Source: TSMC’s October 8 monthly revenue filing ↗Source: TSMC’s explanation of CoWoS packaging ↗
Microsoft · AI safety · Regulation
Your washing machine apparently has more safety paperwork than a frontier AI model.
Microsoft President Brad Smith is calling for urgent global AI safety rules. In a BBC interview published Monday, reported by Reuters, he argued that AI faces fewer established safety requirements than milk and washing machines. His proposals include independent evaluators, notification to victims when a rogue AI agent breaches their systems, and an emergency brake. The comparison is his argument for stronger oversight, not a jurisdiction-by-jurisdiction inventory of existing law. These are proposals, not a new universal inspection regime already in force.
The engineering question behind the metaphor is more useful than the comparison itself: how do you stop a system that is acting outside its intended boundaries? In our view, a dependable emergency brake should operate independently of the model’s willingness to cooperate. Revocable tool access, restricted environments and controls that can interrupt execution are examples of design approaches; their presence alone does not prove adequate containment. The relevant evidence is whether the complete system actually prevents or stops consequential failures under realistic conditions.
Accountability also reaches beyond the organization running the agent. An operator may know the intended task, while an affected third party encounters the unintended action. Inspection and breach notification could make those failures easier to investigate, provided the rules identify who must report, what evidence is required and who can verify the response. Monday’s interview puts those responsibilities on the agenda without resolving their implementation. The machine may be intelligent. The emergency brake should not need to be. Calling for it is a starting point; demonstrating it is the next job.
Source: Reuters on Smith’s BBC interview and proposed safety rules ↗