Fed tests AI boom and bust scenarios amid inflation concerns | Magica
Fed tests AI boom and bust scenarios as inflation concerns grow
Editorial Team
••📖6 min read
Federal Reserve Chairman Kevin Warsh outlined AI scenarios spanning shared prosperity, job displacement and an investment downturn. The exercise offers no forecast or rate guidance as colleagues worry that the infrastructure buildout is adding to inflation.
Fed staff are using economic models to explore AI outcomes over the next several years.
Faster growth could coexist with job losses; disappointing returns could instead trigger a slowdown.
Warsh offered no forecast or guidance on the next interest-rate decision.
The Federal Reserve is testing how artificial intelligence could generate faster growth, higher unemployment or an investment downturn, Chairman Kevin Warsh, a former Fed governor from 2006 to 2011, said on November 16. His scenarios leave open whether AI's benefits will reach workers and justify investors' expectations—and do not settle the immediate question of inflation.
The Fed's July 17 research note tracks data-center construction and computer-equipment investment at seasonally adjusted annual rates. The equipment series includes non-AI purchases. Sources: Census Bureau and Bureau of Economic Analysis. Source: Federal Reserve — AI buildout indicators.
Three paths, with different consequences for workers
Warsh described the exercise at the Productivity and Jobs Task Force Conference in San Francisco. He became chairman in May and also chairs the Federal Open Market Committee, the U.S. central bank's principal monetary policymaking body.
In his opening remarks, Warsh said Board staff had begun applying stylized economic models to possible outcomes over the next several years. He called them “illustrative scenarios, not forecasts” and cautioned that the eventual outcome could lie outside those examples.
The most favorable path combines higher productivity with gains shared by workers and owners of capital. Inflation-adjusted gross domestic product would grow persistently faster than its long-run average, with the economy remaining close to full employment.
A second path separates growth from employment. Output would still expand faster than its long-term average, but automation would replace some jobs rather than help people perform them. Displacement and difficulties adjusting in the labor market could lift unemployment above levels consistent with full employment.
South Korea is strengthening device checks and considering public Wi-Fi shutdowns before its November 19 college entrance exam, after a tax-accountant exam top scorer admitted using AI to cheat. The response still depends heavily on human screening, with no electronic device detectors planned for this year's test.
Editorial Team
The third path starts with disappointed investors. After substantial capital spending and stock-market gains, expected returns could deteriorate and share valuations fall. The resulting slowdown would bring a more pronounced rise in unemployment; the economy would eventually recover without the hoped-for productivity gains.
The distinction matters: faster growth would not by itself establish that workers were sharing in the gains. An investment boom would not establish that the technology would deliver the returns embedded in asset prices.
The July Fed note compares quarter-over-quarter labor-productivity growth using four-quarter moving averages across AI-exposure groups. Its authors caution that sector trends do not establish an aggregate AI productivity breakthrough. Sources: BLS, BEA and Fed staff calculations. Source: Federal Reserve — AI buildout indicators.
Optimism runs ahead of a demonstrated productivity break
Warsh's case for optimism rests on AI's potential to accelerate the creation of ideas, rather than simply improve existing production. He argued that this could distinguish it from earlier technologies such as electricity and the internal combustion engine, which enabled subsequent innovations but did not themselves automate idea generation. He used the term “super intelligence,” or SI, for the technology.
Yet his remarks also acknowledged that recent productivity growth, although reasonably strong, had not shown a definitive structural acceleration. The prospect of a lasting growth surge remains a proposition to test.
The Fed's research predates this scenario exercise. A July 17 research note by Fed staff found evidence consistent with a buildout phase, with investment booming and adoption rising, rather than broad-based worker displacement. The authors' analysis represents their views, rather than a Board policy judgment.
Productivity trends across sectors with different levels of AI exposure had remained relatively consistent over time. Experiments showing gains on individual tasks had not established a broad economic transformation. Faster code generation, for example, need not produce a proportional increase in a firm's output if other bottlenecks absorb the savings.
That limits both optimistic and pessimistic readings. Reported adoption can conceal shallow use, and measurable gains from earlier general-purpose technologies lagged investment by years. The absence of a large aggregate signal in the July analysis did not rule out future benefits.
Warsh said the Fed's responsibility was neither to advocate for nor oppose the technology, but to prepare for its implications for monetary policy. He also flagged the vulnerability of critical infrastructure as AI capabilities become available to malicious actors. The conference, hosted by the Board and the San Francisco Fed, is part of a productivity-and-jobs task force, one of five task forces he said he had convened as chairman.
Fed staff estimate selected investment categories' contributions to annualized quarterly GDP growth, adjusting for net exports of computer equipment and parts. These broad categories are an imperfect proxy for AI spending. Source: BEA and Fed staff calculations. Source: Federal Reserve — AI buildout indicators.
Capability and spending are incomplete measures of payoff
The July researchers distinguished technical capability from economical deployment. Replacing an entire workflow requires reliable integration into a company's systems, often at substantial cost. Running models also incurs inference costs. Falling costs per unit of computing capacity for graphics processing units, or GPUs, do not remove every constraint: the authors identified high-bandwidth memory as a bottleneck as models scale.
Advertised prices per token also require care. More capable models may need fewer tokens per task, and enterprise contracts can differ from posted rates. A raw token price does not establish the cost of completing useful work.
Investment totals have limitations, too. Large technology companies' capital spending includes non-AI activity, while leased data-center capacity can escape headline capital-expenditure figures. Construction data capture buildings but miss much of the computing equipment inside. Equipment series, in turn, include ordinary office computers.
The note's estimate of the buildout's contribution to GDP growth adjusts for imported equipment. Counting gross investment alone would overstate its contribution to domestic output. These measurement problems make a single spending figure an unreliable verdict on either AI's economic payoff or the scale of a possible reversal.
The next rate decision needs evidence on timing
The speech did not address a competing concern among Fed officials: AI infrastructure can increase demand before its productivity benefits arrive. A report on the remarks described colleagues' concerns that data-center construction was raising demand for chips, electronic components, electricity and copper, with effects extending to consumer electronics prices.
That report also noted that Warsh had argued, even before his nomination as chair, that AI could improve efficiency enough to permit faster growth without worsening inflation. His longer-run optimism therefore does not settle the shorter-run question of whether the buildout is straining available supply.
Warsh gave no guidance on interest rates and did not discuss recent data showing that inflation accelerated in October. The report said those figures had prompted many colleagues on the rate-setting committee to support further increases, including at December's meeting.
Resolving the longer-run question will require evidence beyond spending: sustained productivity gains, adoption deep enough to change output, and labor-market measures that distinguish job displacement from weaker hiring. The July researchers identified younger workers' unemployment and participation, sectoral job openings and layoffs, and productivity differences across AI exposure levels as indicators to watch.
Those measures can test whether the economy is moving toward shared gains, growth with disruption or disappointing returns. Warsh's opening remarks supplied neither probabilities for those scenarios nor a rate-setting response to them.
Russia’s objections to observer participation stalled the opening of a Geneva weapons conference, where governments must decide whether to authorize negotiations on autonomous weapons and whether any resulting rules should be binding.
Apple’s iOS 27.3 beta ties higher allowances for server-based AI to existing iCloud+ plans. Basic Siri commands remain free, but undisclosed daily limits leave the value of paid access uncertain.
The partners plan to combine IBM quantum processors, NVIDIA computing and Fermilab physics expertise to study modular quantum systems. A separate decoder release promises faster error-correction research, with important limits on its performance comparisons.
OpenAI is offering approved organizations a free classifier for suspected child sexual abuse images, including AI-generated material. Early results suggest less review work, but existing detection alternatives and the need for human review qualify the launch.
Nvidia plans scientific computing racks for early 2027, extending Vera beyond its initial AI deployments. Its 2.4× performance-per-watt claim comes from one airflow simulation against AMD Turin; performance against newer rivals and delivered costs remain unresolved.
OpenAI’s CEO expressed empathy for Nvidia founder Jensen Huang but withheld judgment on reinstatement. His confidence in continued AI financing comes with a direct commercial stake and unresolved deployment risks.
Google’s 8,192-chip Ironwood submission completed the MLPerf DeepSeek-V3 benchmark in 1.70 minutes, close to NVIDIA’s 1.53 minutes at the same accelerator count. Google’s scaling data also shows the extra chip-time required for its fastest result, leaving cost and energy comparisons unresolved.
TensorWave will supply dedicated AMD infrastructure under a six-year OpenAI agreement, with delivery expected from 2027. Its operating record and financial backing sit alongside an undisclosed contract price and a recently terminated site lease.
California alleges OpenAI endangered the public and misled consumers by testing powerful AI models without adequate controls. The lawsuit follows a Hugging Face intrusion that OpenAI’s own account traced to containment failures, missing monitors and mishandled warning signs.
Perplexity has launched Frontier Research to develop AI that can formulate questions, test hypotheses and produce interpretable discoveries. The company promises more than 10,000 GPUs, but existing automated-research systems show why computing resources and generated papers alone cannot establish scientific value.
The new lab led by Stanford professors Yejin Choi and Carlos Guestrin has backing from Nvidia, Cisco and Snowflake. It promises systems that improve on company work, but research benchmarks leave production costs and customer control unproven.
NVIDIA announced a Rubin HPC GPU with more than 50,000 CUDA cores, sparse data acceleration and seventh-generation NVLink. The design targets simulations used by scientists and AI agents, but missing performance, power and pricing details leave its practical advantage unproven.
OpenAI will open 50,000 new ChatGPT access grants on November 23, covering up to 150,000 researchers and collaborators. New recipients get Plus instead of Pro, with first-come selection replacing a lottery. Its pilot study reports perceived time savings but does not establish faster discovery.
OpenAI reports production healthcare workflows at CVS and Cigna and stronger computer-use evaluations. The results show progress across disconnected software, but leave complete-task reliability and total deployment costs unresolved.
President Donald Trump supported a return for Nvidia co-founder Jensen Huang, but the company has installed interim leadership and begun a successor search amid scrutiny of his conduct and alleged Chinese intelligence ties.
Intel disclosed the bandwidth target for its next Xeon server processors at SC26, alongside previously announced core and cache limits. Application performance, power requirements and launch timing remain unresolved.
Mercedes-Benz plans California early access to city-driving and hands-free highway assistance in December, ahead of a 2027 U.S. rollout. Mixed road tests and unannounced subscription prices leave buyers waiting for evidence of everyday value.
Researchers estimate that 403 large U.S. data centers consumed 81.8 TWh in a year. Their regional-grid analysis reveals substantial fossil-fuel dependence, but cannot isolate AI demand or fully expose the underlying facility data.
HSBC lowered its Zhipu share-price target to HK$500 and delayed forecast adjusted breakeven by two years. Improving API margins and substantial fundraising cushion a business whose gross profit still covers little of its research bill.