The math changed. It did not change gradually—it changed at once, in 2020, when remote-first became a survival policy, and it has not changed back. Four years of data now confirm what labor economists and university placement officers began observing in 2022: the San Francisco premium that structured US AI talent geography for three decades is no longer functioning as a magnetic force. It is functioning as a cost. The evidence is distributed across eight US secondary markets that ENTRA's Remote Issue has covered this month—Philadelphia, Dallas-Fort Worth, Indianapolis, San Diego, Detroit, Charlotte, Minneapolis-Saint Paul, and Portland—each of which now delivers net purchasing power within 5 to 12 percent of a San Francisco hire at 30 to 45 percent lower gross compensation. For employers whose constraint is compute budget and research runway, that delta is not a talent-quality compromise. It is a reallocation of capital from housing subsidies into model training.
The aggregate picture: across eight secondary US AI markets, the average senior ML engineering base salary in Q2 2026 runs approximately $159,000, versus $245,000 in San Francisco, per Robert Half's 2026 Salary Guide city data. That is a 35 percent gross compensation discount. After applying state income tax differentials, cost-of-living adjustments, and housing deltas at each location, the average net purchasing power gap compresses to approximately 7 to 10 percent—a gap that closes almost entirely when employers apply the gross savings to equity grants, profit-sharing, or remote work stipends.
The Compensation Map
The eight markets ENTRA covered in July 2026 span five distinct employer profiles: financial technology and quant AI (Philadelphia, Charlotte), industrial and automotive AI (Detroit, Indianapolis), life sciences AI (San Diego, Minneapolis), enterprise cloud AI (Dallas-Fort Worth, Portland), and defense-adjacent AI (San Diego). Their compensation ranges and housing costs produce a consistent pattern.
| Market | Illus. Sr. ML base | 2BR rent / mo | State + local tax | Illus. net PP | |---|---|---|---|---| | San Francisco (CA) | $245,000 | $5,700 | ~9.3% | ~$118,500 | | Philadelphia (PA) | $185,000 | $2,000 | ~6.80% | ~$120,400 | | San Diego (CA) | $195,000 | $2,900 | ~9.3% | ~$111,900 | | Dallas-Fort Worth (TX) | $160,000 | $1,650 | 0% (TX no income tax) | ~$107,400 | | Minneapolis-Saint Paul (MN) | $160,000 | $1,400 | ~7.85% | ~$104,200 | | Portland (OR) | $165,000 | $2,000 | ~9.9% | ~$104,900 | | Charlotte (NC) | $150,000 | $1,500 | ~5.25% | ~$103,200 | | Indianapolis (IN) | $140,000 | $1,100 | ~3.15% | ~$101,800 | | Detroit (MI) | $130,000 | $1,050 | ~4.25% | ~$97,600 |
(ENTRA illustrative estimates using approximately 15% federal effective rate, reflecting pre-tax retirement contributions and standard deductions. State and local tax rates per respective state revenue authorities, 2026. Housing figures per Zumper, ApartmentAdvisor, and Apartments.com, Q2 2026. Not tax advice.)
San Diego is the partial outlier: California's 9.3 percent effective marginal rate at this income level and high-cost housing in neighborhoods adjacent to Torrey Pines and La Jolla compress the purchasing power advantage versus San Francisco to approximately 6 percent, the narrowest gap in the set. San Diego's inclusion reflects its dominant position in biotech AI, defense AI, and Qualcomm's semiconductor AI cluster—an employer profile that is not replicable in lower-cost markets. The other seven markets deliver net purchasing power within 12 to 18 percent of San Francisco at 29 to 47 percent lower gross compensation.
The practical reading for a 10-person distributed AI team: placing senior ML engineers at these secondary-market salaries generates $86,000 to $115,000 in gross labor cost savings per role versus a San Francisco hire, or $860,000 to $1.15 million across the team annually. At Anthropic's or OpenAI's reported GPU cluster costs of $1,000 to $3,000 per H100 GPU per month, a 10-engineer team savings at these rates funds 300 to 1,100 additional GPU-months per year without a capital raise.
The Employer Calculus
The gross-to-net compression is the mechanism. Employers are not paying secondary-market engineers less and getting less—they are paying less and getting equivalent output, while the difference in gross cost flows to research infrastructure, model training runs, or additional hiring in other skill areas. That logic has been articulated explicitly by at least one frontier AI lab's public communications in 2025; ENTRA declines to characterize the source.
The distributed AI model that Anthropic, OpenAI, Salesforce, Nvidia, Google, Amazon, Microsoft, Intel, and Databricks have each implemented in different forms does not require co-location. It requires bandwidth, async discipline, and occasional in-person summits. The question secondary market employers ask is not "can the engineer do the work remotely" but "can the engineer attend the quarterly planning session." Philadelphia engineers are two hours from New York and four and a half hours from Boston. Portland engineers are three and a half hours from Seattle. Indianapolis engineers are one hour from Chicago. The quarterly travel budget replaces the lease subsidy.
The most sophisticated secondary-market argument is not cost. It is option value. A senior ML engineer placed in Philadelphia at $185,000 in 2026 is a Drexel or Penn graduate who would have left for San Francisco in 2018. The remote role keeps that engineer in a city with a lower cost floor, which means the employer retains the engineer at $200,000 in 2028 rather than losing them to a Bay Area counter-offer at $280,000. Retention pricing in secondary markets is more stable over time because the cost-of-living floor is lower.
The University Supply Chain
Every secondary market in ENTRA's July 2026 briefing series anchors to a university AI pipeline with direct employer-placement relationships. The pipeline architecture is not new—Intel has recruited from Oregon State for three decades, JPMorgan has recruited from Carnegie Mellon and Penn since the 1990s—but the volume and diversity of the output has shifted.
Penn's inaugural Ivy League AI undergraduate degree reached its first full cohort in 2026. Drexel's co-op model produces engineers with applied AI experience before the open-market hiring window opens. Indiana University and Purdue together graduated more than 800 AI-adjacent engineers in 2025-2026. University of Michigan, which anchors the Detroit corridor, produces one of the largest CS graduate cohorts in the Big Ten. The University of Minnesota's CS program feeds Minneapolis-Saint Paul's medical AI and fintech-AI clusters. Oregon State University routes semiconductor-adjacent AI graduates directly into Intel's Hillsboro team.
The aggregate US secondary-market university pipeline—across the eight cities ENTRA covered—produced approximately 4,200 bachelor's and master's-level AI and ML-adjacent graduates in 2025-2026, per ENTRA's composite review of institutional data. That supply does not flow exclusively to local employers. It flows to wherever the best remote offer lands first. Gulf and European employers who source these programs directly—bypassing the Bay Area intermediary step—access this pipeline at graduation rates before the relocate-to-SF premium inflates the hire.
The retention dynamic is the long-running structural question. Campus Philly's graduate retention data shows 50 percent of Philadelphia-area graduates staying post-graduation in 2026. That number likely rises to 55 to 60 percent if remote roles at competitive compensation become the norm for new graduates, per ENTRA's modeling based on comparable trends observed in Austin, Atlanta, and Denver between 2021 and 2024. Secondary cities that anchor remote pipelines compound their own supply: the engineer who stays adds to the talent density that attracts the next employer, which attracts the next round of graduates.
The 2027 Inflection
Three dynamics will determine whether secondary-city dominance in US AI talent geography consolidates or reverses by end of 2027.
Return-to-office mandates. If the six-to-nine-month RTO push that Amazon and several large banks implemented in late 2024 and 2025 extends to frontier AI labs—and there is no current evidence it will—the remote option that makes secondary markets competitive disappears. The talent in secondary markets does not disappear with it; it migrates. The question becomes whether it migrates to San Francisco or to a different secondary market with a local AI anchor employer. Detroit and Philadelphia are better positioned to retain talent through a hypothetical RTO wave than Portland or Indianapolis, given JLR's UK parent and Vanguard's local anchor role, respectively.
Gulf and European employer pipeline entrenchment. Gulf sovereign technology employers—G42, NEOM's technology arm, Saudi Aramco Digital—have begun sourcing US secondary markets in 2026, per ENTRA's market intelligence. If that sourcing institutionalizes into campus recruiting agreements, signing bonus structures, and multi-year remote contracts by 2027, it creates a non-US demand anchor that competes with Bay Area employers for the same graduate supply. That competition does not require secondary cities to have local offices or local funding ecosystems—it requires airport infrastructure and timezone overlap, both of which secondary cities have.
San Francisco's counter-move. The Bay Area has not stood still. San Francisco's median two-bedroom rent has stabilized at approximately $5,700 in mid-2026 after a post-2022 correction, but the city's employer-provided housing programs, equity compensation structures, and frontier AI lab density have adapted to post-remote realities. Anthropic, OpenAI, and Google DeepMind retain SF presence as their primary research anchor for a reason: early-stage model research benefits from in-person density in ways that product AI engineering does not. The secondary-market thesis wins on product AI, applied ML, and AI infrastructure engineering. It does not yet win on frontier research, where co-location and serendipitous hallway collaboration remain underweighted factors in the remote-work literature.
The 2027 signal to watch is not headcount announcements. It is where the 2027 graduating class of Penn, Michigan, Indiana, Oregon State, and UT Austin sends its top AI decile. In 2022, that cohort went to San Francisco or stayed home to remote-work for a Bay Area employer. In 2026, it goes to Philadelphia or Portland or Indianapolis to remote-work for a Gulf sovereign fund or a London AI team. By 2027, the choice set may include a Coventry automotive AI company, a Vilnius fintech platform, or a Colombo Gulf corridor employer—all competing for the same Purdue CS graduate at rates that would have been unthinkable in 2019.
The map is not done moving.
Robert Half 2026 Salary Guide senior ML engineering base compensation ranges for all eight markets reviewed by ENTRA, June-July 2026. San Francisco two-bedroom rent ($5,700/month) from Apartments.com June 2026 Rent Report. Philadelphia two-bedroom rent ($2,000/month) from Zumper, July 2026. San Diego, Dallas, Minneapolis, Portland, Charlotte, Indianapolis, Detroit two-bedroom rent figures from Zumper and ApartmentAdvisor, Q2-Q3 2026. State income tax rates per respective state revenue authorities for tax year 2026: California (marginal rate approximately 9.3% at this income level, per Franchise Tax Board), Texas (0%, no state income tax), Pennsylvania (3.07% state + 3.73% Philadelphia city wage tax = 6.80% combined for residents), Minnesota (7.85% marginal at this income, per MN DOR), Oregon (9.9% flat above $125,000, per Oregon DOR), North Carolina (5.25% flat, per NC DOR), Indiana (3.15% flat, per IN DOR), Michigan (4.25% flat, per MI DOR). GPU cluster cost estimate ($1,000-$3,000/H100/month) reflects publicly reported training infrastructure pricing from cloud providers, Q2 2026; not confirmed by specific labs. Penn AI degree programs (Ivy League BSE in AI, MSE in AI via Raj and Neera Singh Program) confirmed per previous ENTRA reporting and original sources. Drexel co-op program size and graduation projections per previous ENTRA reporting. Campus Philly graduate retention data (50%) per Campus Philly regional retention report as cited in ENTRA's July 2026 Philadelphia briefing. Aggregate secondary-market university pipeline estimate (approximately 4,200 AI-adjacent graduates, 2025-2026) reflects ENTRA's composite review of institutional data; not confirmed by individual institutions. Gulf employer secondary-market sourcing (G42, NEOM technology arm, Saudi Aramco Digital) per ENTRA market intelligence, Q2 2026; individual employer activity not confirmed.
For the full city-by-city briefings underlying this analysis, see the ENTRA Remote Issue series: Philadelphia · Portland · Dallas-Fort Worth · Indianapolis · Detroit · Charlotte · Minneapolis-Saint Paul · San Diego. For the European parallel: how Vilnius and Stockholm are absorbing the same talent redistribution dynamic, see Gulf Remote AI Talent Architecture 2026.
Global Career PlatformFind AI talent. Find your next role.
Booking is hotels. · Airbnb is apartments. · ENTRA is global careers.