Goldman Sachs' 2026 quantitative engineering graduate band in London opened at £72K base with a first-year discretionary bonus floor of £35K — a total-comp floor of approximately £107K (~$135K) before any signing or equity supplement. For the most competed-for CS and ML graduates in the current cycle, the package ceiling, with discretionary bonus at or above target, clears £125K (~$158K) in year-one total cash. JPMorgan's AI risk and model validation graduate track, housed in its Canary Wharf technology function, is pricing similarly: £68K–£76K base with a first-year bonus range of £28K–£45K, yielding a year-one total-comp picture of £96K–£121K (~$121K–$153K). Neither figure appeared in the 2024 ENTRA graduate survey. Both are active in the 2026 cycle.
The comparison that matters is not Goldman versus JPMorgan — it is Goldman versus DeepMind. DeepMind's Research Scientist Graduate Programme pays £82K–£88K base with a Google RSU grant of £60K–£95K over four years, producing a first-year total-comp figure of £97K–£115K. The headline figure is competitive. The structure is not: DeepMind's year-one cash take-home is £82K–£88K base plus benefits; Goldman's year-one cash take-home is £107K–£125K. For a 2026 CS graduate weighing a King's Cross research role against a Canary Wharf engineering role, the cash-in-year-one differential runs in finance's favour by £15K–£30K — and that differential is what London's AI labs are now navigating.
What Finance Is Actually Paying
The Goldman Sachs figure requires decomposition, because the quantitative engineering band is not the whole Goldman picture. The bank's 2026 London graduate intake spans three AI-adjacent functions with meaningfully different comp profiles.
The Quantitative Engineering track — the highest-paying — recruits CS and ML graduates directly into Goldman's Marcus and FICC technology divisions. The 2026 base for this track sits at £70K–£75K, with a target first-year discretionary bonus of £35K–£50K for graduates who join structured products and algorithmic execution teams. The ceiling, for the most competed-for profiles, runs above that: candidate-side conversations tracked by this bureau through Q1 2026 include one documented offer of £75K base plus a £50K first-year bonus guarantee, aggregating to £125K (~$158K) year-one cash. Goldman's standard graduate benefits package adds a further £8K–£12K in pension, private medical, and lifestyle benefits, bringing the effective total compensation above £130K for this profile.
The AI Risk and Model Validation track at JPMorgan — housed in the Chief Data and Analytics Office's London function — runs at a slightly lower base but comparable bonus structure. The 2026 new-grad band for this track opens at £68K base with a target first-year discretionary bonus of £28K–£38K, yielding a year-one cash figure of £96K–£114K. The JPMorgan Model Risk Technology graduate programme, which feeds into the bank's SR 11-7-compliant model validation function, is specifically recruiting ML graduates with transformer familiarity and risk-domain interest — a profile that, two years ago, would have defaulted to a data science role at Monzo or a research track at DeepMind. In 2026, JPMorgan's risk-tech graduate programme is actively competing for it.
Revolut occupies a distinct position in this analysis. Unlike Goldman and JPMorgan, Revolut is not an investment bank — it is a digital banking entity that received its UK banking licence from the FCA in July 2025 after a nearly three-year review. Its inclusion here is deliberate: the comp repositioning Revolut has executed since the FCA authorisation has moved it into a different competitive tier than Monzo or Wise. Revolut's 2026 new-grad ML engineering band, updated as part of the post-FCA internal compensation review confirmed by two people familiar with the process, now runs at £75K–£85K base — a lift from the £68K–£75K range reported in the May 9 ENTRA briefing and a figure that now sits at or above ElevenLabs' Voice Research Residency base of £75K. With discretionary year-one performance awards that Revolut has reinstated for its technology graduate cohort, the total first-year figure clears £90K–£100K.
Why Finance Is Winning This Cohort Now
Three structural factors are driving the 2026 shift — and none of them are the story that circulated in 2022, when Goldman "going AI" was treated as a branding exercise rather than a structural talent move.
First: the research-to-production gap has closed inside finance. Goldman's ML Engineering function within its Platform Solutions and Global Markets divisions is, in 2026, running production ML at a scale and technical ambition that was not present in 2022. The bank's internally developed large language model infrastructure — used for earnings summary generation, risk report automation, and client-facing Alpha Intelligence tooling — requires the same transformer-optimization, inference-pipeline, and fine-tuning skills that DeepMind and ElevenLabs recruit for. A 2026 Imperial MSc AI graduate whose thesis touched RLHF or model quantization is a genuine fit for Goldman's AI Engineering team, not a mismatch corrected by two years of on-the-job finance domain learning. Goldman's technical interview format has reflected this: the firm added a machine learning systems design round to its quantitative engineering interview loop in the 2025–26 recruitment cycle, per two candidates who completed the process. The signal is deliberate.
Second: the bonus structure front-loads cash in a way that RSU grants cannot. DeepMind's Google RSU grants are real and meaningful — £60K–£95K vesting over four years — but the year-one cash equivalent of a £20K–£23K RSU instalment does not feel like £125K on a 28 February payslip. Goldman's year-one discretionary bonus, paid in January following the calendar-year performance cycle, arrives eleven months after joining as a single cash payment. For a 2026 graduate carrying student debt, renting in Zone 2, and making a first mortgage calculation, the temporal structure of finance comp is a material advantage. The AI labs understand this — DeepMind's benefits team introduced an accelerated RSU vesting option for specific graduate profiles in the current cycle, per one candidate-side account — but the structural cash-timing advantage of finance has not been neutralised.
Third: the Skilled Worker visa infrastructure at Goldman and JPMorgan is faster. Goldman Sachs is one of the oldest and most operationally sophisticated Tier 2 sponsors on the Home Office register — its immigration function has processed Skilled Worker applications for hundreds of technologists annually since the visa's introduction, and its average certificate-of-sponsorship issuance timeline for graduate new-starters runs at four to six weeks from offer acceptance. JPMorgan's Canary Wharf immigration team operates similarly. For an international CS or ML graduate — a cohort that, at Imperial College London, represents approximately 43 percent of the 2026 MEng cohort, per ENTRA's analysis of published student composition data — Goldman's established visa processing pipeline is a material operational advantage over an AI lab building its Skilled Worker infrastructure more recently. ElevenLabs' Skilled Worker sponsor licence, confirmed via the Home Office Tier 2 register and active as of May 2026, was established in 2024. Goldman's has been active continuously for more than fifteen years.
The salary threshold context: every role discussed here clears the Skilled Worker £38,700 minimum floor by a factor of two to three. The operative immigration variable is not whether the offer clears the threshold — it does — but how quickly the bank's compliance function can move a certificate of sponsorship. At Goldman and JPMorgan, the answer is weeks. At a Series C AI company processing its first cohort of international graduates, the answer may be months.
The Lab Counter-Moves
DeepMind, ElevenLabs, and Wayve are not watching the comp shift passively. Each has made a specific structural move in the 2026 cycle in response to finance's escalation — and the moves are different enough to be worth mapping separately.
DeepMind triggered its discretionary top-of-band process for a larger share of 2026 Research Engineer candidates than in any prior cycle, per two people familiar with its graduate programme management. The Research Engineer track — at £75K–£82K base, lower than the Research Scientist band — is the one most directly exposed to Goldman's quantitative engineering competition, because it recruits MEng and MSc graduates rather than PhD holders, and the overlap between Goldman's target profile and DeepMind's Research Engineer profile is substantial. The discretionary response has involved accelerated RSU grants — moving the standard four-year vest to a three-year vest for competed-for candidates — and, in documented cases, a one-time transition payment described internally as a "research commitment supplement." The mechanism is unusual in DeepMind's graduate history and reflects how seriously the labs are treating the finance escalation.
ElevenLabs is competing on speed and equity narrative rather than on cash matching. The company's counter-offer to a 2026 Imperial ML graduate holding a Goldman quantitative engineering offer is not "we match the cash" — it cannot, at pre-IPO economics. It is "you are holding pre-IPO options at a $3.3B strike price on a company at an $11B Series D valuation, and you are making this decision before the IPO is announced." The equity argument is real: EMI options issued to ElevenLabs' January 2026 Voice Research Residency intake are, at the current implied secondary-market valuation, already in-the-money by a factor of three on the grant-date strike. Whether a 2026 graduate believes that trajectory continues through an IPO is the psychological crux — and Mati Staniszewski's recruiter presence on LinkedIn, combined with the company's publicly disclosed revenue trajectory crossing $500M ARR in Q1 2026, has made the argument more rather than less credible through the spring cycle.
Wayve has explicitly expanded its Skilled Worker sponsorship infrastructure for the 2026 cohort — a direct response to the immigration-processing-speed argument that Goldman's recruiters are using. The company's 2026 new-grad processing timeline, per one person familiar with Wayve's HR immigration function, has been reduced from a twelve-to-fourteen-week average in 2025 to an eight-to-ten-week target in 2026, achieved by pre-certifying immigration counsel review for the most common graduate visa profiles (Indian and Chinese nationals on Tier 4 transitioning to Skilled Worker post-graduation). The move is calibrated: Wayve pays £78K–£90K base for its Autonomous Systems Engineering Track, below Goldman's year-one cash floor, but the visa-processing differential was a documented friction point in 2025 international graduate offer acceptances, and reducing it is a prerequisite for competing.
What the Numbers Say About Direction
ENTRA's Q1 2026 recruiter survey — nine London ML and data science agencies — captures offer-acceptance patterns for the current cycle. Among CS and ML graduates who received simultaneous offers from a Tier 1 investment bank (Goldman, JPMorgan, Morgan Stanley) and a UK AI lab (DeepMind, ElevenLabs, Wayve), the finance offer was accepted approximately 31 percent of the time — up from an estimated 19 percent in the equivalent 2024 data. The share of graduates choosing finance over labs is not a majority. But it has grown by more than half in two years, and the direction of movement is unambiguous.
The profile accepting the finance offer has also shifted. In 2024, the finance-accepting cohort in this crossover population skewed toward graduates with structured-products or quantitative finance domain interest — people for whom Goldman was always the plan and the AI lab offer was an exploratory hedge. In 2026, the recruiter survey captures a more evenly split cohort: roughly half domain-preference (Goldman was always the target), roughly half pure-compensation decision (the AI lab was the preference, the Goldman cash differential closed it). The second half is the cohort the labs are losing on economics alone — and it is the one that DeepMind's discretionary top-of-band process, ElevenLabs' equity narrative, and Wayve's visa-speed improvement are all designed to address.
Cambridge attribution in this dataset carries a specific note: the crossover population — graduates holding simultaneous finance and AI-lab offers — is concentrated at Imperial College London and UCL, not Cambridge. Cambridge ML PhD and MPhil graduates, whose research profiles route more naturally to DeepMind's Research Scientist track than Goldman's quantitative engineering track, appear in this crossover population at lower rates. The finance-vs-labs competition is, at the Cambridge ML PhD level, less relevant. At the Imperial MEng and UCL MSc level — the Research Engineer and quantitative engineering pipeline — it is the defining competition of the 2026 cycle.
What's Next
Goldman's 2027 London graduate salary review is already under internal discussion, per one source familiar with the firm's compensation committee process. The bank's talent strategy for its AI Engineering function — which it has positioned publicly as a primary growth area for the London technology organisation — requires sustained comp competitiveness with the AI labs in the 2027 and 2028 cycles. A base uplift to £78K–£82K, with a corresponding bonus floor increase, is a realistic outcome of that review.
The AI labs' collective response will be visible by September 2026, when the first offers for the 2027 intake go out. DeepMind's Research Engineer band — the one directly exposed to Goldman competition — has the most room to move within the Google RSU framework, and the accelerated vest experiment in the current cycle is likely a precursor to a structural band uplift. ElevenLabs' response will depend on its IPO timeline: if the company files in the 2026–27 window, the equity argument becomes self-executing. Wayve's response is operational — visa speed and post-Series-C option pool depth — rather than purely compensatory.
For a 2026 Imperial MEng or UCL MSc graduate sitting with a Goldman quantitative engineering offer and a DeepMind Research Engineer offer on the same desk, the decision in May 2026 is genuinely ambiguous in a way it was not in 2023. Goldman's year-one cash advantage is real and substantial. DeepMind's research identity and four-year RSU trajectory are also real. The cohort that has historically resolved that ambiguity in the labs' favour is, this year, resolving it differently — and at a 31 percent finance-acceptance rate among crossover candidates, London's AI labs are paying attention.
Compensation data sourced from candidate-side conversations, recruiter agency surveys (Q1 2026, nine London ML and data science agencies), and ENTRA Job Signal Index tracking. Goldman Sachs and JPMorgan declined to comment on graduate compensation bands. ElevenLabs confirmed its Voice Research Residency structure. Wayve confirmed its Skilled Worker sponsorship processing improvements. Goldman Sachs Tier 2 sponsor history confirmed via Home Office register. ElevenLabs Skilled Worker sponsor status confirmed via Home Office Tier 2 register, May 2026. Imperial College London MEng cohort composition data per ENTRA analysis of published university statistics; not independently confirmed by Imperial.
For the fintech-vs-AI-labs comp war, see Fintech's Graduate War With Big AI: London 2026. For DeepMind's Research Engineer track, see Google DeepMind 2026 Graduate Intake: 65 UK Positions Decoded. For the ElevenLabs versus DeepMind Cambridge PhD competition, see ElevenLabs vs DeepMind: How Cambridge ML PhDs Are Choosing in 2026.