2026 H-1B Sponsorship Report for Data Science & Machine Learning

Across all 6,312 postings, the distribution is Highly likely: 1,131 / 6,312 (17.9%); Likely: 3,300 / 6,312 (52.3%); Possible: 724 / 6,312 (11.5%); Unclear: 750 / 6,312 (11.9%); Unlikely: 407 / 6,312 (6.4%). Every retained posting has a persisted classification from the same taxonomy used in Openbound.

Highly likely and Likely are useful filters for prioritizing verification. Possible reflects limited supporting evidence; Unclear is an evidence gap. Unlikely can reflect an explicit restriction or other negative evidence. An employer’s historical filings do not override a current posting that rules out sponsorship. Rounded shares may not sum to exactly 100%.

Machine Learning is the largest title-defined group, with 1,702 / 6,312 postings (27.0%). Each label describes title text rather than independently verified duties. Analytics and Business Intelligence are combined; their titles often do not identify a more specific specialty.

407 / 505 Research Scientist postings (80.6%) carry favorable signals, compared with 195 / 365 Data Analysis postings (53.4%). Machine Learning has 1,315 / 1,702 (77.3%), while Data Engineering has 465 / 780 (59.6%). These are different current inventories, not interchangeable career paths.

For groups with at least 20 postings, Machine Learning has 1,315 / 1,702 (77.3%) Highly likely or Likely; Analytics / Business Intelligence has 1,013 / 1,509 (67.1%) Highly likely or Likely; Data Science has 1,036 / 1,451 (71.4%) Highly likely or Likely; Data Engineering has 465 / 780 (59.6%) Highly likely or Likely; Research Scientist has 407 / 505 (80.6%) Highly likely or Likely; Data Analysis has 195 / 365 (53.4%) Highly likely or Likely. These differences describe the current mix of employers and postings. They do not establish that moving into a different specialty would improve a candidate’s hiring outcome.

Subtypes follow a fixed title-rule precedence: machine learning; research scientist; data engineering; data science; data analysis; then analytics or business intelligence. A title that says both machine learning and data science is assigned to Machine Learning once. The broader data cohort follows the existing product’s first-matching occupation family.

This ranking combines current data openings with attributable employer history. It sorts by verified-history openings first, then Highly likely + Likely openings, total cohort openings and employer name. It does not rank historical LCA volume or promise that every listed employer sponsors every data role.

The leading three employers account for 1,066 / 6,312 postings (16.9%) in this cohort. TikTok leads this inventory-with-evidence ranking with 363 current openings, 363 attached to verified history, and 341 carrying favorable signals.

The top 20 include technology platforms, consulting firms, biopharma and financial services employers. That is useful when a search is limited to companies described as AI labs. Deloitte also illustrates why employer history and posting signals must remain separate: 73 / 73 current openings attach to verified history, but only 23 / 73 (31.5%) carry favorable signals.

Verified history requires a verified identity, mapped LCA or USCIS history, and positive recent filings or initial/continuing approvals. Continuing approvals can relate to existing workers. This evidence is employer-level, lagged and not necessarily data-specific; the separate favorable column describes the current opening’s classification.

The recorded primary states with the most openings are CA: 1,847 / 6,312 (29.3%); NY: 690 / 6,312 (10.9%); WA: 509 / 6,312 (8.1%). A primary-state count is useful for navigating the inventory, but it does not capture every eligible worksite of a multi-location or nationwide remote posting.

Each posting contributes to exactly one stored primary-state bucket. 1034 / 6,312 records have no stored primary state and remain in the unspecified bucket. The chart displays the ten largest buckets; expand the table for all states. Before planning a move, check the source’s permitted location and work authorization language.

Unspecified: 4,611 / 6,312 (73.1%); Remote: 994 / 6,312 (15.7%); Hybrid: 544 / 6,312 (8.6%); On-site: 163 / 6,312 (2.6%). Among the 994 Remote postings, 546 (54.9%) carry Highly likely or Likely classifications.

Workplace type is unspecified for 4,611 / 6,312 postings (73.1%). The extraction keeps these unknowns rather than treating a missing remote flag as on-site work. This missing coverage, along with differences in employers and roles, prevents a causal claim about workplace arrangement and sponsorship.

If remote work is essential, combine the workplace filter with a favorable sponsorship classification, then confirm permitted states and the employing entity. A U.S. remote posting can still require existing authorization and can still exclude international or certain state locations.

Choose a search that matches your work: modeling, pipelines, experimentation or reporting. Data Science alone contributes 1,451 openings and Data Engineering 780; expanding to all data titles adds roles with different requirements. Filter for sponsorship evidence after relevance, rather than chasing a specialty’s higher share.

Build an employer shortlist from current relevant openings. For each role, check the source posting’s sponsorship restriction, the employing legal entity and the evidence on the company page. A sponsor’s historical scale is less useful when it has no suitable current opening.

For the 750 Unclear openings, ask a focused recruiter question early: does this specific role consider candidates needing sponsorship for your pathway? Missing evidence does not mean refusal, but resolving it before a long interview loop saves time.

Separate location from immigration support. For Remote openings, confirm permitted U.S. states; for multi-location roles, confirm the actual worksite. This snapshot counts listings, not vacancies, offers or hires.