Lumen — Talent Intelligence

AI inference infra — Fireworks vs Baseten vs Together vs ModalVC Firm 1

Analyst

Compare Fireworks, Baseten, Together, and Modal on the workforce signals. What's distinctive about each of the four?

LinkedIn Talent Data Insights

On the workforce metrics, Modal is the smallest but most engineer-pure and most retentive. Baseten is growing fastest. Together is biggest but most stretched. Fireworks sits in the middle on most things — with the highest retention risk.

Headline numbers (May 2026 capture).

- Modal: 147 employees, +101% YoY (73 → 147), 10% attrition, 90 hires. Smallest by far. Lowest attrition in the quartet. Engineering 50% of workforce — most engineer-pure.
- Fireworks AI: 207 employees, +92% YoY (108 → 207), 22% attrition, 146 hires, 27 open jobs. Mid-pack on everything. SF Bay (110) heavy. Engineering 43%.
- Baseten: 257 employees, +195% YoY (87 → 257), 17% attrition, 201 hires, 112 open jobs. Fastest-growing of the quartet — nearly tripling. Highest open-jobs count, by a lot. Engineering 36% (lowest — more sales + HR build).
- Together AI: 351 employees, +50% YoY (234 → 351), 23% attrition, 195 hires, 54 open jobs. Biggest. Slowest-growing in % terms but on a larger base. Highest attrition. Engineering 43%, with a notable Netherlands cluster.

What the data says about each.

Modal Labs is the most defensible-looking on these signals. NYC-headquartered (72 of 147), with a distinctive Stockholm cluster (11) tracking founder Erik Bernhardsson's Spotify pedigree — Spotify is in fact Modal's #2 top past employer (16 employees). 50% Engineering function is the highest in the quartet. 10% attrition is the lowest by 7 percentage points. Lowest LinkedIn Recruiter Open-to-Work share at 14%.

Fireworks AI is the most middle-of-the-pack. 92% YoY growth, 22% attrition, broadly distributed hires from AWS, Meta, Amazon, TikTok. Engineering 43%. Most concerning signal: highest Open-to-Work share in the quartet at 24% (50 of 210 captured profiles), and 28 of those are flagged Active Talent (actively searching).

Baseten is the operational scaler. +195% headcount in 12 months. The growing-skills list is dominated by ramp signals — Bash +1,000%, Reliability Engineering +750%, Application Deployment +700%, Recruiting +700% — the last one is the tell: they are scaling their own recruiting function to keep up. 112 open jobs is more than the other three combined. 41 of 268 captured profiles are Active Talent (15%) — the highest concentration of actively searching candidates in the quartet despite a moderate 17% attrition.

Together AI is the most lateral-hire-heavy and most stretched. 165 Senior-level employees against 114 Entry-level is the most senior-skewed mix in the quartet — they've been around longer (founded 2022) and pulled more from established companies (Microsoft 34 past, Apple 25, Amazon 23 — these are unusually large past-employer counts for a 351-person company). 23% attrition is the highest. Notable Netherlands cluster (24 in The Randstad) — only company in the quartet with a meaningful European base.

The verdict on workforce signals. Modal looks most defensible (small, retentive, engineer-pure, founder-pulled talent). Together looks most established but most stretched (biggest, highest attrition, senior-heavy). Baseten is the operational scaler (fastest growth, most open jobs, scaling its own recruiting infrastructure). Fireworks is mid-pack everywhere with the highest forward retention risk.

Workforce snapshot — May 2026 capture

Metric	Fireworks	Baseten	Together	Modal
Employees (LinkedIn)	207	257	351	147
Headcount growth YoY	+92%	+195%	+50%	+101%
Employees one year ago	108	87	234	73
Attrition rate	22%	17%	23%	10%
Hires (12mo)	146	201	195	90
Departures (12mo)	35	29	66	11
Net change (12mo)	+111	+172	+129	+79
Open jobs	27	112	54	—
Engineering %	43%	36%	43%	50%
HQ	San Mateo	San Francisco	San Francisco	New York

Headcount — Together biggest, Modal smallest

351

257

207

147

Together

Baseten

Fireworks

Modal

Headcount growth YoY % — Baseten nearly tripling

195%

101%

92%

50%

Baseten

Modal

Fireworks

Together

Attrition rate — Modal half the rate of the rest

23%

22%

17%

10%

Together

Fireworks

Baseten

Modal

Open jobs — Baseten hiring 2x harder than the rest combined

112

Baseten

Together

Fireworks

Modal

How valuable were this section's answers?

click 👍 / 👎 to leave feedback

Analyst

Walk me through the talent flows. Where is each company sourcing talent from, and are any of them trading with each other?

LinkedIn Talent Data Insights

Fireworks AI — top 12mo flows

Hiring From

Amazon Web Services (AWS)

Meta

Amazon

Vercel

TikTok

Google

Hathora

MongoDB

Grafana Labs

Snowflake

Fireworks AI

Departures To

OpenAI

Nebius

OpenRouter

Stealth AI Startup

Baseten — top 12mo flows

Hiring From

Parsed

IMC Trading

Databricks

dbt Labs

Microsoft

Meta

Sigma

Google

Apple

Vellum

Baseten

Departures To

Factory

Google

Apple

OpenAI

Together AI

Datadog

Together AI — top 12mo flows

Hiring From

Meta

Amazon

Amazon Web Services (AWS)

SingleStore

ByteDance

Stanford University

Instacart

Cloudera

GitHub

Together AI

Departures To

NVIDIA

Meta

Google DeepMind

Google

AMD

NinjaTrader

Anthropic

FriendliAI

Amazon

Temporal Technologies

Modal — top 12mo flows

Hiring From

Twirl

Applied Intuition

Together AI

Databricks

Vercel

Amazon

Citadel

Butter

Robinhood

FLORA

Modal

Departures To

Cursor

The departures cut sharpens the picture: Together AI is losing senior talent to AI competitors (NVIDIA, Anthropic, Google DeepMind, AMD, FriendliAI). Fireworks is losing a concentrated cohort to OpenAI. Modal has only 1 captured departure across the entire 12mo window — the most retentive in the quartet.

The departures view (12mo, all destinations captured).

- Fireworks AI: OpenAI -4, Nebius -3, OpenRouter -2, Stealth AI Startup -1 (offset by +1 hire). The OpenAI loss is the most coherent signal — a small but pointed migration to OpenAI's inference org. Nebius pulls a European-AI-compute crew.
- Baseten: Factory -3 (small AI startup), Google -1, Apple -1, OpenAI -1, Together AI -1, Datadog -1. Diffuse — six destinations, no concentration. Six total captured departures on a 257-person base.
- Together AI: NVIDIA -4, Anthropic -2, Google DeepMind -2, AMD -2, FriendliAI -2, NinjaTrader -2, Meta -2 (offset by +5 hires), Google -2, Amazon -1, Temporal -1. Ten destinations captured. The NVIDIA/Anthropic/DeepMind/AMD/FriendliAI cluster is the headline — Together is losing senior talent to the most aggressive AI/chip competitors in the market.
- Modal: Cursor -1. Only 1 departure captured in the LinkedIn 12mo data. Consistent with the lowest attrition in the quartet (10%) and the smallest base (147 employees). Most retentive workforce in the comp set on this signal.

Direct intra-quartet trades (12mo captured). Modal pulled 3 hires from Together AI; Baseten lost 1 to Together AI. Other pairs show no captured trades. Almost no movement between these four directly — they are not yet a natural ecosystem for talent migration.

The hires cut — top 10 sources per company.

- Fireworks pulls broadly across hyperscaler/big-tech. Top 12mo sources: AWS 6, Meta 6, Amazon 4, Vercel 3, TikTok 3, Google 3, Hathora 3, MongoDB 2, Grafana Labs 2, Snowflake 2. Pure FAANG/hyperscaler funnel.
- Baseten pulls from a more idiosyncratic mix. Top 12mo sources: Parsed 7 (small-company sweep — Parsed is a ~20-person company), IMC Trading 5 (HFT — quantitative engineering talent), Databricks 4, dbt Labs 3, Microsoft 3, Meta 3, Sigma 3, Google 2, Apple 2, Vellum 2. The Parsed and IMC patterns read as targeted recruiting, not broad funnel.
- Together pulls the broadest mix of senior FAANG and infra talent. Top 12mo sources: Meta 5, Amazon 5, AWS 5, SingleStore 5 (small-company sweep — possibly acqui-hire), ByteDance 4, Stanford University 4, Instacart 4, Cloudera 3, LinkedIn 3, GitHub 3. Three hyperscalers tied at 5; the SingleStore cluster is the most distinctive single-source concentration.
- Modal pulls from a narrow lateral mix. Top 12mo sources: Twirl 4 (small AI company sweep), Applied Intuition 3, Together AI 3, Databricks 3, Vercel 2, Amazon 2, Citadel 2, Butter 2, Robinhood 2, FLORA 2. Very few hires from FAANG directly. Most come from other small AI / fintech / infrastructure companies — consistent with Modal's founder-recruited model.

The past-employer (pedigree) view. Looking at all current employees' past employers — a longer-arc signal than 12mo flows:

- Fireworks employee pedigree: Google 24, AWS 20, Meta 17, Microsoft 16, Facebook 12. Pure FAANG.
- Baseten employee pedigree: Microsoft 17, Google 17, Twitter 7, Sigma 5. Microsoft + Google heavy.
- Together employee pedigree: Microsoft 34, Apple 25, Amazon 23, Google 21, Topsy 12. Heavy hyperscaler/FAANG mix.
- Modal employee pedigree: Spotify 16, Google 10, Microsoft 7, Better 7. The Spotify count is the founder signal — Erik Bernhardsson was Spotify’s head of ML / founder of Annoy, and pulled a meaningful chunk of his Spotify network into Modal.

The pattern. Fireworks and Together look like "raid the FAANG" companies on the inflow side. Modal looks like a "founder-network" company. Baseten is in between. On the outflow side, only Together is leaking to direct AI competitors at scale — the data reads as senior-talent extraction by NVIDIA, Anthropic, and Google DeepMind, three of the most active recruiters in the AI infrastructure market.

Past-employer pedigree — top sources of current employees (longer-arc signal than 12mo flows)

Rank	Fireworks	Baseten	Together	Modal
#1	Google 24	Microsoft 17	Microsoft 34	Spotify 16 (founder signal)
#2	AWS 20	Google 17	Apple 25	Google 10
#3	Meta 17	Twitter 7	Amazon 23	Microsoft 7
#4	Microsoft 16	Sigma 5	Google 21	Better 7
#5	Facebook 12	—	Topsy 12	—

How valuable were this section's answers?

click 👍 / 👎 to leave feedback

Analyst

What does the growing-skills data say about where each company is investing right now?

LinkedIn Talent Data Insights

The fastest-growing skills at each company expose what each is actually building today. All four are investing in cloud-infra depth — but Baseten's signal stands out: they're scaling their own recruiting machine.

The top fastest-growing skills (1-year growth, YoY).

- Fireworks: Analytical Skills +400%, Amazon S3 +350%, Ansible +260%, Amazon EC2 +233%, Cluster Analysis +220%. Cloud-ops and analytics build. Pure infrastructure ramp.
- Baseten: Bash +1,000%, Reliability Engineering +750%, Application Deployment +700%, Recruiting +700%, Bayesian statistics +650%. The recruiting skill jumping +700% is the standout — Baseten is hiring recruiters as fast as it's hiring engineers, consistent with the 112 open-jobs count (more than the other three combined).
- Together: AWS Lambda +208%, Amazon S3 +187%, API Development +177%, REST +167%, Amazon EC2 +162%. The most uniformly AWS-cloud-infra ramp of any of the four. Together is building serverless / API surface depth.
- Modal: Node.js +1,000%, Salesforce.com Implementation +700%, AWS +400%, React Native +400%, Redux.js +400%. The most product/frontend-skewed of the four — Modal is building dashboard / UI surface area and starting to layer in Salesforce-driven sales motion.

The signals on each company's stage.

- Fireworks is in pure infra-build mode. The skills are all backend cloud (S3, EC2, Ansible).
- Baseten is in operational-scale mode. The recruiting +700%, Application Deployment +700%, Reliability Engineering +750% pattern is what "we are running fast and need more hands" looks like.
- Together is doubling down on cloud infrastructure depth across AWS Lambda, S3, EC2, API/REST surface — a more mature platform build than Fireworks.
- Modal is the only one where the skill ramp signals commercial maturity — Node.js + React Native + Salesforce implementation suggests Modal is building real product UI and a sales motion, not just infra.

The top-skills inventory (current employees, by skill). Together has the deepest absolute skill stock (166 Python, 101 Java, 99 JavaScript, 99 Deep Learning, 93 C++). Modal has the strangest top-skill mix — 59 employees tagged with "Technical Staff Management" — consistent with a smaller, more senior-management-heavy team.

Fireworks — Top 5 fastest-growing skills (cloud-ops build)

Analytical Skills

400%

Amazon S3

350%

Ansible

260%

Amazon EC2

233%

Cluster Analysis

220%

Baseten — Top 5 fastest-growing skills (operational scaling — including Recruiting itself)

Bash

1000%

Reliability Engineering

750%

Application Deployment

700%

Recruiting

700%

Bayesian statistics

650%

Together — Top 5 fastest-growing skills (AWS / serverless build)

AWS Lambda

208%

Amazon S3

187%

API Development

177%

REST

167%

Amazon EC2

162%

Modal — Top 5 fastest-growing skills (product + sales motion building, not just infra)

Node.js

1000%

Salesforce Implementation

700%

Amazon Web Services (AWS)

400%

React Native

400%

Redux.js

400%

Top-5 skill inventory — total employees tagged with each skill

Rank	Fireworks	Baseten	Together	Modal
#1	Python 80	Python 103	Python 166	Python 61
#2	Java 64	Java 67	Java 101	Tech Staff Mgmt 59
#3	C++ 54	C++ 52	JavaScript 99	Java 47
#4	Deep Learning 54	Data Analysis 51	Deep Learning 99	Algorithms 32
#5	Machine Learning 52	SQL 48	C++ 93	Software Development 32

How valuable were this section's answers?

click 👍 / 👎 to leave feedback

Analyst

How does the geographic footprint compare? All four are US AI-infra companies — does the data show meaningful differentiation?

LinkedIn Talent Data Insights

Three of the four are SF-anchored, with NYC as the secondary hub. Modal is the only outlier — NYC-anchored, with a distinctive Stockholm cluster that maps directly to founder pedigree.

The headline geographic split.

- Together: SF Bay 202 (58%), The Randstad NL 24 (7%), NYC 18 (5%). Largest absolute SF concentration. The Randstad cluster is unique in the quartet — no other AI-infra peer here has a comparable European base. The NL operation reads as a deliberate European foothold.
- Baseten: SF Bay 134 (52%), NYC 49 (19%), Seattle 11 (4%). SF-anchored with the largest NYC presence in absolute terms after Modal. The 49 NYC count vs. 134 SF is a more balanced bi-coastal split than Fireworks or Together show.
- Fireworks: SF Bay 110 (53%), NYC 21 (10%), Seattle 6, London 5, Boston 5. Standard SF-heavy US-tech distribution. NYC presence is smaller than Baseten's in both absolute and relative terms.
- Modal: NYC 72 (49%), SF Bay 26 (18%), Stockholm 11 (7%), Boston 3. The only NYC-primary company in the quartet. The 11-person Stockholm cluster is the founder pedigree showing up in the data — Erik Bernhardsson is Swedish, ex-Spotify, and 16 of Modal's 147 employees have past experience at Spotify (their #2 past-employer source). Stockholm is Spotify's HQ.

Geographic flow signal. Fireworks' fast-growing locations are San Francisco Bay Area and NYC — they are deepening their existing two hubs, not adding new ones. Baseten's locations grew most in SF (134 from a smaller base last year) and NYC (49). Together's fast-growing locations are The Randstad NL and SF — they're scaling Europe as well as headquarters. Modal's fast-growing locations are NYC, SF, and Stockholm — all three of their existing hubs growing in parallel.

What this says. Together's NL cluster suggests they see the European inference market as worth a dedicated office, not just remote hires. Modal's Stockholm footprint is a network effect of the founder, not strategic geographic expansion — but it gives them a Northern European talent pipeline the others lack. Fireworks and Baseten are running a standard US bi-coastal playbook with no comparable signal.

Top metros by headcount (May 2026 capture)

Metro	Fireworks	Baseten	Together	Modal
San Francisco Bay Area	110	134	202	26
New York City Metro	21	49	18	72
Greater Seattle	6	11	6	—
London Area, UK	5	—	7	—
Greater Boston	5	3	—	3
Greater Stockholm	—	—	—	11
The Randstad, Netherlands	—	—	24	—
Austin, Texas Metro	2	—	7	1
Los Angeles Metro	—	6	6	—

Fastest-growing locations — where each is expanding

Company	Fastest-growing locations	Signal
Fireworks	San Francisco Bay, NYC	Standard US bi-coastal deepening
Baseten	San Francisco Bay, NYC	Standard US bi-coastal scaling
Together	The Randstad NL, San Francisco Bay	Active European expansion
Modal	NYC, San Francisco Bay, Stockholm	Three-hub model with founder-pulled Nordic pipeline

How valuable were this section's answers?

click 👍 / 👎 to leave feedback

Analyst

What does the LinkedIn Recruiter Open-to-Work signal say about retention risk at each, and how does the senior/entry mix shape the picture?

LinkedIn Talent Data Insights

The Recruiter Open-to-Work signal mostly confirms the Talent Insights attrition picture — but with a meaningful twist on Baseten (high active-talent share despite moderate attrition).

Aggregate Open-to-Work share (LinkedIn Recruiter, May 2026 pull).

- Fireworks: 50 of 210 captured profiles flagged Open-to-Work (24%). Highest forward retention risk. Active Talent (actively searching, not just open): 28 of 210 (13%).
- Together: 75 of 383 (20%). Mid-pack. Active Talent: 33 (9%).
- Baseten: 40 of 268 (15%). Below average OTW share, but Active Talent is 41 of 268 (15%) — the highest concentration of actively-searching employees in the quartet despite the lowest OTW%. Suggests the people who are signaling departure are doing so actively, not casually.
- Modal: 22 of 158 (14%). Lowest OTW share. Active Talent: 8 (5%) — the lowest actively-searching share. Most retentive on a forward-looking basis.

The senior/entry mix. This shapes how to read the retention data — senior-heavy workforces typically have higher lateral-move propensity than entry-heavy ones.

- Together: Senior 165, Entry 114, Director 43, VP 23, CXO 16. The most senior-skewed workforce by far. 165 Senior vs 114 Entry inverts the typical pattern. Combined with 23% attrition, this reads as a workforce that has been through enough hiring cycles to accumulate senior bench — but is also losing the most. The high senior concentration likely explains why attrition is the highest: senior people move more easily.
- Modal: Senior 82, Entry 44. Senior-skewed despite being the smallest. Director 7, VP 4, CXO 4 — a thinner exec layer than Together's. Lowest attrition (10%) suggests this senior-skew is sticky, not churning.
- Fireworks: Entry 79, Senior 67. Entry-heavy mix. Combined with the 22% TI attrition and 24% OTW%, the picture is: a more junior workforce, less stable, with the highest forward retention risk.
- Baseten: Entry 112, Senior 86. Most entry-skewed in absolute and relative terms. 17% attrition is moderate, but the 15% Active-Talent share (highest in the quartet) is the warning signal — the junior workforce is actively shopping.

The synthesis on retention.

- Modal = stable senior workforce, low forward risk, lowest attrition. Most insulated.
- Together = senior-skewed and stretched. Highest attrition, mid OTW. The senior layer is the most vulnerable to outside offers.
- Fireworks = junior workforce with the highest forward retention risk. Needs to fix retention before it scales further.
- Baseten = junior workforce in scaling mode. Highest Active-Talent share is the canary — the recruiting +700% growth-skill signal makes sense in this context.

Open-to-Work share (LinkedIn Recruiter, May 2026 pull)

Fireworks — Open to Work

24% of 210 captured profiles

Baseten — Open to Work

15% of 268 captured profiles

Together — Open to Work

20% of 383 captured profiles

Modal — Open to Work

14% of 158 captured profiles

Recruiter spotlights — retention risk indicators

Spotlight	Fireworks	Baseten	Together	Modal
Captured profiles (Recruiter total)	210	268	383	158
Open to Work	50 (24%)	40 (15%)	75 (20%)	22 (14%)
Active talent (actively searching)	28 (13%)	41 (15%)	33 (9%)	8 (5%)
TI attrition rate (1y)	22%	17%	23%	10%

Seniority mix — Together unusually senior-skewed for its size; Fireworks/Baseten entry-heavy

Level	Fireworks	Baseten	Together	Modal
Senior	67	86	165	82
Entry	79	112	114	44
Director	19	18	43	7
Manager	12	11	20	6
VP	4	3	23	4
CXO	7	4	16	4

Workplace types — Modal's tiny remote share signals a tight in-person culture

Workplace type	Fireworks	Baseten	Together	Modal
On-site	35	28	58	9
Hybrid	35	29	58	9
Remote	22	21	45	6

How valuable were this section's answers?

click 👍 / 👎 to leave feedback

Analyst

Pull it together. What's the workforce verdict on the four companies, and which one looks best-positioned on these signals?

LinkedIn Talent Data Insights

On workforce signals alone, Modal looks the best-positioned of the four. Modal is the smallest but has the most defensible workforce profile — engineering-pure, retentive, founder-pulled, lowest forward retention risk. The other three each have a meaningful workforce concern.

The four-line verdict.

- Modal = smallest (147) but most defensible. 50% Engineering (highest), 10% attrition (lowest), 14% OTW (lowest), 5% Active Talent (lowest). NYC + Stockholm dual-hub with founder-pulled Spotify talent (16 ex-Spotify employees in a 147-person company). The skill ramp shows commercial product-build (Node.js +1000%, Salesforce Implementation +700%) — they are starting to layer in sales motion on top of infra. Risk: smallest market presence, slowest absolute hiring (90/year), no listed open jobs visible in TI capture.
- Together = biggest (351) and most stretched. 23% attrition is the highest. 165 Senior vs 114 Entry is the most senior-skewed mix in the quartet — they have the deepest experience bench but the most attractive workforce to poach from. 24-person Netherlands cluster gives them a European foothold no one else has. Risk: senior workforce + high attrition + high OTW% on a 383-profile Recruiter base = they are leaking experienced talent.
- Baseten = fastest scaler (+195% YoY, 112 open jobs). The growing-skills list reads as operational ramp — Recruiting +700%, Application Deployment +700%, Reliability Engineering +750% — they're scaling internal processes alongside engineering. Risk: 15% of profiles are Active Talent (highest in the quartet), and the workforce is the most entry-heavy. The recruiting machine is real, but the junior workforce is actively shopping.
- Fireworks = mid-pack on everything, with the worst forward retention signal. 24% Open-to-Work is the highest in the quartet (50 of 210 captured profiles), and 28 are flagged Active Talent — 13% of the workforce is actively searching. Risk: this is a workforce that hasn't yet stabilized; the broad FAANG hiring funnel hasn't translated to retention.

The one workforce signal that could move the picture on each:

- Modal: if the Stockholm/Spotify pipeline thins out (the founder network is finite), the company's distinctive talent advantage shrinks. Watch the Stockholm count over 12 months — if it doesn't grow, the moat is structural rather than scalable.
- Together: if attrition stays at 23%, the company will be replacing nearly a quarter of its workforce annually — and the senior layer (most attractive to poach) is the most exposed. The single most actionable signal is whether attrition normalizes toward 15-17%.
- Baseten: if the 15% Active Talent share doesn't drop as the company scales, that's a signal that growth is outpacing culture/comp. The Recruiting +700% skill ramp tells us they know — the question is whether they can backfill faster than people are leaving.
- Fireworks: the 24% OTW + 13% Active Talent combination is the alarm bell. Without a retention intervention, the company will hit a workforce stall well before it solves product-market fit at its target scale.

The summary scorecard.

Workforce signals scorecard — Modal best-positioned overall

Dimension	Fireworks	Baseten	Together	Modal
Growth velocity (YoY %)	+92%	+195% (best)	+50%	+101%
Retention (1 − attrition)	78%	83%	77%	90% (best)
Forward retention (1 − OTW%)	76%	85%	80%	86% (best)
Active-talent share (lower is better)	13%	15% (worst)	9%	5% (best)
Engineering %	43%	36%	43%	50% (best)
Open-jobs intensity (hiring momentum)	13%	44% (best)	15%	—
Geographic differentiation	Standard US	Standard US	+Europe (NL 24)	+Stockholm (founder)
Distinct talent advantage	Broad FAANG pull	Scaling speed	Senior depth + EU base	Founder-network pull, retention
Biggest workforce risk	Highest OTW%	Highest Active Talent %	Highest attrition	Founder-network finite

How valuable were this section's answers?

click 👍 / 👎 to leave feedback

Continue the analysis

How valuable were this section's answers?

Helpful, missed the mark, or somewhere in between? Tell me what you'd want more or less of.

Not at all helpfulExtremely helpful

Prefer email? raybould@gmail.com

Methodology

All workforce data captured from LinkedIn Talent Insights and LinkedIn Recruiter on 2026-05-19.

Captured profiles are LinkedIn-visible employees tagged to a company. This set typically diverges slightly from active headcount because recent ex-employees may still list the company on their profile, and conversely some recent hires haven't yet updated.

Talent flow numbers are captured movements between companies over the indicated window (12, 6, or 24 months). All four companies are too small for individual flows over 10 — small absolute numbers can be noisy. The signal is in the patterns: who pulls from whom, and what kind of company (FAANG, infra peer, niche acqui-hire).

The four companies cover the same competitive surface: managed inference / model-serving infrastructure for foundation models. They all compete with each other and (laterally) with the hyperscalers' own inference offerings.