OpenAI disclosed plans in March 2026 to grow from roughly 4,500 employees to 8,000 by December — a 78% headcount increase inside a single calendar year. Anthropic, at an estimated 3,000 to 5,000 employees generating annualized revenue approaching $30 billion, is doing something closer to the opposite: running what SaaStr's Jason Lemkin called, in a widely-shared April 2026 post, "the most efficient company anyone has ever seen at this revenue scale." The two companies are running competing theories of what an AI lab is supposed to be.
The H1 2026 data on headcount, compensation, and office footprint tells the story of that experiment in motion.
What OpenAI Is Building
The Financial Times, citing two people with direct knowledge of the matter, reported on March 21 that OpenAI's new hiring plan allocates the bulk of the 3,500 net-new roles across product development, engineering, research, and sales — with a new category of "technical ambassador" roles added to the slate. Technical ambassadors, per OpenAI's own job postings and a subsequent analysis by AInvest, are not salespeople. They are deployment engineers whose job is to sit inside enterprise client organizations and help businesses extract value from OpenAI tools. The role category exists because OpenAI's fastest-growing revenue segment is enterprise API and ChatGPT Enterprise — and enterprise AI deployment fails at the integration layer, not the model layer.
To house the incoming headcount, OpenAI has leased more than 1 million square feet of office space in San Francisco's Mission Bay, according to reporting by The Real Deal and CoStar in March 2026. The deal for a 280,000-square-foot sublease at the former Dropbox headquarters on 1800 Owens Street pushed OpenAI's San Francisco footprint past the seven-figure threshold. The company also signed a separate campus lease in Mountain View in February, per the SF Standard, giving it a two-city Bay Area infrastructure capable of absorbing the planned additions.
The equity structure underpinning the hiring sprint is without precedent in private-market tech. Per a Fortune report dated February 18, 2026, citing Wall Street Journal data, OpenAI paid an average of $1.5 million in stock-based compensation per employee in 2025 — the highest average among any major tech startup on record, and roughly six times the inflation-adjusted equivalent at Google before its 2004 IPO. With roughly 4,000 to 4,500 employees at the time, the total annual equity bill ran approximately $6 billion, representing more than half of annualized revenue at the time. The mechanism is PPUs — Profit Participation Units — which vest over four years at 25% per year with no cliff and can be liquidated during periodic tender offers. The structure is designed to make OpenAI equity feel like liquid compensation, which is both the retention tool and the recruiting pitch.
On the compensation stack: Levels.fyi data current to May 2026 puts OpenAI software engineer total comp at $249,000 at L2 to $1.28 million at L6, with the median falling around $555,000. Senior research scientists and staff engineers at the L5 to L6 tier pull packages in the $1.1 million to $1.4 million range per the comp reset covered separately by ENTRA in April. A "typical" mid-to-senior offer described by a recruiter with knowledge of current packages, who was granted anonymity to discuss confidential hiring terms, runs $300,000 base plus a $2 million PPU grant over four years — roughly $800,000 annualized total comp.
What Anthropic Is Building
Anthropic has not announced a comparable headcount target. What it has announced is revenue. By the first quarter of 2026, Sacra estimated Anthropic's annualized revenue at approximately $30 billion on a gross-reporting basis. The company's confirmed December 2025 headcount was approximately 2,300, per Fortune. Estimates for May 2026 range from roughly 3,000 (GetLatka sales intelligence) to 5,028 (Tracxn data cited by SaaStr in April), with the wide range reflecting Anthropic's practice of not disclosing official figures and the methodological differences between full-time-employee counts and extended workforce. The company carries 376 to 392 open roles on its careers page as of late May 2026, indicating continued active growth — but at nothing approaching OpenAI's stated pace.
The efficiency figure is the anchor. At $30 billion annualized and 3,000 to 5,000 employees, Anthropic's revenue per employee sits between $6 million and $10 million — higher than Google, Salesforce, or any company in the Forbes Global 2000 at comparable revenue scale. Google crossed $30 billion in annual revenue with more than 32,000 employees. Salesforce needed approximately 79,000. Epoch AI published a data note in Q1 2026 confirming that both Anthropic and OpenAI generate more revenue per employee than any major public tech company, with Anthropic's per-head figure leading OpenAI's.
The compensation architecture is comparable but structured differently. Anthropic total comp ranges from approximately $300,000 to $490,000 for software engineers, with senior research scientists pulling a median of $746,000 according to Levels.fyi data. Base salaries run $180,000 to $300,000 depending on role and level. Equity is granted as RSUs against a company that closed a Series G in February 2026 at an implied valuation approaching $380 billion, per Bloomberg and The Information. A senior engineer receiving $247,000 annually in RSUs at current valuation is betting on a company that has more than quintupled its stated value in eighteen months. Anthropic's Glassdoor compensation score, per JobsByCulture data, runs 4.8 out of 5.0 — the highest in a database of 118 companies.
Dario Amodei's operating philosophy on headcount is legible from the outside. In a widely-cited interview with Lex Fridman, Amodei described Anthropic's intent to keep teams small relative to compute operated: every person on the team should feel the leverage of what they're doing. That framing is not just a culture pitch. It is a deliberate constraint on hiring rate — a bet that marginal research output scales faster with model compute than with additional researchers, and that coordination cost rises faster than the research return from headcount addition past a certain density.
The Open Question
The two companies are now generating comparable annualized revenue on dramatically different staffing bases, which means the per-employee productivity implied in Anthropic's numbers is either structural — reflecting a fundamental difference in operating model — or temporary, likely to converge as both companies add enterprise sales infrastructure and the revenue composition shifts from API toward more labor-intensive deployment.
OpenAI's bet on the technical ambassador role points directly at that convergence. Enterprise revenue at scale requires human deployment infrastructure. OpenAI's decision to build it in-house, rather than through a partner channel, is an explicit bet that the labor cost of that infrastructure is recoverable through margin on the enterprise contract. It also explains why the 3,500 net-new roles in the 2026 plan are weighted toward product, engineering, and go-to-market rather than research — the research headcount at OpenAI was already substantially larger than Anthropic's, and the incremental headcount is designed to convert research output into revenue, not to generate more research output.
Anthropic's theory is the inverse. Keep the research-to-total-headcount ratio high, let the model output sell itself through API and Claude.ai subscriptions, and hold off on the enterprise sales layer until the product is differentiated enough to support a thinner go-to-market motion. The 376 open roles on Anthropic's careers page in May 2026 skew heavily toward research, engineering, and policy — the company's public listings include Research Engineer, Research Scientist, and Alignment Science roles as the highest-volume categories, with enterprise-facing roles a smaller share of the open slate.
Three Numbers to Watch in H2 2026
3,500. That is the net number of hires OpenAI needs to close between now and December to hit its 8,000 target. At 3,500 net hires in roughly six months, OpenAI would need to average approximately 580 new employees per month. In the first half of 2026, the company has not publicly disclosed a monthly hire rate. Whether it can execute at that pace — while maintaining the compensation floor that makes its offers competitive, and while integrating new hires into a culture that was built as a 1,000-person research organization — is the operational question of the year.
$6 million. That is the floor of Anthropic's revenue-per-employee range at the conservative end of current estimates. If Anthropic grows headcount toward 6,000 by year-end through its current open-role slate while revenue growth decelerates from the pace of the past 18 months, the per-employee figure compresses. If revenue continues tracking at the rate Sacra estimated for Q1, the figure holds or expands. The compression or expansion of that number will be the clearest signal of whether the lean model is structurally sustainable or a temporary artifact of a market in which Anthropic's model quality commanded premium pricing before competition intensified.
$1.5 million. That is the average OpenAI equity grant per employee in 2025. If OpenAI adds 3,500 employees in 2026 at anything approaching that average — even at half the rate for new hires — the total equity bill approaches $8 billion to $9 billion annually by December. At OpenAI's current revenue trajectory, that is a cost structure that requires continued revenue growth at a rate that few private companies have sustained past $10 billion annualized. The equity structure that makes OpenAI the most attractive compensation environment in tech history is also, at scale, a cost structure with no precedent in private markets.
The market will have a clear answer on which theory is right by Q1 2027, when both companies will be reporting against the headcount targets set in H1 2026. Until then, the two hiring philosophies are running in parallel — one optimizing for speed of scale, one for return per head — in the same San Francisco talent market, competing for the same researchers, at compensation bands that are now structurally above $1 million at the senior level regardless of which lab you are joining.
Sources: FT / Silicon Republic: OpenAI to double headcount — Fortune: OpenAI plans to double headcount — Fortune: OpenAI $1.5M equity compensation — The Real Deal: OpenAI Mission Bay lease — CoStar: OpenAI 1M sq ft San Francisco — SF Standard: OpenAI Mountain View campus — SaaStr: Anthropic efficiency — Epoch AI: Revenue per employee — Levels.fyi: OpenAI software engineer salary — Levels.fyi: Anthropic research scientist salary — JobsByCulture: Anthropic compensation 2026 — AInvest: OpenAI technical ambassador — Bloomberg / The Information: Anthropic Series G ~$380B valuation