OpenAI is paying $225K to people who can solder a wire harness and read a Linux terminal

The Broadcom Deal: OpenAI's First Custom Silicon Signals a Full-Stack Compute Pivot

OpenAI and Broadcom unveiled Jalapeño on June 24, the company's first custom-designed AI inference chip. It's an application-specific integrated circuit built to serve large language models, and engineering samples are already running workloads including GPT-5.3-Codex-Spark in OpenAI's labs. The companies designed the chip from scratch in nine months, which OpenAI calls the fastest ASIC development cycle it's aware of for high-performance semiconductors.

This is no side project. Greg Brockman, OpenAI's president, framed Jalapeño as the opening move in a plan to "build the full stack behind its models and products." OpenAI has spent the better part of four years as one of the world's largest buyers of Nvidia GPUs. It still is. But the Broadcom deal, first announced in October 2025, commits the two companies to deploying enough custom accelerators to require 10 gigawatts of power, with initial deployment targeted for late 2026 and expansion through 2029. Broadcom CEO Hock Tan told Bloomberg that his earlier projection of 1.3 gigawatts in 2027 "may prove conservative" because demand is, in his words, "simply insatiable."

The economics explain why. Inference, serving model outputs to ChatGPT users, Codex queries, and API calls, is where OpenAI's compute bill lives every single day. Nvidia's GPUs remain the default for training and for workloads that demand flexibility. But ASICs trade that flexibility for lower cost per query and the ability to tailor the silicon to the specific kernels, memory patterns, and serving systems OpenAI runs. Tan, in an interview with Reuters, placed Jalapeño on par with Nvidia's Blackwell line and Google's tensor processing units. OpenAI's own self-reported numbers ("substantially better" performance per watt) haven't been independently verified. A detailed technical report is expected in the coming months.

OpenAI designed the chip architecture. Broadcom handled silicon implementation and contributed its Tomahawk networking silicon for large-scale production. Celestica builds the boards, racks, and systems. And OpenAI's own models helped accelerate parts of the design and optimization process, a recursive loop the company bets will get stronger over time. Richard Ho, who leads OpenAI's hardware program, said the architecture was optimized around "the kernels, memory movement, networking, and serving patterns that matter most for frontier AI models." A next chip generation is already planned for 2028, with annual releases after that.

The move mirrors what hyperscalers have already done. Google has TPUs. Amazon has Trainium and Inferentia. Meta has MTIA. Microsoft has Maia. OpenAI now joins that list with Broadcom as its manufacturing and networking partner, the same role Broadcom fills for several of those other companies. For Nvidia, the risk isn't that Jalapeño beats the Blackwell line on a benchmark. It's that the biggest AI operators stop treating a general-purpose GPU stack as the only viable answer for inference at scale.

Inside the San Francisco Robotics Prototyping Lab: Roles That Bridge AI and Physical Systems

OpenAI's robotics team is hiring in San Francisco for roles that sit at the intersection of AI and physical hardware. The Prototyping Lab Technician, Robotics posting lists a compensation range of $225K–$295K plus equity. It is a hands-on build role that requires interpreting schematics, wiring harnesses, and mechanical drawings, plus comfort with the Linux command line. The team aims to unlock general-purpose robotics and push toward AGI-level intelligence in dynamic, real-world settings, and the work spans structural, electrical, sensing, and actuation subsystems.

A second opening, the 3D Printing Lab Technician, Robotics, focuses on electromechanical assembly and prototype builds. A third, the Robotics Lab Manager & Technician, describes running "the beating heart of our prototyping environment," building mechanical assemblies, wiring harnesses, and debugging early-stage robots. All three are based in San Francisco and are in-person five days a week.

The job descriptions reveal what OpenAI's hardware ambitions actually demand at the lab level. Technicians fabricate fixtures and test equipment, execute low-volume builds and rework activities, and run hardware experiments alongside engineering teams. The emphasis on "rapid iteration" and "high-velocity, early-stage hardware development environments" signals that the robotics division is still in the prototype phase, not scaling production. These are not manufacturing roles. They test whether general-purpose robotic systems can move from concept to working physical hardware.

The compensation bands tell their own story. A prototyping technician role at $225K–$295K base puts these positions at or above many senior software engineering salaries in the Bay Area. OpenAI is paying a premium for people who can bridge the gap between AI research and physical system constraints, and who are willing to do it on the lab floor, not remotely. The company lists 709 open roles across all departments, with these robotics positions among the newest additions, suggesting the hardware push is still in its early stages but moving fast.

The Workforce Signal: What OpenAI's Hiring Mix Reveals About Post-LLM Engineering

OpenAI's robotics hiring surge in San Francisco is not a broad exploration. It is a targeted rebuild of a capability the company abandoned in 2019, when it concluded it lacked enough physical-world data to train useful robotic systems at scale. Now it is hiring back into exactly the roles that solve that original problem, and the composition of the job postings reveals where the broader AI industry's talent demand is heading.

The open roles span actuator design, electrical engineering, simulation environments, control systems software, data acquisition, and machine learning for distributed data systems. Base salaries for some positions reach $310,000 before equity, according to Crypto Briefing. That compensation level signals a talent war with Tesla, Google DeepMind, Figure AI, and a growing cohort of humanoid startups all fishing in the same pool of engineers who understand both inference and physical systems.

What the roles actually show:

The overlap is the story. OpenAI's Broadcom chip deal, focused on LLM inference, and its robotics prototyping lab both demand engineers who can optimize model execution under real-world constraints (power, latency, thermal limits, sensor noise). The skills transfer in both directions. An engineer who tunes inference on a custom ASIC for a data center can tune inference on an actuator controller on a robot. The hiring mix reflects that convergence.

Caitlin Kalinowski, OpenAI's Member of Technical Staff who posted the initial robotics roles on X in January 2025, described the first openings as two senior IC roles, an EE Sensing Engineer and a Robotics Mechanical Design Engineer, plus a TPM Manager to stand up the training lab. Brockman shared the post. The company's careers page now lists roughly 706 total roles, with robotics and hardware positions concentrated in San Francisco.

The earlier constraint that killed OpenAI's first robotics effort, lack of real-world data, has weakened. Vision-language models are stronger, sensors are cheaper, and simulation tools have improved. The hiring push is OpenAI's bet that the data bottleneck is now solvable in-house, not through partners alone.

That shift has implications for the talent market. Engineers with experience in calibration, sensor fusion, simulation realism, real-time control, and robot data pipelines were already in demand at Figure AI, 1X, Tesla, Agility Robotics, and Physical Intelligence. OpenAI entering that same labor market at $210K to $310K base will push compensation higher across the sector and make specialized hiring harder for smaller players.

The next signal to watch is whether OpenAI rebuilds senior robotics leadership. Kalanowski reportedly resigned in March 2026, according to Startup Fortune, after raising concerns about OpenAI's Pentagon work and guardrails around surveillance and lethal autonomy. If the company continues adding roles across perception, controls, and data acquisition while filling that leadership gap, the industry will have stronger evidence that this is a sustained strategy, not a recruiting stunt.

How OpenAI's Hardware Bet Stacks Against Anthropic, Google, and SpaceX

OpenAI's dual move, custom silicon with Broadcom and a robotics prototyping lab in San Francisco, puts it in a strategic position that none of its major AI competitors have matched in scope. The chip deal addresses inference cost and supply control. The robotics hiring push addresses what happens when AI leaves the data center. Together, they signal OpenAI is building toward vertically integrated AI infrastructure, from transistor to physical agent.

Anthropic shows no comparable hardware ambition on its public job board. The company's recent postings lean toward enterprise partnerships, legal infrastructure, and compute sourcing; a Strategic Deals Lead role for Compute, Networking & Memory in San Francisco suggests it's negotiating for existing capacity, not designing its own. Anthropic's safety-first brand positions it as the responsible AI company, but that positioning doesn't require custom chips or robotics labs. The company appears to be doubling down on software differentiation and regulatory positioning rather than hardware integration.

Google remains the most direct comparison, and the most formidable. Google's custom TPUs have been in production for years, its DeepMind division runs robotics research, and its data center infrastructure gives it a level of vertical integration OpenAI is only now attempting to build. Google's advantage is time: it has been shipping custom AI silicon at scale while OpenAI was still entirely dependent on merchant GPUs. OpenAI's Broadcom deal is a catch-up move as much as a strategic one. The robotics lab hiring suggests OpenAI knows it can't match Google's head start in isolation, and it needs physical-world data and embodied AI capability to compete on the next frontier.

SpaceX occupies a different competitive category entirely. Zero G Talent's board lists 119 SpaceX roles added in the past week, spanning propulsion analysts in Hawthorne and transport technicians in McGregor, Texas. SpaceX builds physical hardware at scale (rockets, not robots), but its engineering culture and manufacturing rigor set a bar that AI companies entering hardware for the first time will be measured against. The salary ranges tell part of the story: SpaceX's Raptor propulsion roles pay $135,000–$190,000, while OpenAI's Senior RTL Engineer for interconnect design tops out at $445,000. OpenAI is pricing hardware talent at AI-software rates, a signal that it's competing for the same silicon design engineers that the broader tech and defense industries also need.

The competitive picture that emerges is this: Google has the most mature full-stack AI hardware operation, Anthropic has chosen not to compete on hardware at all, and SpaceX sets the physical-engineering standard that any AI company building real-world systems will eventually face. OpenAI is betting that controlling its own inference stack and building embodied AI capability is the path to staying independent and vertically integrated. The 44 roles OpenAI added in the past week, including a Prototyping Lab Technician and a Senior RTL Engineer (both in San Francisco), are the first visible hires in that direction.

The Skills That Will Define the Next AI Hiring Wave

OpenAI's job postings tell a story that most workforce forecasts miss. The signal isn't in the company's press releases; it's in the line items. A Prototyping Lab Technician role asks for someone who can troubleshoot electrical systems, interpret wiring diagrams, build wire harnesses, and work the Linux command line, all in the same posting. That's not a software job with a hardware hobby attached. That's a new job category.

The salary range for that role ($225,000 to $295,000, per Zero G Talent's current board data) puts it in line with senior software engineering positions at the same company. OpenAI is paying top-of-market for people who can physically build and fix things, not just model them. That pricing signal matters more than any CEO quote about "AGI in dynamic real-world settings."

The convergence skills to watch:

• Inference optimization meets power electronics. OpenAI's Broadcom partnership means someone has to design, test, and deploy custom silicon at the board level. The Senior RTL Engineer, Interconnect Design posting in San Francisco pays up to $445,000 a year, a chip architect's salary. But the prototyping technician role requires hands-on troubleshooting of the power delivery and sensing subsystems that sit on top of that silicon. The workforce demand spans from transistor-level design to soldering a wire harness, and the middle layers between them.

• Electromechanical troubleshooting as a core competency, not a nice-to-have. The prototyping technician posting lists building, modifying, and repairing electromechanical assemblies spanning structural, electrical, sensing, and actuation subsystems. Defense and robotics operators reading this should note: the same skill set that qualifies someone to maintain a UGCV or robotic assembly line qualifies them to prototype the next generation of AI-driven physical systems. The transferable skill isn't "AI knowledge." It's the ability to read a schematic, identify a fault, and fix it fast.

• Simulation-to-hardware loop fluency. OpenAI's robotics team is hiring for roles that span custom actuator design, simulation realism engineering, and large-scale data acquisition. The gap between simulation output and physical prototype is where most teams lose months. People who can close that gap (who understand why a simulated actuator behaves differently when it has backlash, thermal drift, and a noisy encoder) will be the bottleneck hires.

• Lab operations and process discipline. The prototyping technician role includes maintaining lab equipment, tooling, inventory, and workspaces, plus developing processes that improve build quality, repeatability, and documentation. That reads like a manufacturing engineering job description. It is. The difference is the pace: OpenAI wants someone comfortable in fast-paced environments with a wide variety of projects. Defense contractors and aerospace firms already run this model. The talent pool exists; it just hasn't been recruited into AI yet.

What this means for operators and engineers tracking the next wave:

If you're an embedded systems engineer, a robotics technician, or someone who's spent years making physical systems work in unforgiving environments, OpenAI's hiring mix is a market signal that your skill set is about to get a lot more competitive. The company posted 44 roles in the past week alone on Zero G Talent's board, and the hardware-adjacent positions aren't outliers; they're a pattern.

The next AI hiring wave won't just be about who can fine-tune a model. It's going to be about who can make the model move something in the real world, reliably, repeatedly, and at scale. That's a different workforce. And based on what's being posted right now, it's being built in San Francisco, in person, five days a week, at salaries that say OpenAI is serious about it.

Working in AI? Zero G Talent tracks the openings: browse AI jobs, openings at SpaceX, OpenAI and Anthropic, and the people building the field.