November 20, 2026:


BrightAI, the Palo Alto, California, company that mounts battery-powered sensors on utility poles, pipelines, compressors and tanks, announced a $97 million round on Wednesday, November 18, led by Smith Point Capital. Founder Alex Hawkinson told Fortune that GE Vernova, already a BrightAI customer, insisted on investing even though BrightAI had not planned to take strategic money this round. The sensors run AI on the device, so they can flag faults without a constant internet connection, which matters to anyone who pays for inspection crews or depends on pipes, poles and pumps, but nearly every performance figure in the announcement comes from BrightAI itself.
BrightAI plans to use the new money for offices and field operations in key Asian and European markets. Fortune reports that the round values the company at $750 million, but it attributes that figure to one unnamed person familiar with the matter, and The Next Web’s coverage of the round gives no valuation.
Smith Point Capital, the growth firm run by former Salesforce co-CEO Keith Block, led the round. New investors are Cobalt Capital, Marlinspike Partners, Zero Infinity Partners and GE Vernova. Existing backers Khosla Ventures, Inspired Capital and Upfront Ventures added money.
GE Vernova, whose equipment helps generate about a quarter of global electricity, plans to use BrightAI across its Power, Wind and Electrification businesses. The company is both investor and customer. That dual role means any results GE Vernova reports from the rollout will come from a shareholder as well as a customer.
BrightAI’s product, Stateful, has three main parts. Edge AI sensors attach to poles, pipelines, compressors and tanks and analyze conditions locally. Autonomous robots and drones inspect sites too remote or hazardous for crews, and BrightAI claims more than five times the productivity of manual inspection. Wearable AI assistants guide technicians through repairs and record their work. The platform also includes foundation AI models, and Webrazzi says it identifies root causes of detected problems.
BrightAI says Stateful runs at more than 50,000 locations and connects millions of sensors and systems. Its November 2024 launch release counted more than 250,000 AI endpoints across more than 25,000 sites, which would put the company’s reported site count at roughly double in two years. Customers named in coverage include GE Vernova, Reworld, Osmose Utilities Services, Azuria Water Solutions and Kodiak Gas Services, and The Next Web adds Diamondback Energy.
BrightAI’s sensor pods draw on cameras, vibration, ultrasonic, gas and thermal sensing, according to analyst firm Sacra’s company report, and do a first pass of analysis on the device. Only structured alerts, not raw feeds, travel to the cloud, where a model of each monitored asset supports a four-step loop: monitor continuously, diagnose using models trained on repair history, dispatch a work order with part numbers and instructions, and learn from the completed repair.
The processor behind that design is the Electron E1 from Efficient Computer, which BrightAI partnered with in September 2025. IEEE Spectrum describes the E1 as an array of stripped-down tiles that lack instruction fetching and branch prediction, joined by a programmable network so data moves from tile to tile, with a compiler that maps programs written in C onto the array. Efficient claims 10 to 100 times better energy efficiency than commercial ultralow-power CPUs on typical embedded tasks such as matrix multiplication, Fourier transforms and convolutions, and Spectrum’s report cites no third-party benchmarks. University of Illinois researcher Rakesh Kumar told Spectrum the main challenge for the startup is economic. Efficient Computer has described BrightAI as its first real-world deployment, though neither BrightAI’s announcement nor coverage of the round says how many already-deployed units carry E1 chips.
The tradeoff is where the cost lands. Local inference means a sensor on a remote pole does not have to push every byte over a cellular or satellite link, which cuts bandwidth needs. The cost shifts to keeping a physical fleet running: Sacra’s report lists battery replacement as an ongoing service cost and says battery capacity is not specified. Cisco notes that physical AI sensors degrade or collect debris and need maintenance and model retraining to stay accurate. For an operator, the question is less what data costs than what it takes to service thousands of devices in the field.
BrightAI’s case rests on two company figures: customers spot trouble an average of 51 days before a serious failure would have occurred, and more than half of scheduled inspections find nothing wrong. Hawkinson put the second figure higher in his Fortune interview, saying two of three scheduled trips find nothing to fix, while technicians he spoke with estimated 50% to 70%. The article quotes no independent analyst assessing either figure. Hawkinson’s own framing, as Fortune reported it, is blunter: “The Romans are laughing at us.”
Independent research urges caution about this category. McKinsey has found that predictive-maintenance savings are often canceled out by the cost of false alarms that cannot be avoided. BrightAI’s market sizing also needs reconciling. Its survey of 12,158 operators across 16 industries found 98% lack adequate visibility into their assets, and Fortune cites more than $7 trillion as waste and more than $7.5 trillion as addressable annual impact without explaining the difference. Outside research cited by BrightAI, including from the U.S. Department of Energy and the EPA, puts infrastructure failure costs at about $2.5 trillion a year.
A buyer evaluating any vendor in this category can ask which asset classes the models have been validated on, how false alarms are counted against the savings, and who audited the average warning time.
Webrazzi, The Next Web and The AI Insider all report $184 million in total funding. BrightAI’s July 2025 release put lifetime funding at $78 million after its Series A, and adding $97 million gives $175 million, which leaves about $9 million unaccounted for; no coverage explains the gap. The disclosed rounds were smaller still: CB Insights lists $66 million raised across three rounds.
One more data point complicates the picture. Forge Global, a private-market trading platform, lists a Series A-2 dated December 4, 2025, for $90.35 million at a $549.93 million post-money valuation, drawn from the company’s most recent certificate of incorporation, and says it has found no independent public confirmation that the round closed. None of the coverage of the November round cited in this article mentions such a round.
The valuation is similarly thin. Fortune’s $750 million figure rests on one anonymous source. Bloomberg reported in July 2025 that the Series A valued BrightAI at around $300 million, also citing a person familiar. Revenue offers no help in judging either number. BrightAI said in 2025 that it passed $100 million, and its June 2025 release described that as lifetime revenue, while SiliconANGLE reported that annualized revenue had topped $100 million. Fortune’s article discloses no revenue figure.
The Next Web reports that nearly 60% of BrightAI’s target market sits outside North America. Fortune reports the company is opening offices abroad but names no countries or office locations, and says headcount doubled in 2025. BrightAI’s current footprint is a Palo Alto headquarters, an operations center in Midland, Texas, and a briefing center in Washington, D.C.
Selling connected hardware into Europe brings regulation attached to the product itself. The EU Cyber Resilience Act covers products with digital elements, and since September 11, 2026, manufacturers must report actively exploited vulnerabilities and severe incidents, with an early warning due within 24 hours to the national response team and the EU cybersecurity agency ENISA. The act applies in full from December 11, 2027. Law firm Taylor Wessing lists IoT sensors among the products it covers. Germany’s chamber of commerce federation DIHK says the EU Data Act’s access-by-design rule for new connected products applies from September 12, 2026.
Sacra’s report, an analyst profile rather than a BrightAI statement, says Series A money was earmarked for sensor localization to meet IEC and CE standards and notes that government and public-utility contracts may require data to stay in-country. Data-residency rules are widespread: more than 100 countries have some form of data sovereignty law, and Wikipedia’s overview notes that the U.S. CLOUD Act creates tension with the EU’s GDPR for providers operating in both jurisdictions. Which markets BrightAI enters first will decide which of those rules apply to its sensors and cloud.
Hawkinson founded BrightAI in 2019. He previously founded SmartThings, an idea that came after a 2011 power outage let pipes freeze and burst at his family’s cabin in Leadville, Colorado; he sold the company to Samsung in 2014. Samsung’s price was never disclosed, though trade publications estimated at up to $200 million. Fortune reports BrightAI was bootstrapped before it took outside money, and its November 2024 seed round, led by Upfront Ventures, was its first outside capital.
BrightAI says it builds its own sensors, computing, AI models and field tools so that customers’ operational data stays off other vendors’ platforms, and Hawkinson’s framing is that what the system learns about an operator’s equipment should belong to that operator. The principle has a history. At a 2014 Washington Post forum, Hawkinson said SmartThings believed users should “100% own their own data.” What is missing is the contract language. Fortune’s article does not discuss data ownership, and BrightAI’s homepage text does not state ownership terms, though it links to a Trust & Security section.
For a reader deciding how much weight to give BrightAI’s pitch, the picture is mixed. The technical design is credible in principle: analyzing signals on the device is an established way to cut bandwidth and keep monitoring running where connectivity is poor. The numbers that would prove the business case, from the 51-day warning to the $750 million valuation, are company-reported or single-source, and total funding is off by $9 million between BrightAI’s own release and press reports. GE Vernova’s rollout across its Power, Wind and Electrification businesses is the clearest early test, because independent results from a large industrial customer would be the first outside evidence for the performance claims. Until those arrive, operators, investors and regulators in Europe and Asia have good reasons to ask for audited false-alarm rates, published data-ownership terms and a security review of the sensors before signing multi-year contracts.
Predictive maintenance assesses the current condition of equipment to estimate when it needs service, unlike preventive maintenance, which relies on average or expected life statistics. Common techniques include vibration, thermal and acoustic monitoring.
BrightAI and Hawkinson say what the system learns about an operator’s equipment should belong to the operator. Coverage of the round does not publish contract terms, and BrightAI’s homepage text does not define ownership, so operators would need to confirm it in their own agreements.
In the United States, a joint advisory from seven agencies, including the FBI, CISA, NSA, EPA and Department of Energy, was published April 7, 2026 and updated July 22. It warns that Iranian-affiliated actors are exploiting internet-connected operational technology devices, including programmable logic controllers, in the water and energy sectors; it does not involve BrightAI. Edge devices are often resource-constrained, which can limit security choices. In Europe, the Cyber Resilience Act’s reporting duties began September 11, 2026, and the act will apply in full from December 11, 2027. Neither BrightAI’s announcement nor coverage of it describes how its sensors are secured.
Stateful targets electricity, water, oil and gas, waste management and heavy industry, combining edge AI sensors, robots and drones, wearable assistants and foundation AI models.