Google on Wednesday launched Gemini 3.8 Flash and a new cybersecurity model just as Washington stepped up scrutiny of foreign-made power equipment used in the U.S. data-center buildout. The two moves sit on opposite sides of the same machine: one side racing to sell more AI, the other side tightening control over the grid, the parts, and the chokepoints that keep the whole thing running.
Who Gets the Spoils
Google said Gemini 3.8 Flash is its third Flash model in six weeks. The company described it as focused heavily on coding and agentic tasks, and called it Gemini's best reasoning and coding model yet, with significant improvements over 3.7 Flash in software engineering and multi-step tasks. Tulsee Doshi, senior director of product management at Google DeepMind, said the recent Flash models have "really surprised us in positive ways in their performance," adding that they "give us opportunities to lean into them."
The pricing tells its own story. Google set Gemini 3.8 Flash at 75 cents per million input tokens and $3.75 per million output tokens, the same introductory price as its last Flash model. Gemini Enterprise is adding pay-as-you-go pricing, token discounts of up to 20%, monthly caps on agent spending and a zero-dollar base subscription option. According to materials provided to CNBC, Google also directly targeted Microsoft and Anthropic, saying their recurring seat fees and separate product licenses make their offerings more expensive and less flexible.
Nearly three-quarters of Google Cloud customers are already using its AI products, the company said, and Google Cloud CEO Thomas Kurian told CNBC those customers are spending about 50% more than their original commitments. DeepMind's Demis Hassabis told the G20 Innovation meeting on Wednesday that Gemini could increasingly serve as a general-purpose layer coordinating cheaper, specialized models and agents. It was the first time Hassabis had spoken publicly since DeepMind's reorganization last month, when he moved from CEO of DeepMind to chairman of the unit.
What the Defenders Get
Google is also pushing Gemini 3.8 Flash Cyber, saying it can detect and patch software vulnerabilities at frontier-level performance while running substantially faster and more affordably than larger models. Doshi said, "We're really excited about being able to provide an offering to defenders that is a fraction of the cost, much faster, while still showcasing that frontier-level performance." Because those capabilities could also be misused, Google is initially limiting access to a small group of trusted government and enterprise defenders through its new Fairwind Program.
That gatekeeping matters. The company is selling speed and scale, but only to the institutions it trusts to hold the keys. The rest of the public gets the bill, the risk, and the spectacle.
Berkshire Hathaway CEO Greg Abel also gave Alphabet a public vote of confidence. Abel told CNBC's Becky Quick on Wednesday that Berkshire sees Alphabet as a winner in AI, saying, "We have a lot of visibility from within our companies as to how we're using AI, what type of benefits it's delivering, so that brought incremental interest, and then we saw Google as a significant player."
Alphabet's ad business got another boost on Wednesday when a federal judge rejected the Justice Department's push to force Google to sell its AdX exchange, opting for behavioral remedies rather than the structural breakup regulators had sought. The decision followed a separate antitrust ruling last year, when a judge rejected calls to force Google to divest Chrome. Antitrust attorney Wyatt Fore, a partner at Schinder Cantor Lerner, told CNBC it was a "big deal" that courts in both cases opted against breaking up the company, adding that Google is heading into the AI race with "no hands tied behind its back."
Who Pays for the Buildout
At the same time, the U.S. AI data-center boom is exposing heavy reliance on Chinese-made equipment. Chinese firms supply large portions of the parts needed to develop these facilities, including transformers, switchgear, batteries and optical tech, analysts told CNBC. The power stack in a data center includes transformers, which change high-voltage electricity from the grid to lower levels needed for servers and cooling equipment; switchgear, which are centralized systems with switches, fuses and circuit breakers; and batteries, which provide backup power.
Ben Boucher, senior analyst, supply chain for Wood Mackenzie, said, "There are a few key areas that the AI infrastructure buildout is increasingly dependent on Chinese imports," adding, "The most notable ones include substation transformers, which hyperscalers typically have on site to step transmission voltage down." Yury Dvorkin, associate professor at Johns Hopkins University, said, "The mid-term exposure is concentrated in grid connectors: transformers, switchgear, and batteries," and added, "China's share of certain transformer and switchgear categories runs near 30%, and it accounts for over 40% of U.S. battery imports." He said there is also deeper exposure upstream in copper, electrical steel and battery cathode materials.
Optical technology is another area of dependence. Companies including Zhongji Innolight and Eoptolink lead global data-center optical transceiver revenue, and Chinese firms collectively account for roughly two-thirds of global unit supply, according to research firm Counterpoint. Neil Shah, vice president of research at Counterpoint, said in an August report that "Western competitors like Coherent and Lumentum possess advanced photonic designs, but currently lack the cleanroom capacity, automated packaging infrastructure, and yield scale required to absorb Innolight and Eoptolink's volume within a 12-to-24-month horizon."
Washington is moving to reduce that dependence as AI infrastructure becomes more important to the U.S. economy and national security. President Donald Trump signed an executive order last week declaring a national emergency around the "extraordinary foreign threat" to the U.S. involving bulk-power system equipment produced abroad. The order authorized the Energy Department to prohibit or impose conditions on certain transactions involving some components used in the grid and data centers. Trump said in a statement connected to the order that "Since my first term, the threat to the United States regarding foreign supply of bulk-power system electric equipment has become even more acute," and added that "The rapid growth of advanced manufacturing, data centers, artificial intelligence, and defense production has increased the Nation's dependence on abundant, reliable electricity and magnified the consequences of a successful attack or supply disruption on the bulk-power system."
Trump had already said in April that "transformers, transmission lines and conductors, substations, high-voltage circuit breakers, power control electronics" were "essential to the national defense" in a Presidential Determination. A White House spokesperson told CNBC, "Reshoring manufacturing that's critical to our national and economic security has been a top priority for President Trump, and the Administration continues to deliver with a robust and nimble agenda of tax cuts, tariffs, and deregulation," adding, "Trillions in investments across key sectors – from steel to semiconductors to autos – prove that the Administration's strategy is paying off."
The Federal Communications Commission added power inverters produced in foreign countries to its Covered List in July, saying the equipment and services on the list pose an unacceptable risk to national security. CNBC also reported that the Trump administration is drafting a ban on U.S. imports of new Chinese optical transceivers, which allow data to travel at high speed over fiber optic cables between data centers. The FCC did not respond to a request for comment and the White House did not address the report.
Hitachi Energy announced in September 2025 that it would invest $1 billion to expand production of critical grid infrastructure in the U.S., including $457 million for a new large power transformer facility to meet demand from the AI buildout. Siemens Energy said in February it would invest $1 billion in U.S. production for grid and gas turbine equipment for AI infrastructure and data center expansion.
S&P Global said in June that U.S. data-center capacity is forecast to grow from 62 GW in March 2026 to 152 GW by 2030 because of high-density AI workloads. Wood Mackenzie said power transformers and substations are already in an estimated market shortage of 15% and 8%, respectively, in 2026, and that restrictions on units made in China would further worsen supply chain challenges. Boucher said in an August blog that "A bulk of the impacts will be centered around data centers who have been using Chinese units to minimize lead times," and added, "The 100 MVA+ segment is where the shortage is already most acute, and it is precisely where data centers have been turning to Chinese manufacturers to manage lead times."
Laveena Iyer, senior analyst at The Economist Group, said, "Scrutiny over China's presence in the U.S. data centre power stack has only risen over the past eight months or so, this is a shift from the previous focus that centred solely on the compute stack." She also said an optical transceiver ban would force U.S. hyperscalers to look for alternatives locally, potentially pushing up costs, and added that "The U.S. administration appears keen to de-risk its AI infrastructure before Chinese tech is fully embedded in it, quite like the 5G rollout where removing Chinese telecoms gear later on led to network rollout delays and rising costs for telecoms companies."
CNBC said it contacted the U.K. embassy of the People's Republic of China for comment on Trump's executive order and on China's involvement in the supply of parts for the AI data-center buildout.