Mark Zuckerberg is buying Meta a seat at the AI table. Does he know what to do with it?

Meta CEO Mark Zuckerberg is spending tens of billions of dollars to make Meta a leading frontier AI lab, including a $14 billion stake in Scale AI and plans for $130 billion to $145 billion in capital expenditures in 2026, but faces intense competition from OpenAI, Anthropic, and Google. Zuckerberg's August manifesto argues for distributing AI as 'personal superintelligence' rather than centralizing it, a vision that requires Meta to leapfrog rivals in a race it entered late after pivoting to the metaverse.

Meta’s CEO, Mark Zuckerberg, posted an AI manifesto https://www.fastcompany.com/91587997/mark-zuckerberg-wants-you-to-trust-meta-with-your-personal-ai-good-luck-with-that to his company’s website in August. The 6,537-word missive resembled similar proclamations from CEOs such as OpenAI’s Sam Altman https://blog.samaltman.com/the-gentle-singularity and Anthropic’s Dario Amodei https://darioamodei.com/essay/machines-of-loving-grace . Unlike theirs, however, it argued that the power of future AI https://www.fastcompany.com/section/artificial-intelligence models should be put directly in the hands of as many people as possible, in the form of “personal superintelligence” that would benefit everyone from small-business owners to 8-year-olds like Zuckerberg’s own daughter. “Rather than centralizing superintelligence, we should distribute it widely and give every person the ability to direct it,” he wrote. “This has the potential to begin a new era of personal empowerment where individuals can use this powerful new capability to reach their full potential, pursue their interests, and improve their lives and the world more than ever before.” Nine months earlier, during a Meta earnings call, Zuckerberg had foreshadowed how he would make this possible: “I’m very focused on establishing Meta as the leading frontier AI lab.” Even by Zuckerbergian standards, this is a stretch goal. It would require Meta to leapfrog—at a bare minimum—Anthropic and OpenAI, and maybe Google https://www.fastcompany.com/91502632/google-most-innovative-companies-2026 . Over time, Microsoft, Amazon, and SpaceX could also get in its way, along with a few Chinese AI companies. Other contenders will likely emerge. Zuckerberg can hardly be accused of failing to anticipate the rise of artificial intelligence: He established the company’s first lab dedicated to the technology in 2013. But he soon became distracted by virtual reality and the metaverse, a concept that so captured his imagination that Facebook Inc. became Meta in October 2021. Thirteen months later, OpenAI launched ChatGPT, and the race to command generative AI began in earnest. Since then, Meta has been engaged in a furious, sometimes chaotic game of catch-up, backed by its biggest spending spree ever. Meta declined to make executives available for on-the-record interviews for this article. In June 2025, the company paid $14 billion for a 49% stake in the training data company Scale AI, its second-largest deal after its $22 billion purchase of WhatsApp in 2014. Scale’s cofounder and CEO, Alexandr Wang, joined Meta as head of the newly formed Meta Superintelligence Labs, or MSL. Another recent splurge—$2 billion for Manus https://www.fastcompany.com/91515944/manus-ai-cleaned-up-my-computer-for-a-price , the maker of a buzzy AI agent—was ultimately undone https://www.fastcompany.com/91533167/why-manus-has-become-a-crucial-prize-in-the-global-ai-race by Chinese regulators. To lure top-level employees from its rivals, Meta has offered compensation packages unprecedented even in Silicon Valley. Some AI superstars are reportedly https://www.wired.com/story/mark-zuckerberg-meta-offer-top-ai-talent-300-million/ getting first-year pay of $100 million or more. The bulk of Meta’s AI investment, however, is going to the massive computing resources necessary to deploy AI at scale. The company plans to spend between $130 billion and $145 billion on capital expenditures in 2026, more than double its spending in 2024 and 2025 combined. Since 2024, it’s broken ground on nine data centers, including a 10-million-square-foot, $50 billion project in Richland Parish, Louisiana https://datacenters.atmeta.com/richland-parish-data-center/ . Meta has the resources to compete. In July, a report from the influential research firm SemiAnalysis argued that OpenAI and Anthropic were in a two-way competition for the title of frontier AI’s leading lab, but that Meta’s data, talent, and computing resources gave it the best chance of becoming their peer. Meta’s ad business remains a potent engine of growth, accounting for 98% of its revenue in the second quarter of 2026, which saw net income of $15.8 billion. By the end of 2026, market research firm eMarketer predicts https://www.emarketer.com/press-releases/meta-to-surpass-google-in-digital-ad-revenues-for-first-time-ever/ , Meta will pass Google to become the world’s largest digital advertising company. But money alone can’t buy AI supremacy. “The story of Meta in the past decade-plus has been just a series of swings and misses,” says investor and Spyglass newsletter https://spyglass.org/ author M.G. Siegler. “And then just constantly trying to figure out what is next and not being able to find it.” The company’s previous bets on new frontiers haven’t paid off, from enterprise software and cryptocurrency to, most prominently, the metaverse. Over the As Meta searches—once again—for its second act, it’s now all in on AI. In less than four months, it has shipped three versions of Muse Spark, its new flagship large language model. But so far, the company hasn’t managed to incorporate magical AI of any kind in its products. It’s also years behind other AI labs in the lucrative domains of AI coding and cloud services. As Zuckerberg steers Meta toward a goal—personal superintelligence—that he’s only beginning to explain in more than the sketchiest of details, he’s navigating a period of change unlike any in the company’s history, with a playbook that’s anything but proven. When Facebook was young https://www.fastcompany.com/59441/facebooks-mark-zuckerberg-hacker-dropout-ceo and in full “move fast and break things” mode, it was too busy to focus on artificial intelligence. “For many years, when people proposed building a research lab or division or something, I didn’t want to do it,” Mike Schroepfer, who joined the company in 2008 and served as CTO from 2013 to 2022, told me in 2018 https://www.fastcompany.com/90276668/facebook-is-five-years-into-its-ai-journey . “I couldn’t paint a clear enough line between the work they do and what was needed for the company.” That began to change in 2012, around the time the company acquired https://www.fastcompany.com/1830138/facebook-buying-instagram-1-billion Instagram for $1 billion, neutralizing a startup that could have become its greatest competitive threat. With photo sharing already among the most popular activities on Facebook’s namesake social network, Zuckerberg soon bought https://techcrunch.com/2012/06/18/facebook-scoops-up-face-com-for-100m-to-bolster-its-facial-recognition-tech/ Face.com, an Israeli facial recognition startup, and assembled a small research team dedicated to exploring computer vision. As Facebook passed a billion monthly users and neared its 10th anniversary, Zuckerberg’s interest in AI grew more expansive. In late 2013, he created a lab called FAIR for “Facebook AI Research,” or, in the Meta era, “Fundamental AI Research” and hired one of the field’s most respected scientists, Yann LeCun, to run it. “Mark was kind of thinking about ‘What should Facebook do for the next 10 years?’” LeCun explained to me in 2015 https://www.fastcompany.com/3052885/mark-zuckerberg-facebook . “AI was seen as a really strategic set of technologies for the future.” The lab aimed to develop AI that was “better than human level at all of the primary human senses: vision, hearing, language, general cognition,” Zuckerberg told me in 2015. FAIR quickly made its mark by creating PyTorch https://pytorch.org/ , an open-source framework for building, training, and running deep learning algorithms. It became an industry standard embraced by other companies, including a low-profile lab called OpenAI. But FAIR’s work didn’t transform Facebook as much as another of Zuckerberg’s strategic priorities. Until Apple’s 2007 introduction of the iPhone led to the smartphone boom, Facebook had been a PC-centric product. By 2012, though, its products dominated mobile, with Facebook and Instagram accounting for 26% of app usage in the U.S., according to Comscore. Google’s combined properties, by contrast, accounted for 10%. Zuckerberg didn’t regard that as a triumph: Facebook was, after all, still merely a tenant on Apple and Google’s platforms, not a landlord. His attempt to create a “ Facebook phone https://www.fastcompany.com/1764057/facebook-social-are-front-and-center-htcs-new-status-handset ,” with its own layer of software on top of Google’s Android, became one of Facebook’s most famous misfires, with AT&T marking it down from $99 to 99 cents within a month of release. “One of my big regrets is that Facebook hasn’t had a major chance to shape the mobile operating system ecosystem,” Zuckerberg lamented in 2015 https://www.fastcompany.com/3052885/mark-zuckerberg-facebook . He was explaining to me why he’d acquired Oculus VR the previous year for what was later revealed to be close to $3 billion. The startup’s virtual reality headset was a mind-blowing breakthrough—so dazzling it seemed it might be the next epoch-shifting device. If Oculus turned out to be the iPhone of VR, Facebook would become its Apple. After patiently investing in the technology, Meta sold more than 10 million units of its $299 Quest 2 headset in a little over 14 months following its debut in fall 2020, according to Counterpoint Research. Even so, VR felt stunted—a snazzy form of gaming, but not the smartphone’s inevitable heir. Zuckerberg was undaunted, however, and announced https://www.fastcompany.com/90691334/facebook-meta-new-name Facebook’s rebrand as Meta in October 2021. It was an expensive foray: Between 2020 and 2025, Meta lost $84 billion on Reality Labs, the division responsible for metaverse hardware; in March 2026, it announced it was shutting down the Quest version of its Horizon Worlds social network, the closest thing it had to a working prototype of the metaverse. A day later, after users complained, it said https://www.cnbc.com/2026/03/19/meta-backtracks-on-decision-to-end-horizon-worlds-vr-ans-spoke-up.html it would keep existing spaces and games open “for the foreseeable future.” In his quest to challenge the smartphone’s primacy as a platform, Zuckerberg also pushed into smart glasses in partnership with eyewear giant EssilorLuxottica https://www.fastcompany.com/91270766/essilorluxottica-most-innovative-companies-2025 . There was nothing incredible about the initial results, a pair of Ray-Ban sunglasses with built-in cameras, microphones, open-ear audio playback, and a companion app. They just had the advantage of actually existing. Zuckerberg remained salty about the smartphone’s primacy as a platform: “I think the fact that we basically take this thing out of our pocket and kind of have our head stuck in it for a while is just not how we all want to interact,” he told https://www.fastcompany.com/90673958/facebook-smart-glasses-ray-ban-stories-luxottica me in 2021. The fact that Apple shipped 235 million iPhones that year, according to market research firm IDC, suggested not everyone agreed. Still, Meta’s glasses have turned out to be a workable, hands-free platform for AI. In 2023, the company gave them a Siri-esque voice assistant. In 2025, it introduced a model that put tiny displays in the wearer’s range of sight. Today, Meta and EssilorLuxottica have 69% of the nascent market for smart glasses, according https://www.idc.com/resource-center/blog/smart-glasses-surge-the-xr-market-is-rewriting-its-own-rules/ to IDC, which expects the category to grow to almost 40 million units a year by 2030. That’s nowhere near smartphone territory, but at least Zuckerberg, who already has a head start on Google https://blog.google/products-and-platforms/platforms/android/android-xr-io-2026/ and Apple https://www.cnet.com/tech/mobile/apple-smart-glasses-development-bumps-reported-delay/ , won’t be deprived of his fair shot at shaping a new ecosystem. Whether such products will become central to how most people interact with AI, as he keeps predicting, is anyone’s guess. While Zuckerberg was selling the world on headsets and smart glasses, a twist of fate helped propel Meta near the forefront of AI labs—at least for a time. In February 2023, Meta announced https://ai.meta.com/blog/large-language-model-llama-meta-ai/ a new LLM called Llama for Large Language Meta AI . At first, it planned to share its creation with academic researchers on a case-by-case basis. But a week later, as the world was still wrapping its mind around the three-month-old ChatGPT, a leaked version showed up on the dark web site 4chan https://www.theverge.com/2023/3/8/23629362/meta-ai-language-model-llama-leak-online-misuse . Suddenly, Llama was available to anyone who could figure out how to download and install it. And anyone who did could modify the code to suit their own purposes. Llama’s unintentional release led to fears of it being misused by bad actors. Some of that played out, including the creation of a Llama-powered Discord bot that spewed hate speech https://www.vice.com/en/article/people-used-facebooks-leaked-ai-to-create-a-based-chatbot-that-says-the-n-word-basedgpt/ . But there was also excitement about the freely available model’s potential to democratize the spread of useful forms of AI. OpenAI, Anthropic, and Google tightly control their primary models. Zuckerberg realized he could drive improvements more quickly by sharing Meta’s work. The prospect rattled other companies. “Paradoxically, the one clear winner in all of this is Meta,” wrote Google engineer Luke Sernau in an internal memo https://newsletter.semianalysis.com/p/google-we-have-no-moat-and-neither published by SemiAnalysis. “Because the leaked model was theirs, they have effectively garnered an entire planet’s worth of free labor.” Less than five months later, Meta leaned into that advantage. Rather than vetting users, the company simply released Llama 2 to the public as an open-weight model. AI purists use this term for models such as Llama, whose training data remains proprietary, making them less than fully open source. The following April, it did the same with Llama 3, which AI developer and blogger Simon Willison https://simonwillison.net/ calls “absolutely spectacular.” Buoyed by its euphoric reception, Zuckerberg wrote that he expected Llama to become the industry’s most advanced model by 2025. Instead, Meta’s AI momentum evaporated. In January 2025, China’s DeepSeek released https://www.fastcompany.com/91268664/deepseek-called-into-question-big-ai-trillion-dollar-assumption-openai a surprisingly powerful open-weight AI model, startling the industry and encroaching on Meta’s turf. Meta responded a few months later with Llama 4, touting https://ai.meta.com/blog/llama-4-multimodal-intelligence/ its strong performance in industry-standard benchmarks that evaluate areas such as reasoning, coding, and image generation. But a kerfuffle broke out when it became clear that Meta was highlighting numbers for a version of the model it didn’t actually release. FAIR chief LeCun later told https://www.ft.com/content/e3c4c2f6-4ea7-4adf-b945-e58495f836c2?syn-25a6b1a6=1 the Financial Times https://www.fastcompany.com/91268664/deepseek-called-into-question-big-ai-trillion-dollar-assumption-openai that the results were “fudged a little bit.” According to Users widely regarded the model they did get as a step backward. “What I heard is that they made some mistakes in the data they used, and that’s one of the reasons Llama 4 was not a great model,” says Andrew Dai, a former Google DeepMind scientist who was among the industry talent Zuckerberg courted via e-mail to join Meta he ended up leaving Google to found a startup . When it came to that underlying training data, Meta was as proprietary as any of its competitors: “If you look at the papers for Gemini and ChatGPT, there’s usually only one paragraph, or a couple of sentences, about the data,” Dai notes. After Llama 4’s release, Meta “fell out of the race, and traditionally, when you’ve fallen out of the race, you can’t catch up,” says Ethan Mollick, a Wharton professor and the author of Co-Intelligence and the forthcoming Co-Existence . LeCun told the FT that the incident led to Zuckerberg losing faith in FAIR’s team. Soon, Zuckerberg was having exploratory conversations with Scale AI’s Wang that led to Wang coming on board as head of Meta Superintelligence Labs MSL . No longer the architect of Meta’s AI strategy, LeCun left to found his own AI startup https://www.bbc.com/news/articles/cdx4x47w8p1o . Wang’s hire https://www.fastcompany.com/section/hiring was greeted with skepticism https://futurism.com/artificial-intelligence/zuckerberg-fall-out-new-ai-hire by outsiders and even some insiders: One Scale employee tells Fast Company that his former employees saw Wang as a “ marketing https://www.fastcompany.com/section/marketing figure” for Meta. But when the new lab shipped its first model, Muse Spark 1.0, in April 2026, it got generally positive reviews https://www.businessinsider.com/used-meta-muse-spark-ai-rate-lunch-suggest-dinner-2026-4 : “It was not a frontier model, but it was a good reentry,” says Mollick. It was also as hermetically sealed as anything from Anthropic and OpenAI. The same was true of Muse Spark 1.1 and 1.2. The latter also made a splash by dramatically undercutting other AI providers’ pricing, especially for customers who agreed to let Meta train future models on their input. But the arrival of Muse Spark didn’t mark the demise of Meta’s investment in open-weight AI. In his August manifesto, Zuckerberg wrote that opening up AI is crucial to its future on multiple fronts, from minimizing cybersecurity risks to keeping the U.S. at the forefront of model development. The same day he posted the piece, Meta released a new open-weight model, Muse Glimmer https://research.meta.ai/blog/introducing-muse-glimmer-open-agentic-model , that was efficient enough to run on a high-end PC or Mac. It also promised to make an open-weight version of Muse Spark 1.2 available within weeks. As the company marshals resources toward this new push, the transition has been bruising for employees. Weeks after releasing Muse Spark, Meta laid off 8,000 people https://www.fastcompany.com/91545576/heres-how-meta-is-justifying-its-layoffs-to-thousands-of-employees —about 10% of its workforce—with teams working on metaverse hardware among those most affected, though some employees in AI roles were also cut. Seven thousand other staffers were moved into roles focused on accelerating model development. Meta also began rolling out workplace software to track employees’ keyboard strokes and mouse movements, using them as model training fodder. It later suspended the program. Back in 2015, Zuckerberg had proudly told me that his success as a CEO had come from the effort he put into “building a culture where people think about the mission in the same way that I do.” Now that synchronicity seems to be gone, at least for the moment. According to a current MSL staffer, employees in the lab regard Zuckerberg’s “personal superintelligence” talk as sloganeering, not a road map for building useful products. One Silicon Valley insider says the richly compensated AI recruits he knows at the company are “mercenaries,” there “to ride this out for as long as Zuck’s interests hold and take as much money off the table as possible.” Morale among longer-serving staffers, who have seen friends and mentors depart by edict or choice, has cratered, says the MSL employee. Meta’s leadership has been forced to acknowledge the turmoil. “We obviously did an atrocious job explaining the vision,” admitted CTO Andrew Bosworth in an internal memo reported https://www.wired.com/story/andrew-bosworth-meta-employees-unrest/ by Wired in June. In another memo https://www.reuters.com/business/metas-zuckerberg-admits-mistakes-made-ai-transformation-2026-06-12/ , quoted by Reuters the same month, Zuckerberg said he was “focused on providing as much stability as possible” and didn’t expect more layoffs—“company-wide” ones, at least—in 2026. But he also girded employees for more disruptions: “I don’t want to overpromise, because the world is changing in ways that are out of our control.” Even as Zuckerberg rips Meta apart and rebuilds it on the fly, he remains vague about what the company would do with a transformative AI model. In July 2025, he wrote https://about.fb.com/news/2025/07/personal-superintelligence-for-everyone/ that personal superintelligence would help “you achieve your goals, create what you want to see in the world, experience any adventure, be a better friend to those you care about, and grow to become the person you aspire to be.” More than a year later, in his manifesto, he said the technology would be “a personalized tutor,” “a coach with a PhD in every subject and unlimited patience,” and “a superintelligent learning assistant that knows exactly how to teach you new job skills.” He also wrote that Meta would offer this tech “for free or as affordably as possible,” suggesting an ad-based business model. In a way, Zuckerberg’s lack of specificity makes sense. Though definitions of superintelligence vary, the term is widely understood https://www.fastcompany.com/91363668/superintelligence to apply to AI that not only matches human capability across all disciplines—which is known as artificial general intelligence https://www.fastcompany.com/91044103/what-is-artificial-general-intelligence-openai-gpt-4-musk-lawsuit , or AGI—but greatly exceeds it. Only superintelligent AI, which doesn’t yet exist, may know for sure what it could do for us. Little that Meta has shipped to date—such as Pocket https://techcrunch.com/2026/08/20/meta-brings-pocket-an-app-that-lets-you-vibe-code-and-share-games-to-us-users/ , an AI app that lets users create simple games—feels like a precursor to any of this. And after stumbling its way through more than 20 years of privacy-related controversies, the company is still creating new, avoidable ones, sowing distrust for personal superintelligence even before it arrives. In June, Wired found https://www.wired.com/story/meta-smart-glasses-face-recognition-nametag-connections/ facial recognition software, a technology Meta shelved in 2021 over “growing social concerns,” embedded in the Meta AI smartphone app, apparently in preparation for its use with the company’s smart glasses. The same month, 404 Media reported https://www.404media.co/hackers-simply-asked-meta-ai-to-give-them-access-to-high-profile-instagram-accounts-it-worked/ that hackers had discovered that it was trivially easy to trick Meta’s AI support assistant into helping them take over high-profile Instagram accounts. In July, uproar https://www.bbc.com/news/articles/cp9lee19y1yo over a new feature that let Instagram users generate photorealistic images of other users—no permission required—prompted Meta to yank it https://www.cbc.ca/news/business/meta-instagram-change-ai-settings-9.7265448 after two days. Then there are the growing legal woes Meta faces over charges that its platforms are harmful to children. They include $942 million in fines in New Mexico https://www.fastcompany.com/91515805/meta-to-pay-for-exposing-children-to-sexual-exploitation-new-mexico-trial-endangering-children and a suit filed by 29 states, which Meta just settled https://www.nytimes.com/2026/08/26/technology/meta-settlement-social-media-addiction-lawsuit.html?unlocked article code=1.8VA.kXHt.Hj0ytF9e77nn&smid=nytcore-ios-share by agreeing to pay up to $17.1 billion in penalties and make changes to its products. In his manifesto and elsewhere, Zuckerberg has been taking potshots at unspecified others who believe AI competes with humanity rather than augmenting it. In contrast, Meta believes in “putting power in people’s hands to pursue their own aspirations,” he wrote in a Wall Street Journal op-ed https://www.wsj.com/opinion/the-ai-future-is-for-everyone-a0c24e20 . But Zuckerberg’s own public profile of late makes him an odd spokesman for the kind of personal growth he has said AI can fuel. Around the time of Donald Trump’s return to the White House—which Zuckerberg celebrated as cohost of a Washington, D.C., black-tie reception—he began letting it all hang out. On Joe Rogan’s podcast, he groused about corporate America’s “cultural neutering” and insufficient appreciation of aggressive “masculine energy.” In June, he was among the guests at the White House UFC match https://www.fastcompany.com/91256283/meta-dana-white-board-trump-ally-ufc-boss held on the president’s 80th birthday. However personal superintelligence pans out—and however long that takes—Meta is already under pressure from investors to demonstrate that its AI initiatives won’t be a money pit forever. After the company’s second-quarter earnings call didn’t calm their nerves, its stock dropped by 10% in a day. In early August, Meta released Muse Code https://developer.meta.com/ai/lp/muse-code/?utm source=search&utm medium=Muse-Code-Tier-0&utm campaign=paid&utm term=phrase-match-muse-code-meta&gad source=1&gad campaignid=24111147749&gbraid=0AAAAAC2gQWBJPlQOqbUnfK3i7otCsfNPe&gclid=CjwKCAjw-rTUBhAiEiwADv8gBDuuz9MBStDUF-65NSt3voPTAdrIPaHsceNHTF1t7zYBnaEihuvkGxoCE4MQAvD BwE , its first AI-powered coding agent, but like every other product in the category, it will be hard-pressed to steal mindshare and users from Anthropic’s Claude Code https://www.fastcompany.com/91502017/anthropic-most-innovative-companies-2026 , the industry’s current darling. And while Meta now offers Muse Spark as an API for third-party developers, that represents the tiniest of steps toward spinning up a cloud computing business akin to Google Cloud. Taking on the major AI cloud providers “is not the kind of thing Meta has the skill set or the software or the people to do,” says Creative Strategies CEO and principal analyst Ben Bajarin. Even if Meta doesn’t develop a full-blown enterprise AI platform, the technology could bolster its existing operations. In June, for instance, it introduced https://www.cnbc.com/2026/06/03/meta-business-agent-is-zuckerberg-latest-effort-to-diversify-from-ads.html a service called Meta Business Agent that allows companies to provide automated customer service via WhatsApp, Messenger, and Instagram. But until Zuckerberg can show how the company’s investment will pay off, he’ll remain vulnerable to critics. “For him, it’s ‘How can I be a big tech player and not have my own model?’” says the aforementioned Silicon Valley insider. “That’s completely egotistical. It doesn’t make business sense to pay the kind of money he’s paying to train your own models.” Making frontier AI pay for itself “will require having killer products, which Meta has done in the past, but not in a while, besides acquisitions,” says investor/writer Siegler. “Maybe the glasses. Maybe. Still, all of this is smart for them to try. And they are shipping these new models in record time.” Lack of clarity about where AI is headed may not weigh heavily on Zuckerberg’s mind. Once you’ve turned your Harvard dorm-room coding project into a $1.5 trillion behemoth that monetizes the attention of 3.6 billion people a day, improvising your way into the future must feel only natural. “At the beginning of Facebook, I didn’t have an idea of how this was going to be a good business,” Zuckerberg told me in 2015. “I just thought it was a good thing to do.” Today, Zuckerberg isn’t a lonely visionary. Instead, he’s just another Silicon Valley CEO trying to turn AI into a good business. The awesome resources at his disposal guarantee nothing. And thanks to his own boundless ambition, the hardest part is still to come.