Weekly Top Stories - 07/10/26

Welcome to Galaxy Research's Weekly Top Stories. Subscribe to get this newsletter delivered to your inbox every Friday morning.

In this week's edition, Will Owens explains why onchain activity is so back, despite bear-market summer doldrums; Lucas Tcheyan analyzes big shifts in the AI market; and Thad Pinakiewicz looks at Kraken’s arbitration victory over the auditor that dumped it during the Gensler era.

Got feedback on this newsletter? Email [email protected]. We’d love to hear from you.

Market Update

The total crypto market cap stands at $2.28tn, up 1.57% from last week (when it stood at $2.24tn). Bitcoin's network value is 4.54% of gold's market cap. Over the last seven days, BTC is up 3.94%, ETH is up 3.16%, and SOL is down 2.43%. Bitcoin dominance is 56.68%, down 10 basis points from last week. (Data: Messari, Galaxy Research)

Memecoins Are ‘So Back,’ but for How Long?

Who could have guessed? Robinhood Chain, which went live July 1, was built explicitly for tokenized stocks, RWAs, agentic trading, and regulated financial products, yet this week, its most active use case has been memecoins.

This outcome comes on the heels of a broader resurgence in onchain user activity, primarily driven by the astounding success of social trading. Fomo (the leading social trading app with Apple Pay on-ramps, leaderboards, and multi-chain support) just closed a $75m Series B round at a $550m valuation and has been breaking all-time highs in user metrics.

This activity has been led by $ANSEM, a Pump.fun token named for prominent Crypto Twitter key opinion leader (KOL) Ansem (@blknoiz06), who adopted and championed it despite not launching it himself. This token went from zero to >$400 million market cap in a matter of days and is now trading around $300 million market cap at time of writing. As a reminder, no memecoin launched in 2026 has surpassed a $1 billion market cap.

At launch, the conversation focused on tokenized stocks and other (comparatively) buttoned-down financial products. But on Tuesday night, traders figured out that Vlad Tenev, CEO of Robinhood, had followed the X account for CASHCAT, one of the top memecoins on the nascent Robinhood Chain. This was perceived as a bullish catalyst in memecoin trading and led to traders scrambling to bridge their assets over to Robinhood Chain to “ape” into the token.

Tenev then came out and explicitly tweeted about memecoins on his company's new blockchain. His post sent activity on Robinhood Chain soaring. Many individuals on crypto Twitter were amped, replying with catchphrases like “we are so back.”

Decentralized exchange (DEX) volumes on Robinhood Chain reflect the shift. After launching quietly on July 1, daily volume settled into a modest $35m–$70m range through its first week, then spiked nearly 10x in a day to $560m+ on July 8, coinciding with the CASHCAT frenzy. And on Thursday afternoon, fomo announced the integration of Robinhood Chain, another sign of retail demand for the chain.

Our Take

The rebound of sentiment in the crypto “trenches" is real. Let’s start with what’s genuinely new: the rails are better than they’ve ever been. Fomo has simplified onboarding into an Apple Pay transaction and discovery into a social feed, collapsing the gap between “sees a token” and “owns a token” to almost nothing. Its addition of perps trading for non-U.S. users has further expanded the platform’s reach.

Now, the deja vu. The demand rushing into memecoin trading is the same demand it’s always been: speculative, narrative-driven, and pointed at whatever’s moving fastest. Syncracy Capital co-founder Ryan Watkins bluntly made the skeptical case against focusing on this activity. He questioned why anyone should be excited about degens playing the same zero-sum games and argued Robinhood could do better than rekindle the 2024 meta that then lived primarily on Solana. It’s a fair critique. A chain with Robinhood’s distribution highlighting memes onchain is a statement about what the market actually demands from any new blockchain you hand it.

Which is why the durability question is the real story. Memecoin surges are structurally front-loaded. That’s the base case. ANSEM’s peak week on fomo roughly doubled any prior weekly high on the platform in 2026. Attention is the most abundant and least durable asset in crypto. The bet worth watching isn’t whether these tokens hold their market cap. It’s whether the infrastructure underneath these flows retains the users that memecoins brought in the door once volumes cool. So far, Solana’s success in this regard has been the exception.

Speculation has always been crypto’s best user-acquisition funnel; the problem is keeping people onchain and moving them into more sustainable activity. If new rails convert even a fraction of this meme-driven traffic into retained wallets or real RWA usage, this cycle will look different than the ones before. If they don’t, “onchain activity is back” and “the casino got upgraded” are once again describing the same thing.

The rails keep getting better. The behavior filling them hasn’t changed much. Whether this cycle breaks that pattern is the only question that matters. - Will Owens

OpenAI Releases GPT-5.6 as Frontier Labs Face New Headwinds

OpenAI began its public rollout of GPT-5.6 Thursday, opening its newest AI model family to everyone after weeks of restricted access. The lineup splits three ways under a new naming scheme where the number marks the generation and the name marks a capability tier. Sol (not to be confused with the Solana blockchain’s native asset) is the flagship for hard coding, reasoning, and cybersecurity work. Terra is a cheaper everyday model that roughly matches GPT-5.5 at half the price. Luna is the fastest and cheapest of the three. Sol runs $5 per million input tokens and $30 per million outputs. Luna runs $1 and $6, respectively.

OpenAI previewed GPT-5.6 in late June to about 20 government-approved partners while U.S. agencies reviewed the models, and the Department of Commerce cleared a broad launch this week after additional testing. OpenAI has said plainly that it doesn't think this should become the standard way models reach the market.

It was a crowded week for artificial intelligence releases. OpenAI also shipped a new generation of voice models. SpaceXAI released Grok 4.5 on Wednesday, its first model built specifically for coding and agents, trained alongside Cursor (the startup SpaceX is buying for $60 billion) and priced at $2 and $6 to undercut Anthropic's Opus 4.8. Anthropic extended included access to its Mythos-class Claude Fable 5 on paid plans to July 12, pushing back the date it moves to metered usage credits. Meta put out an agentic coding model of its own.

Our Take

The pitch from frontier labs has historically been simple. They build the most capable models, so customers should pay to use them. Increasingly, that capability is becoming as much of a liability as an advantage.

The more powerful these models become, the more government scrutiny they attract, the more strategic risk they create for enterprise customers, and the more pressure they face from cheaper open-source alternatives that are "good enough" for a growing number of use cases. GPT-5.6 may be an important model release, but it also arrives at a moment when the basic business model of frontier AI is becoming more complicated.

Start with Washington. On June 2, President Trump signed an AI cybersecurity executive order that gives the federal government a more formal role around the release of the most capable frontier models. The order directs agencies to create a “voluntary” framework where developers can work with the government to determine whether a model qualifies as a "covered frontier model," provide federal officials with access for up to 30 days before release to trusted partners, and coordinate on who gets early access. It also says, explicitly, that this is not a licensing, preclearance, or permitting regime. The process is optional on paper, but it still normalizes a pre-release government review loop for the models that labs most want to commercialize.

GPT-5.6 shows how blurry that line can get. OpenAI says it previewed the models' capabilities with the U.S. government and began with a limited group of trusted partners whose participation had been shared with Washington before broader availability. It also says GPT-5.6 Sol, Terra, and Luna advanced the frontier in software engineering, scientific research, and cybersecurity, exactly the kinds of capabilities that make governments nervous, and it rates all three tiers, not just the flagship, "High" for cyber and biological risk. There’s an inherent tension building here. Frontier labs want to market how powerful their models are, but the more powerful the model, the harder it is to release it to customers. Galaxy Head of Firmwide Research Alex Thorn wrote two weeks ago about the pitfalls of AI governance revolving around private letters from the Commerce Department, and at the G7, French President Emmanuel Macron warned that European governments would stop buying American AI if access could be cut off on a day's notice, while Canada's Prime Minister Mark Carney called dependence on concentrated AI a strategic mistake. The curb is not unique to the United States. China's Ministry of Commerce has been talking to Alibaba, ByteDance, and Z.ai about restricting overseas access to their most capable models, the open cyber-capable ones included, for the same national-security reason Washington restricted Anthropic's.

This slow-motion crackdown matters because enterprises are already rethinking what it means to outsource intelligence. The first wave of enterprise AI adoption was mostly about getting access to the best model. The next wave is about control. Who owns the data, who owns the workflow, who captures the learning loop, and who ultimately gets the "alpha" created by using AI inside the business. A growing number of large software company leaders have become more outspoken about the issue in recent weeks. Palantir CEO Alex Karp has criticized OpenAI and Anthropic, advocating for his firm’s own sovereign AI pitch focused on giving enterprises control over their data, models, applications, and deployment environments. In his view, if a company pipes its most valuable processes into a frontier lab's platform, it may be getting intelligence back, but it may also be teaching the platform where the value sits. The deeper version of this, which Microsoft CEO Satya Nadella recently framed as a shift to "frontier ecosystems" rather than frontier models, is that a company's real edge becomes the private evaluations and weights it trains on its own data, not whichever model runs underneath.

Sentiment on AI token spending has also round-tripped in a few months. The euphoric phase, when companies posted internal leaderboards celebrating the "10x" engineer who ran up the biggest multi-agent coding bill, is over. Firms started burning annual AI budgets in a single quarter, and the variable cost of all those tokens became impossible for finance departments to ignore. By some estimates, frontier providers earn 70% to 80% gross margins on usage-based token revenue. Coinbase is one of the cleanest examples. CEO Brian Armstrong posted in late June that the company cut its AI spending nearly in half while token usage kept climbing, and did it by changing the defaults rather than capping engineers. Most queries now route to open-weight models like GLM 5.2 and Kimi 2.7 through an internal gateway, with frontier models reserved for planning and the hard problems. A new consumption pattern is emerging. Default the bulk of the work to cheap open models; escalate to the frontier only when it matters.

Put those pressures together, and the frontier-lab story looks more complicated than the leaderboards suggest. These companies still win when they build the best models. But the best models also attract the most government scrutiny, create dependency risks for customers, and give enterprises the strongest reason to diversify across cheaper, controllable alternatives.

“We have the best model” used to be the whole pitch. Increasingly, it’s also the biggest risk factor. - Lucas Tcheyan

Kraken Wins Arbitration Against Auditor That Bolted During ‘OCP 2.0’ Era

Kraken's parent, Payward, went to Delaware's Court of Chancery this week to collect $22 million from Mazars USA, the rare auditor that owes money to a client it walked out on rather than the other way around.

Here is a fun way to lose $22 million as an auditor. You audit a company for three years, sign off cleanly twice, get within days of a third opinion, and then put down your pen and leave. You tell the client, in writing, that you have no disagreement with its management, no doubts about its integrity, and no evidence of fraud. Then an arbitrator spends just five days deciding what that exit was worth, and hands you the bill.

That is ostensibly what happened, but court filings tell a more revealing story (despite redactions so heavy you’d think you were reading UFO or JFK assassination documents). Kraken hired Mazars to audit its 2022 books, and by late 2023 Mazars was, on its own account, nearly finished and expecting another clean opinion. Then in November the Securities and Exchange Commission sued Kraken for running an unregistered exchange, and Mazars, by then also subpoenaed by both the SEC and a grand jury for its Kraken work papers, quit that December without issuing anything.

The audit contract gave Mazars a short, specific menu of exits: undisputed bills more than 60 days overdue, a determination that "Company personnel have not been forthcoming... or have not been truthful," or a finding that "continuing with the engagement would violate professional ethical obligations." Otherwise, the letter said, "we may not terminate this engagement."

The trouble for Mazars is that Mazars itself, not Kraken, closed those potential exits from the contract. The arbitrator found that Mazars had confirmed, in writing and more than once, that it "had no disagreements with management with respect to any matter of accounting principles or practices," that it "made no communications or determinations with respect to any lapse in management integrity," and that it "did not identify any fraud." Nobody says Kraken stopped paying. That leaves only the professional-ethics catchall, and the rationale that Mazars invoked in its termination notice is redacted.

An audit is not decoration. State money-transmitter licenses, banking relationships, and counterparties all run on the assumption that someone serious has checked the numbers. The arbitrator found Mazars' departure created a "licensing crisis," and tied $12.5 million of the award to Kraken's acquisition of TradeStation Crypto, a platform Kraken bought for the regulatory licenses it suddenly needed once it no longer had its audit. Mazars knew that pulling the audit would threaten the business, could foresee Kraken scrambling to replace what it lost, and now owns the cost of the scramble.

Our Take

To read Mazars’ exit properly you have to put it in the setting the crypto industry calls Operation Choke Point 2.0.

On Kraken's telling, the SEC suit was one instrument in a broader campaign of informal pressure: a January 2023 joint statement from the Fed, FDIC and OCC flagging crypto as a safety-and-soundness concern, FDIC letters telling banks to pause crypto activity, the SEC's SAB 121 pushing custody onto balance sheets, and the disappearance of the settlement rails the industry ran on, with the 2023 closures of Silvergate among other crypto-friendly banks.

An auditor watching its crypto clients turn radioactive, and having already pulled its proof-of-reserves work for the entire sector a year earlier, is behaving exactly as that pressure would predict. And the SEC case that supposedly justified the alarm was later dismissed with prejudice, with no penalties and no admission of guilt.

Now the interesting professional question, which is whether an auditor has to form a view on the merits of its client's legal troubles. Mostly it does not. Auditors handle pending litigation all the time through disclosure and, where warranted, a qualified opinion. There was even a live accounting issue in the mix with Kraken. Mazars' reconciliation of the 2022 numbers had surfaced a $250m discrepancy between several of Kraken's accounts and its GAAP records. But that is precisely what an audit is for. You find the error, you correct it, you flag it in the report. You do not discover a flaw in your own prior work (Mazars conducted Kraken’s 2021 and 2020 audits) and respond by refusing to file any work at all.

The contrast with the famous auditor walkouts is notable here. When Deloitte quit Manpasand Beverages in 2018, days before results, it was over the company withholding data. When EY resigned from Super Micro in 2024 and sent the stock off a cliff, EY said plainly that new information had left it unable to rely on management's representations. (That exit deserves an asterisk, because a special committee later determined that EY's resignation and the conclusions in its letter were not supported by the facts examined in its review, a good reminder that even a dramatic departure is a reason to investigate rather than a verdict.)

But EY at least alleged something. Mazars alleged nothing, affirmed the client was clean, and left regardless.

That brings us to the arbitrator's cleanest line: "This was a judgment call by Mazars, not a mandated result."

No standard forced Mazars out. It chose to go, knowing the choice would blow a hole in the business of an honest, paying client and exposed Mazars to the liability it now owes. The reason is redacted, so we may never know what spooked Mazars. But you do not pay $22 million to walk away from a client that did nothing wrong unless something else in the room was worth more than the fee.

For an industry that spent years insisting Choke Point 2.0 was real rather than paranoia, a neutral arbitrator ruling that the exit was a choice and not a mandate is the closest thing to an admission it is going to get. –Thad Pinakiewicz

Other News

• 🇺🇸 Sen. Wyden (D-OR) urges Senate to keep BRCA dev protections in CLARITY

• 🚨 SEC to propose Regulation Crypto as soon as this month

• ₿ MSTR sells BTC again, this time in size (see Chart of the Week, below)

• 🚫🛢️🕙🐈 CFTC blocks CME plan to offer 24/7 oil futures trading

• 💰🦾 Crypto VC Paradigm raises $1.2b for AI and robotics investments

• 🖨️ Anthropic appoints ex-Fed chair Bernanke to oversight trust

• 🚢 Vanguard, onetime crypto skeptic, seeks to hire first head of digital assets

• ⚖️🔮 Polymarket sued over MSTR market decision (read our take from May) ...

• ⚡... and adds instant BTC Lightning deposits powered by Spark

Chart of the Week: Saylor Sells Again, This Time in Size

Michael Saylor’s Strategy disclosed Monday that it sold ~$216m of bitcoin last week in two chunks, the firm's first material sales ever, dwarfing the $2 million disposal it did in May.

Notably, the sale does not appear to be part of the bitcoin monetization program Strategy announced a week earlier as part of a strategic overhaul. “As of July 5, 2026, the full amount of this capacity remains available,” Strategy said in Monday’s filing. That's up to $1.25 billion the company could still raise by selling more BTC.

For more on the market implications of Strategy’s new, er, strategy, read last week’s newsletter.